Inspiratory muscle training, with or without concomitant pulmonary rehabilitation, for chronic obstructive pulmonary disease (COPD)

Abstract

Background

Inspiratory muscle training (IMT) aims to improve respiratory muscle strength and endurance. Clinical trials used various training protocols, devices and respiratory measurements to check the effectiveness of this intervention. The current guidelines reported a possible advantage of IMT, particularly in people with respiratory muscle weakness. However, it remains unclear to what extent IMT is clinically beneficial, especially when associated with pulmonary rehabilitation (PR).  

Objectives

To assess the effect of inspiratory muscle training (IMT) on chronic obstructive pulmonary disease (COPD), as a stand‐alone intervention and when combined with pulmonary rehabilitation (PR).

Search methods

We searched the Cochrane Airways trials register, CENTRAL, MEDLINE, Embase, PsycINFO, Cumulative Index to Nursing and Allied Health Literature (CINAHL) EBSCO, Physiotherapy Evidence Database (PEDro) ClinicalTrials.gov, and the World Health Organization International Clinical Trials Registry Platform on 20 October 2022. We also checked reference lists of all primary studies and review articles.

Selection criteria

We included randomized controlled trials (RCTs) that compared IMT in combination with PR versus PR alone and IMT versus control/sham. We included different types of IMT irrespective of the mode of delivery. We excluded trials that used resistive devices without controlling the breathing pattern or a training load of less than 30% of maximal inspiratory pressure (PImax), or both.

Data collection and analysis

We used standard methods recommended by Cochrane including assessment of risk of bias with RoB 2. Our primary outcomes were dyspnea, functional exercise capacity and health‐related quality of life. 

Main results

We included 55 RCTs in this review. Both IMT and PR protocols varied significantly across the trials, especially in training duration, loads, devices, number/ frequency of sessions and the PR programs. Only eight trials were at low risk of bias.

PR+IMT versus PR

We included 22 trials (1446 participants) in this comparison. Based on a minimal clinically important difference (MCID) of −1 unit, we did not find an improvement in dyspnea assessed with the Borg scale at submaximal exercise capacity (mean difference (MD) 0.19, 95% confidence interval (CI) −0.42 to 0.79; 2 RCTs, 202 participants; moderate‐certainty evidence).  

We also found no improvement in dyspnea assessed with themodified Medical Research Council dyspnea scale (mMRC) according to an MCID between −0.5 and −1 unit (MD −0.12, 95% CI −0.39 to 0.14; 2 RCTs, 204 participants; very low‐certainty evidence). 

Pooling evidence for the 6‐minute walk distance (6MWD) showed an increase of 5.95 meters (95% CI −5.73 to 17.63; 12 RCTs, 1199 participants; very low‐certainty evidence) and failed to reach the MCID of 26 meters. In subgroup analysis, we divided the RCTs according to the training duration and mean baseline PImax. The test for subgroup differences was not significant. Trials at low risk of bias (n = 3) demonstrated a larger effect estimate than the overall.

The summary effect of the St George's Respiratory Questionnaire (SGRQ) revealed an overall total score below the MCID of 4 units (MD 0.13, 95% CI −0.93 to 1.20; 7 RCTs, 908 participants; low‐certainty evidence). 

The summary effect of COPD Assessment Test (CAT) did not show an improvement in the HRQoL (MD 0.13, 95% CI −0.80 to 1.06; 2 RCTs, 657 participants; very low‐certainty evidence), according to an MCID of −1.6 units. 

Pooling the RCTs that reported PImax showed an increase of 11.46 cmH₂O (95% CI 7.42 to 15.50; 17 RCTs, 1329 participants; moderate‐certainty evidence) but failed to reach the MCID of 17.2 cmH₂O.  In subgroup analysis, we did not find a difference between different training durations and between studies judged with and without respiratory muscle weakness. 

One abstract reported some adverse effects that were considered "minor and self‐limited".

IMT versus control/sham

Thirty‐seven RCTs with 1021 participants contributed to our second comparison. There was a trend towards an improvement when Borg was calculated at submaximal exercise capacity (MD −0.94, 95% CI −1.36 to −0.51; 6 RCTs, 144 participants; very low‐certainty evidence). Only one trial was at a low risk of bias.

Eight studies (nine arms) used the Baseline Dyspnea Index ‐ Transition Dyspnea Index (BDI‐TDI). Based on an MCID of +1 unit, they showed an improvement only with the 'total score' of the TDI (MD 2.98, 95% CI 2.07 to 3.89; 8 RCTs, 238 participants; very low‐certainty evidence). We did not find a difference between studies classified as with and without respiratory muscle weakness. Only one trial was at low risk of bias.

Four studies reported the mMRC, revealing a possible improvement in dyspnea in the IMT group (MD −0.59, 95% CI −0.76 to −0.43; 4 RCTs, 150 participants; low‐certainty evidence). Two trials were at low risk of bias.

Compared to control/sham, the MD in the 6MWD following IMT was 35.71 (95% CI 25.68 to 45.74; 16 RCTs, 501 participants; moderate‐certainty evidence). Two studies were at low risk of bias. In subgroup analysis, we did not find a difference between different training durations and between studies judged with and without respiratory muscle weakness. 

Six studies reported theSGRQ total score, showing a larger effect in the IMT group (MD −3.85, 95% CI −8.18 to 0.48; 6 RCTs, 182 participants; very low‐certainty evidence). The lower limit of the 95% CI exceeded the MCID of −4 units. Only one study was at low risk of bias.

There was an improvement in life quality with CAT (MD −2.97, 95% CI −3.85 to −2.10; 2 RCTs, 86 participants; moderate‐certainty evidence). One trial was at low risk of bias.

Thirty‐two RCTs reported PImax, showing an improvement without reaching the MCID (MD 14.57 cmH₂O, 95% CI 9.85 to 19.29; 32 RCTs, 916 participants; low‐certainty evidence). In subgroup analysis, we did not find a difference between different training durations and between studies judged with and without respiratory muscle weakness.  

None of the included RCTs reported adverse events.

Authors' conclusions

IMT may not improve dyspnea, functional exercise capacity and life quality when associated with PR. However, IMT is likely to improve these outcomes when provided alone.

For both interventions, a larger effect in participants with respiratory muscle weakness and with longer training durations is still to be confirmed.

Plain language summary

Are exercises for strengthening breathing muscles effective for people with chronic obstructive pulmonary disease?

Key messages

• Exercise combined with specific exercises to strengthen breathing muscles may not improve breathlessness, physical fitness and life quality. Strength of breathing muscles and endurance increased but not enough to make a difference to patients. 

• Specific exercises to strengthen breathing muscles compared to no exercise may improve breathlessness, physical fitness and life quality. Strength of breathing muscles and endurance increased, but we don't know if this benefitted patients.

• We don't know whether exercise or specific exercises to strengthen breathing muscles is better for people with weakened breathing muscles who trained for several weeks.

• Future research should focus on people with weakened breathing muscles and studies should include more people.

What is chronic obstructive pulmonary disease (COPD)?

Chronic obstructive pulmonary disease (COPD) is a lung condition characterized by blockages in the airways, which cause shortness of breath and a cough. It appears after the long‐term inhalation of irritating gases like cigarette smoke and chemicals. Training and strengthening the breathing muscles is thought to improve breathing and reduce air obstruction.

What exercise treatments do people with COPD use?

Health professionals use various exercises to help improve people's COPD. 

• Some people undertake a program of general exercise and education to help reduce symptoms and improve their exercise capacity and life quality. 

• Other people try to improve the strength and endurance of the breathing muscles through a series of breathing exercises using specific devices. This is called 'inspiratory muscle training' (IMT). The devices add resistance to breathing to strengthen the diaphragm and the intercostal muscles between the ribs ‐ the muscles used for breathing. People may then be able to breathe in more air with each breath and be active for longer. The devices are also used by people with healthy lungs to improve their sports performance.

What did we want to find out?

We wanted to find out if exercise combined with IMT compared to exercise alone, and IMT compared to no exercise or sham IMT has a better effect on breathlessness, physical fitness and life quality. (A sham device has no effect on breathing muscles. It allows a fair test of the real devices, because people don't know which they are using.)

We also wanted to check whether IMT was associated with any unwanted effects.

What did we do?

We searched for studies that compared

• exercise combined with IMT with exercise alone; and

• IMT with no exercise or sham IMT. 

We compared and summarized the results of the studies and rated our confidence in the evidence, based on factors such as study methods and sizes.

What did we find?

Exercise plus IMT compared with exercise alone
We found 22 studies with 1446 participants, which lasted between 2 and 24 weeks. Exercise ranged from training only on a treadmill, with only a cycle, and a combination of exercises (training with a cycle and treadmill, muscle strengthening, stair climbing, and education). The duration and devices of IMT also varied across the studies.  

We found that this combination:
• probably makes little to no difference to breathlessness (measured with different scales);
• has an unknown effect on physical fitness;
• may make little to no difference to life quality (measured with different scales);
• probably makes little to no difference to strength of breathing muscles. 

IMT versus no training or sham device
We found 37 studies with 1021 participants, which lasted from 2 weeks to a year. IMT varied across the studies regarding devices, resistance, frequency and supervision.

We found out that IMT alone:
• may reduce breathlessness measured with one scale, but it is unclear if it has an effect when measured with two other scales;
• probably improves physical fitness;
• probably improves life quality when measured with one scale, but it is unclear if it has a benefit when measured with another one;
• may make little to no difference to strength of breathing muscles.
 

What are the limitations of the evidence?

The studies used different training durations, resistance, devices, number and frequency of sessions, and physical training programs. This makes it hard to draw firm conclusions. Overall our confidence in the conclusions is reduced because the studies were small, some participants may have been aware of which treatment they were receiving, and generally, there was some diversity in the studies. 

How up to date is the evidence?

The evidence is up‐to‐date to 20 October 2022.

Summary of findings

Summary of findings 1. Pulmonary rehabilitation plus inspiratory muscle training  compared to pulmonary rehabilitation alone for  people with chronic obstructive pulmonary disease.

Pulmonary rehabilitation plus inspiratory muscle training compared to pulmonary rehabilitation alone for people with chronic obstructive pulmonary disease
Patient or population: people with chronic obstructive pulmonary disease (COPD) Setting: community Intervention: pulmonary rehabilitation (PR) + inspiratory muscle training (IMT) Comparison: PR
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with PR	Risk with PR+IMT
Dyspnea assessed with Borg scale at submaximal exercise capacity Scale from 0 to 10 (worse) Follow‐up: range 3 months to 4 months	The mean dyspnea was 4.65	The mean dyspnea was 0.19 points higher (0.42 lower to 0.79 higher)	‐	202 (2 RCTs)	⊕⊕⊕⊝ Moderatea	The combination of PR+IMT probably results in little to no difference in dyspnea measured with Borg at submaximal exercise capacity compared to PR alone, considering an MCID of −1 unit
Dyspnea assessed with mMRC Scale from 0 to 4 (worse) Follow‐up: range 1 month to 2 months	The mean dyspnea ranged from −0.8 to −0.33	MD 0.12 lower (0.39 lower to 0.14 higher)	‐	204 (2 RCTs)	⊕⊝⊝⊝ Very lowa,b	The evidence is very uncertain about the effect of the combination of PR+IMT on dyspnea measured with the mMRC compared to PR alone, considering an MCID between −0.5 and −1 unit
Functional exercise capacity assessed with 6MWD Follow‐up: range 2 weeks to 6 months	The mean functional exercise capacity was 304.72 metersc	MD 5.95 meters higher (5.73 lower to 17.63 higher)	‐	1199 (12 RCTs)	⊕⊝⊝⊝ Very lowd,e	The evidence is very uncertain about the effect of the combination of PR+IMT on the 6MWD compared to PR alone, considering an MCID of 26 meters
Health‐related quality of life assessed with SGRQ total score Scale from 0 to 100 (worse) Follow‐up: range 3 weeks to 6 months	The mean health‐related quality of life was 14.9c	MD 0.13 higher (0.93 lower to 1.2 higher)	‐	908 (7 RCTs)	⊕⊕⊝⊝ Lowf	The combination of PR+IMT may result in little to no difference in  health‐related quality of life measured with the SGRQ compared to PR alone, considering an MCID of −4 units
Health‐related quality of life assessed with CAT Scale from 0 to 40 (worse) Follow‐up: range 3 weeks to 6 months	The mean health‐related quality of life ranged from −3.42 to −3	MD 0.13 higher (0.8 lower to 1.06 higher)	‐	657 (2 RCTs)	⊕⊝⊝⊝ Very lowg,h	The evidence is very uncertain about the effect of the combination of PR+IMT on health‐related quality of life measured with the CAT compared to PR alone, considering an MCID of about −1.6 units
Inspiratory muscle strength assessed with PImax Follow‐up: range 3 weeks to 6 months	The mean inspiratory muscle strength was 67.37 cmH2Oc	MD 11.46 cmH2O higher (7.42 higher to 15.50 higher)	‐	1329 (17 RCTs)	⊕⊕⊕⊝ Moderated	The combination of IMT+PR probably slightly increases inspiratory muscle strength (PImax) compared to PR alone, without reaching the MCID of 17.2 cmH2O
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CAT: COPD [chronic obstructive pulmonary disease] Assessment Test; IMT: inspiratory muscle training; MD: mean difference; mMRC: modified Medical Research Council dyspnoea scale; MCID: minimum clinically important difference; PImax: Maximal Inspiratory Pressure; PR: pulmonary rehabilitation; RCT: randomized controlled trial; SGRQ: St George's Respiratory Questionnaire; 6MWD: six‐minute walk distance
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_423482112637765988.

^aDowngraded by one level for imprecision due to small sample size (rule of thumb: at least 400 participants).
^bDowngraded by two levels for risk of bias because all the trials are at high risk of bias.
^cIncluding change and endpoint scores.
^dDowngraded by one level for risk of bias because most of the evidence is from studies at high risk of bias and with some concern.
^eDowngraded by two levels for inconsistency due to considerable statistical heterogeneity (I² statistic), confidence intervals not overlapping, and significant variations in the direction of the effects.
^fDowngraded by two levels for risk of bias due to a considerable bias across the studies in the measurement of the outcome (lack of blinding) and all the trials are at high risk of bias and some concern.
^gDowngraded by two levels for risk of bias due to considerable bias across the studies in the measurement of the outcome (lack of blinding) and most of the evidence is from studies at high risk of bias and some concern.
^hDowngraded by one level for inconsistency due to considerable statistical heterogeneity and confidence intervals not overlapping .

Summary of findings 2. Inspiratory muscle training compared to control or sham for people with chronic obstructive pulmonary disease.

Inspiratory muscle training compared to control or sham for people with chronic obstructive pulmonary disease
Patient or population: people with chronic obstructive pulmonary disease (COPD) Setting: community Intervention: inspiratory muscle training (IMT) Comparison: control or sham
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with control or sham	Risk with IMT
Dyspnea assessed with Borg scale at submaximal exercise capacity Scale from 1 to 10 (worse) Follow‐up: range 5 weeks to 4 months	The median dyspnea was 1.5	MD 0.94 lower (1.36 lower to 0.51 lower)	‐	144 (6 RCTs)	⊕⊝⊝⊝ Very lowa,b	IMT may improve dyspnea measured with Borg scale at submaximal exercise capacity compared to control/sham but the evidence is very uncertain, considering an MCID of −1 unit (only the lower limit of the 95% CI exceeded the MCID)
Dyspnea assessed with BDI‐TDI: focal score (TDI) Scale from −9 to +9 (better) follow‐up: range 2 months to 6 months	The median dyspnea was 1.2	MD 2.98 higher (2.07 higher to 3.89 higher)	‐	238 (8 RCTs)	⊕⊝⊝⊝ Very lowb,c	IMT may improve dyspnea measured with the BDI‐TDI (Focal score) compared to control/sham but the evidence is very uncertain, considering an MCID of +1 unit
Dyspnea assessed with mMRC Scale from 0 to 4 (worse) Follow‐up: range 8 months to 8 months	The median dyspnea was 0.62	MD 0.59 lower (0.76 lower to 0.43 lower)	‐	150 (4 RCTs)	⊕⊕⊝⊝ Lowb,d	IMT may improve dyspnea measured with the modified mMRC compared to control/sham, considering an MCID between −0.5 and −1 unit
Functional exercise capacity assessed with 6MWD Follow‐up: range 2 weeks to 12 months	The mean functional exercise capacity was 298.4 meters	MD 35.71 meters higher (25.68 higher to 45.74 higher)	‐	501 (16 RCTs)	⊕⊕⊕⊝ Moderated	IMT probably improves functional exercise capacity measured with the 6MWD compared to control/sham, considering an MCID of 26 meters
Health‐related quality of life  assessed with SGRQ total score Scale from 0 to 100 (worse) Follow‐up: range 2 months to 12 months	The median health‐related quality of life was 23.61	MD 3.85 lower (8.18 lower to 0.48 higher)	‐	182 (6 RCTs)	⊕⊝⊝⊝ Very lowe,f	IMT may improve health‐related quality of life measured with the SGRQ compared to control/sham but the evidence is very uncertain, considering an MCID of −4 units (only the lower limit of the 95% CI exceeded the MCID)
Health‐related quality of life assessed with CAT Scale from 0 to 40 (worse) Follow‐up: mean 2 months	The mean health‐related quality of life ranged from −0.5 to 0.3	MD 2.97 lower (3.85 lower to 2.1 lower)	‐	86 (2 RCTs)	⊕⊕⊕⊝ Moderateb	IMT probably improves health‐related quality of life measured with CAT compared to control/sham, considering an MCID of about −1.6 units
Inspiratory muscle strength assessed with PImax  Follow‐up: range 2 weeks to 12 months	The mean inspiratory muscle strength was 51.23 cmH2O	MD 14.57 cmH2O higher (9.85 higher to 19.29 higher)	‐	916 (32 RCTs)	⊕⊕⊝⊝ Lowd,g,h	IMT may increase inspiratory muscle strength (PImax) slightly compared to control/sham considering the MCID of 17.2 cmH2O
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). BDI‐TDI: Baseline Dyspnea Index ‐ Transition Dyspnea Index;CI: confidence interval; CAT: COPD Assessment Test; IMT: inspiratory muscle training; MD: mean difference; mMRC: modified Medical Research Council dyspnoea scale; MCID: minimum clinically important difference; PImax: Maximal Inspiratory Pressure; PR: pulmonary rehabilitation; RCT: randomized controlled trial; SGRQ: St George's Respiratory Questionnaire; 6MWD: six‐minute walk distance
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_423482164894599533.

^aDowngraded by two levels for risk of bias because most of the evidence is from studies at high risk of bias and some concern, there is an issue with blinding, and high risk of bias studies show different estimates to studies at low risk of bias and some concern.
^bDowngraded by one level for imprecision due to small sample size (rule of thumb: less than 400).
^cDowngraded by two levels for risk of bias due to considerable bias across the studies in the measurement of the outcome (lack of blinding) and most of the evidence is from studies at high risk of bias and some concern.
^dDowngraded by one level for risk of bias because most of the evidence is from studies at high risk of bias and with some concern.
^eDowngraded by two levels for risk of bias because most of the evidence is from studies at high risk of bias and some concern, and high risk of bias studies show different estimate to studies at low risk of bias and some concern.
^fDowngraded by two levels for imprecision due to small sample size (rule of thumb: less than 400) and because the 95% CI includes benefit and harm.
^gWe did not downrate inconsistency although substantial statistical heterogeneity because the studies are on one side of the line of no effect. So we are more confident about the direction of the effect.
^hDowngraded by one level for publication bias because the funnel plot and the number of studies give rise to serious suspicions about publication bias.

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is a respiratory condition that includes bronchitis and emphysema. Chronic bronchitis is defined by the presence of a productive cough for at least three months per year for two consecutive years, during which other causes of cough have been excluded (GOLD 2022). Emphysema is damage to the portion of the lungs responsible for gas transfer called alveoli (Berg 2016). COPD is a significant public health issue, especially in low‐ and middle‐income countries, where nearly 90% of deaths from COPD occur (WHO 2020). COPD was the third leading cause of global deaths in 2016 with about 3 million deaths (WHO 2018), and it is expected to remain in the third ranking until 2030 (WHO 2013).

The main risk factors for COPD are tobacco smoking, second‐hand smoking, air pollution, and exposure to fuel oil fumes (GOLD 2022). COPD is characterized by a non‐reversible airflow obstruction in the lungs. Exposure to irritants stimulates mucus production and damages cilia that clear away mucus and dirt; this causes air to be trapped inside airways, leading to hyperinflation (GOLD 2022; Ramos 2014). Airflow obstruction in emphysema is due to the loss of elastin, which increases lung compliance and decreases elastic recoil (Costanzo 2019). In other words, the lungs lose their ability to return spontaneously to their resting position after inhalation. COPD is a cause of disability as it affects people's ability to breathe normally and has systemic, severe, and long‐term effects (Agustí 2005). Post‐bronchodilator spirometry is the primary test to measure airflow obstruction. It confirms airflow limitation if the ratio between forced expiratory volume in 1 second (FEV1) and forced vital capacity (FVC) is less than 70% (GOLD 2022).

Clinically, the most common symptoms are dyspnea, chronic cough, wheezing, and sputum production. Dyspnea — also known as shortness of breath or breathlessness — is the most common symptom reported by patients with COPD and is associated with a deterioration in their quality of life and physical activity (Anzueto 2017). Dyspnea results from multiple mechanisms, such as air trapping and dynamic hyperinflation, and it is associated with a significant load on the respiratory muscles (Padula 2006; O'Donnell 2007). The Global Initiative for Chronic Obstructive Lung Disease (GOLD) suggests a therapeutic strategy based on medical history, clinical symptoms, and life quality (ABCD stages) (GOLD 2022).

Description of the intervention

Inspiratory muscle training (IMT), also known as respiratory or ventilatory muscle training, aims to improve inspiratory muscle strength and endurance through a series of breathing exercises. It was developed in the late 1970s (Andersen 1979; Belman 1980; Leith 1976), and it has been used in people with respiratory diseases such as COPD and asthma. IMT focuses on enhancing the performance of respiratory muscles and on improving respiratory symptoms and exercise capacity (Padula 2006).

There are three main categories of IMT devices: threshold loading devices, passive and electronic flow resistive devices, and isocapnic hyperpnea devices (Belman 1994). Other devices exist, and they are reported in Menzes 2018. Most threshold trainers have an adjustable spring‐loaded valve to set the resistance level from 9 cm of water (cmH₂O) to 41 cmH₂O (or from 7 cmH₂O to 41 cmH₂O) and allow changes in resistance by 2 cmH₂O (Menzes 2018). The threshold pressure is independent of the breathing pattern (Geddes 2005). The passive‐resistive trainer contains holes of different diameters: the biggest hole provides the lowest resistance, whereas the narrowest hole offers the highest resistance (McConnell 2004; Menzes 2018). The respiratory load can be selected by turning the dial towards the chosen hole. However, unlike the threshold device, passive‐resistive trainers depend on the inspiratory flow (Wu 2017). The electronic resistive device is similar to the passive‐resistive trainer, and it has the advantage of dynamically adapting the flow resistance (Menzes 2018). Isocapnic hyperpnea trainers are based on low load and high respiratory flow (60% to 90% of maximal voluntary ventilation (MVV)) so that respiratory muscles contract at a higher speed (for an extended time) (McConnell 2004). That device contains a rebreathing bag to maintain physiological rates of CO₂, so patients breathe both fresh air and some of the expired CO₂. In addition to multiple devices of IMT, protocols for this therapy differ between teams in terms of frequency, duration, and supervision (Langer 2015). 

Various measures are used to evaluate respiratory muscle strength (Laveneziana 2019). Maximal static inspiratory mouth pressure (PImax) is the most commonly used technique to assess the strength of the diaphragm and other inspiratory muscles (Pessoa 2014). It is calculated through a mouthpiece connected to a manometer, either at residual volume or at functional residual capacity (Laveneziana 2019). However, this technique requires the co‐operation of patients. Other approaches to assess inspiratory muscle strength exist, such as sniff nasal inspiratory pressure (SNIP) based on a pressure sensor attached to a catheter placed in the nostril (Maillard 1998), phrenic nerve electric transcutaneous stimulation, and phrenic nerve magnetic stimulation (Caruso 2015).

How the intervention might work

Unlike inspiration, expiration is a passive process. In people with COPD, elastin in the lungs can be reduced, leading to incomplete expiration; this means air is trapped in the airways and leads to hyperinflation (Papandrinopoulou 2012). Static and dynamic distention caused by hyperinflation explains, in part, the pathophysiology of respiratory muscle dysfunction in COPD (O'Donnell 2006). Indeed, inspiratory muscles (the diaphragm, intercostal muscles, and the sternocleidomastoid muscle (SCM)) are exposed to an important load generated from hyperinflation and high airway resistance (Caron 2011). In the early stages of COPD, inspiratory muscles try to adapt to these circumstances. For example, type II muscle fibers of the diaphragm switch into type I, which are highly resistant to fatigue (Clanton 2009). There is also an increase in blood capillaries (Doucet 2004) as well as the oxidative capacity (since type I fibers have high mitochondrial density and enzymes that support the oxidative pathway, so the ability to use oxygen will be increased) (Ottenheijm 2008). At an advanced stage of the disease, oxidative stress, gas exchange abnormalities, and the changes in the chest cavity overcome adaptation mechanisms and the diaphragm will be in a position of impaired mechanical advantage. It loses up to 60% of its muscle tissue and becomes shorter and more horizontal, leading to ineffective mechanical function (Caron 2011; Ottenheijm 2008; Salito 2015). A study showed that IMT induced structural and anatomical changes in external intercostal muscles by changing the distribution of type I fibers and increasing the size of type II fibers (Ramirez Sarmiento 2002). Overall, inspiratory muscle weakness is associated with dyspnea and respiratory failure, despite the ability of the diaphragm to adapt itself to hyperinflation (Bégin 1991; Caron 2011). IMT may improve the strength and endurance of these muscles.

In COPD, it is difficult to work on expiration flow and volume due to mechanical changes. However, it is possible to work on inspiration since it is an active process. In other words, strengthening of the inspiratory muscles increases the inspiratory flow (so there will be an increase in the tidal volume (TV)), decreases the inspiratory time, and improves the expiratory time (Beaumont 2018; Charususin 2016).

Why it is important to do this review

The clinical symptoms most often reported by patients with COPD are dyspnea, a decline in exercise capacity and an impairment in their quality of life (Spruit 2013). It is recommended to start pulmonary rehabilitation (PR) as soon as possible, ideally either during hospitalisation or soon after discharge from the hospital (Spruit 2013). Current guidelines recommend an optimal duration of eight weeks for a PR program (Rochester 2015). Usually, PR consists of physiotherapy, nutritional and psychosocial care, patient therapeutic education, and upper and lower limb training (Beaumont 2015; Spruit 2013).

The American Thoracic Society (ATS) reported that IMT may be beneficial as a stand‐alone intervention and when added to PR in patients with respiratory muscle weakness (Spruit 2013). However, the potential effects of combining IMT and PR are still unclear. Indeed, some recent randomized controlled trials (RCTs) did not find significant improvements in patients with severe COPD (Beaumont 2018; Charususin 2018), and most of the published clinical trials and meta‐analyses recommended further investigation (Beaumont 2018a; Gosselink 2011; Langer 2015; Lötters 2002). Furthermore, there is a lack of certainty in the linear relation between PImax, FEV1, and clinical outcomes. In other words, many RCTs showed that IMT improves PImax, but the extent to which this improvement is clinically significant (i.e. the minimal clinically important difference) has not yet been proved (Beaumont 2018; Schultz 2018). The benefits of unsupervised IMT are also unclear (Langer 2015).

Although many meta‐analyses have been published on different modes and modalities of IMT (Geddes 2005; Geddes 2008; Gosselink 2011; Lötters 2002; O'Brien 2008; Smith 1992), several questions remain unanswered. For example, the optimum duration of a training program has not been established, nor has the effect of IMT on dyspnea and quality of life. There is also a need to investigate its additional effect when added to PR. Moreover, the published meta‐analyses are a few years old, and there are now other studies to be included. A recent clinical trial (Langer 2015), showed that factors other than inspiratory muscle weakness might influence the performance of IMT, such as the variety of protocols, and it is worth exploring these.

Objectives

To assess the effect of inspiratory muscle training (IMT) on chronic obstructive pulmonary disease (COPD) as a stand‐alone intervention and when combined with pulmonary rehabilitation (PR).

Methods

Criteria for considering studies for this review

Types of studies

We included published randomized controlled trials (RCTs), as full‐text articles or abstracts, as well as unpublished RCTs. We included abstracts if they reported at least the number of participants in each group, the duration of the intervention, and the training load. We accepted trials with more than two arms. We excluded observational studies, quasi‐RCTs and cross‐over trials since no washout period for IMT has been established.

Types of participants

We included people with COPD diagnosed according to international standards (GOLD 2022), at any stage of the disease. We placed no restrictions on age, duration, setting, or the kind of pulmonary rehabilitation. We planned to include RCTs with different conditions for the same intervention of interest as long as we could obtain the data of participants with COPD separately.

We classified COPD according to GOLD 2022 stages based on the predicted value of Forced Expiratory Pressure in 1 Second (FEV1):

• GOLD 1 ‐ mild: FEV1 ≥ 80% predicted

• GOLD 2 ‐ moderate: 50% ≤ FEV1 < 80% predicted

• GOLD 3 ‐ severe: 30% ≤ FEV1 < 50% predicted

• GOLD 4 ‐ very severe: FEV1 < 30% predicted

Types of interventions

The review consists of two comparisons, as follows.

• IMT plus PR versus PR

• IMT versus no treatment or sham

First, we included trials that explored the benefit of combining IMT and PR compared to PR only. PR consists of, but is not limited to, exercise training, physiotherapy, therapeutic education, and nutritional and psychosocial care (McCarthy 2015). We included different types of IMT irrespective of the mode of delivery: resistance training (high load, low frequency) or endurance training (low load, high frequency), device (i.e. threshold loading, resistive flow device, isocapnic hyperpnea). We made no restrictions on the duration, supervision (home‐based or in a healthcare setting), or timing (during hospitalization or later) of the intervention. We excluded studies where the training was conducted only once per week (face‐to‐face or distance sessions), regardless of the total duration of the clinical trial. The minimum accepted training load was 30% of PImax or more (Hill 2010). We also excluded trials that used a resistive device without controlling the breathing pattern. If a study conducted an incremental training load that started less than 30% of PImax, we considered only the follow‐up from which the load was equal to our threshold.

According to the proportion of supervised sessions, we defined supervision as:

• under 20%: unsupervised;  

• 20% to 70%: partially supervised; and  

• above 70%: fully supervised.

For the second comparison, we compared IMT with control or sham. We defined sham training as using a resistance of less than 30% of PImax. We accepted control groups if they did not receive any intervention or received an intervention other than exercise training to blind participants (e.g. therapeutic education). We made the exception for breathing exercises if participants in the control group did not receive more than one type of training (e.g. diaphragmatic breathing, pursed lips breathing), and the purpose was not to compare it with IMT.

Types of outcome measures

We analysed the following outcomes in the review, but we did not use them as a basis for including or excluding studies.

Primary outcomes

Dyspnea: the essential scale for our primary analysis is the Borg scale (Borg 1982). We only included the Borg score when it was measured at isotime. We analysed all the scales reported by the trials as long as they were validated, and when possible, we combined them in a meta‐analysis. The other included scales were: Baseline and Transition Dyspnea Indexes (BDI‐TDI) (Mahler 2005), and Modified Medical Research Council (mMRC) (Bestall 1999).

Functional exercise capacity: this can be assessed through multiple tests. We did not exclude studies based on the test used. However, for our analysis, we considered that the most important measurement is the six‐minute walk distance (6MWD) (Holland 2014). Therefore, we included it in the summary of findings table and considered it for subgroup analysis. We included other tests and reported them either in qualitative or quantitative analysis.

Health‐related quality of life: we accepted any scales as long as they were validated. This includes the St. George's Respiratory Questionnaire (SGRQ) (Jones 1992), the chronic respiratory questionnaire (CRQ) (Wijkstra 1994) and the COPD assessment test (CAT) (Jones 2009).

Secondary outcomes

Inspiratory muscle strength: measured by maximal static inspiratory mouth pressure (PImax) (Laveneziana 2019).

Laboratory exercise tests: we were primarily interested in the maximal oxygen uptake (VO₂peak), which could be measured through:

• incremental cycle ergometer test;

• endurance cycle ergometer test;

• treadmill test.

Respiratory muscle endurance:

• Respiratory muscle endurance pressure (P_thmax): measured by incremental load testing (Laveneziana 2019). It is the maximally tolerated pressure when breathing against a continuously increasing load.

• Respiratory muscle endurance time (T_lim): measured by constant load testing (Laveneziana 2019). It is the time an individual can maintain breathing against a fixed load. It can be carried out either through a threshold/resistive or isocapnic hyperpnea device.

• Maximal voluntary ventilation (MVV): this is the total volume of expired air between 12 seconds to 15 seconds of deep and fast respiration. MVV is usually compared to predicted MVV (calculated through the forced expiratory volume at 1 second) (Wood 2017).

Respiratory function:

• forced expiratory volume at 1 second (FEV1)

• residual volume

Adverse events: as defined by the trial authors.

We collected outcomes irrespective of the time frame and summary statistics (change from baseline or final values), with a preference for change score. For each trial, we analysed only the outcomes listed above and not all the outcomes reported in the trial. However, we included all the tests and measurements used for the same outcome.

Search methods for identification of studies

Electronic searches

We searched for all published and unpublished RCTs regarding IMT for COPD, in consultation with the Cochrane Airways Information Specialist. We did not apply restrictions on language or publication status (i.e. published, ongoing, or unpublished).

We searched the following databases for relevant trials in October 2020 and we updated our literature search on 23 August 2021 and on 20 October 2022.

• Cochrane Airways Trials Register (Cochrane Airways 2022), via the Cochrane Register of Studies, all years to date (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• Cochrane Central Register of Controlled Trials (CENTRAL), via the Cochrane Register of Studies, all years to date (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• MEDLINE Ovid SP ALL, 1946 to 12 October 2020 (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• Embase Ovid SP, 1974 to week 41 2020 (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• PsycINFO Ovid SP, 1967 to October week 1 2020 (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) EBSCO, 1937 to 13 October 2020 (searched 13 October 2020, 23 August 2021 and 20 October 2022)

• Physiotherapy Evidence Database (PEDro), 1999 onwards (searched 13 October 2020, 23 August 2021 and 20 October 2022)

We searched the following trials registries.

• US National Institutes of Health Ongoing Trials Register, ClinicalTrials.gov(searched 13 October 2020, 23 August 2021 and 20 October 2022)

• World Health Organization International Clinical Trials Registry Platform(searched 13 October 2020, 23 August 2021 and 20 October 2022)

The database search strategies are listed in Appendix 1.  The search strategy was developed in MEDLINE by the Cochrane Airways Information Specialist in collaboration with the authors and peer‐reviewed by another Cochrane Information Specialist using the PRESS checklist (McGowan 2016). The MEDLINE search strategy was then adapted appropriately for each database.

All databases and trial registries were searched from their inception to the present, with no restriction on language or type of publication. Hand‐searched conference abstracts and grey literature were identified through the Cochrane Airways Trials Register and CENTRAL. When possible, A native‐language speaker translated studies written in a language other than English.

Searching other resources

We checked the reference lists of all primary studies and reviews for additional references. We searched for relevant manufacturers' websites for device information. We searched on PubMed for errata or retractions from included studies published in full‐text and, when possible, reported the date this was done within the review.

Data collection and analysis

Selection of studies

Two review authors from complementary disciplines (OA and WF) independently screened the titles and abstracts of the search results using  Covidence, and they coded them as 'include' (eligible or potentially eligible/unclear) or 'exclude'. We retrieved full‐text study reports of all potentially eligible studies, and two review authors (OA and WF) independently screened them for inclusion while also recording the reasons for excluding ineligible studies. We retrieved the full text of potentially relevant reports and removed duplicate records using Covidence.

When appropriate, we contacted the study authors to request further information. We also contacted the manufacturer of the IMT device when we did not understand the concept of training. We resolved disagreements through discussion, without the need for a third review author. We identified and excluded duplicates and collated multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review.

We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (Figure 1) and the 'Characteristics of excluded studies' table (Moher 2009).

1.

Data extraction and management

Two review authors (OA and AK) used a data collection form that we piloted on at least one study in the review to extract characteristics from included studies. We extracted the data using Covidence and an Excel spreadsheet. We planned to record any missing information as unclear or not described. Each form included the following information.

• General information: study ID, author contact detail, and the person who is completing the form

• Methods: aims of the study, study design, total study duration

• Participants: inclusion criteria, exclusion criteria, total number randomized, number randomized per group, mean age, age range, sex, COPD stage, clusters (if applicable), number missing, reasons for missing participants, number of participants moved from one group to another, reasons moved, baseline imbalances, and subgroup analysis

• Intervention/comparison groups: type of group, type of intervention, type of control, duration, supervision, setting, device, intensity, frequency, type of training (strength/endurance)

• Outcomes: primary and secondary outcomes, baseline characteristics, and time points

• Notes: funding for studies and notable conflicts of interest of trial authors

Two review authors (OA and AK) independently extracted outcome data from included studies. We noted in the Characteristics of included studies tables if outcome data were not reported in a usable way or if some of the data were missing. We resolved disagreements by reaching a consensus or by involving a third review author (SK). If we identified multiple reports from the same study, we would extract data from all reports directly into a single data collection form. One review author (OA) transferred data into the Review Manager 5 file (Review Manager 2020). We double‐checked that data had been entered correctly by comparing data presented in the systematic review against the study reports. A second review author (WF) spot‐checked study characteristics for accuracy against the study report.

Assessment of risk of bias in included studies

Two review authors (OA and TL) assessed the risk of bias independently for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021). We resolved any disagreements by discussion or by involving another review author (SK). We assessed the risk of bias according to the following domains.

• Bias arising from the randomization process

• Bias due to deviations from intended interventions

• Bias due to missing outcome data

• Bias in measurement of the outcome

• Bias in selection of the reported result

We used RoB 2 to assess risk of bias in randomized studies (Sterne 2019). We used the RoB 2 Excel tool to complete the risk of bias assessment. We used RevMan Web to generate traffic light plots of the domain‐level judgments for each outcome (RevMan Web 2022).

Our effect of interest was the assignment to the intervention at baseline and our main outcomes were those listed in the summary of findings tables. We judged each outcome as being at low risk, some concerns, or high risk according to the RoB 2 algorithm. We provided a quote from the study report, together with a justification for our judgment, in the risk of bias table. 

Where information on risk of bias relates to unpublished data or correspondence with a trial author, we noted this in the risk of bias table. To detect reporting bias, we compared the study protocol with the published report, and we contacted the study authors to identify missing or partially reported data. If more than 10 studies were included in the meta‐analysis, we created a funnel plot to explore publication bias. None of the included studies was a cluster‐RCT.

We incorporated the risk of bias assessment in the Results section of the review and it was also part of the GRADE assessment of the certainty of evidence (along with precision, directness, consistency, and publication bias). When considering treatment effects, we took into account the risk of bias for the studies that contributed to that outcome. Our primary analysis included all the studies without taking the risk of bias judgments into account.

Assessment of bias in conducting the systematic review

We conducted the review according to the published protocol and reported any deviations from it in the 'Differences between protocol and review' section of this systematic review.

Measures of treatment effect

All of our outcomes were continuous data. We calculated a mean difference (MD) with 95% confidence intervals (CIs) where studies used the same scale, and the standardised mean difference (SMD) with 95% CIs where studies used different scales to measure the same concept. We interpreted SMD analysis following the rule of thumb based on Cohen's d effect size (Cohen 1988):

• 0.2 represents a small effect;

• 0.5 represents a medium effect;

• 0.8 represents a large effect.

Depending on how studies reported ordinal data, we analysed the scales as continuous (since all of them were longer than five). We presented all results with a 95% CI.

We undertook a meta‐analysis when it was meaningful. That is to say, it made sense to combine the data, and the populations, interventions and outcomes were similar enough to be pooled in the same forest plot. If both change‐from‐baseline and end‐point scores were available for continuous data, we used change‐from‐baseline, unless there was a low correlation between measurements in individuals. 

If adjusted analyses were available (analysis of variance (ANOVA) or analysis of covariance (ANCOVA)), we used these as a preference in our meta‐analyses. If the adjusted MD was reported, we included it in the meta‐analysis using the Generic Inverse Variance method unless adjusted and unadjusted analyses were similar.  

Unit of analysis issues

• Cluster‐RCTs and dichotomous outcomes: were not included in the review.

• Repeated observations on participants: if studies reported outcomes at multiple time points, we chose the longest follow‐up period to keep consistent with the studies. We divided the duration of follow‐up into categories to explore possible differences in the effect estimate. More information is in 'Subgroup analysis and investigation of heterogeneity'.

• Studies with more than two groups:  we included only the relevant arms.

• Two comparisons from the same study within the same meta‐analysis: we combined the active arms or halved the control group to avoid double‐counting.

Dealing with missing data

We requested missing or unclear numerical data from study authors (such as for conference abstracts; randomization, the training load). We did not use imputation because most of the data were available, and it was not possible to receive the participants' data. For studies that reported only the overall effect estimate without providing data for each intervention group, we used the generic inverse variance method to meta‐analyse them.

We used the methods recommended by McGrath 2020 to convert median to mean.

When the data were presented only graphically, and in case we could not get numerical information from the study authors, we used WebPlotDigitizer to extract them from the graphs. 

Assessment of heterogeneity

We assessed statistical heterogeneity through visual inspection by detecting overlapping confidence intervals in the forest plot. We used the Chi² test with a P value of 0.05 to indicate the statistical significance and the I² test (Higgins 2003), to explore statistical heterogeneity (we considered a value over 50% to represent substantial heterogeneity). We performed subgroup analysis to investigate heterogeneity. We also discussed clinical heterogeneity (e.g. COPD stages) and methodological heterogeneity (e.g. duration of the intervention, number of sessions per week, the total number of sessions, the training load). 

Assessment of reporting biases

We created funnel plots to explore possible small‐study and publication biases for functional exercise capacity (6MWD) and respiratory muscle strength (PImax) since they were included in the summary of findings table and more than 10 trials explored these outcomes.

Data synthesis

We expected the included studies to have many variables that could influence the pooled effect estimate. Therefore, we used a random‐effects model for the analysis. We ran a meta‐analysis when it was appropriate; that is, where it made sense to combine different effect estimates and the studies were homogeneous enough to allow reliable interpretation of the analysis. 

Ordinal outcomes were meta‐analysed as continuous data following the guidance in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). Two authors (OA and AK) analysed the data using RevMan 5 (Review Manager 2020), and RevMan Web (RevMan Web 2022).

Subgroup analysis and investigation of heterogeneity

We carried out the following subgroup analyses.

• Duration of intervention: short‐term (less than four weeks), medium‐term (between four weeks and seven weeks and six days), long‐term (eight weeks and longer).

• Respiratory muscle strength (PImax): studies with participants with respiratory muscle weakness (the mean baseline PImax of the participants was less than or equal to 60 cmH₂O) or without respiratory muscle weakness (the mean baseline PImax of the participants was greater than 60 cmH₂O). 

We used the following outcomes in the subgroup analyses.

• Dyspnea: Baseline Dyspnea Index‐Transition Dyspnea Index (BDI‐TDI)

• Functional exercise capacity: 6‐minute walk distance (6MWD)

• Respiratory muscle strength (PImax)

We included within‐study data when available. We used the formal test for subgroup interactions in RevMan Web (RevMan Web 2022).

Sensitivity analysis

We performed the following sensitivity analyses.

• We removed from the primary analysis studies judged to be at high risk of bias and some concerns.

• We compared the results of the random‐effects and fixed‐effet models for the BDI‐TDI, the 6MWD, the SGRQ and PImax.

We considered studies to be at high risk of bias overall if we judged at least one of the domains to be high risk. 

Summary of findings and assessment of the certainty of the evidence

We created summary of findings tables (Table 1, Table 2) including the following outcomes.

• Dyspnea: Borg scale, mMRC and BDI‐TDI

• Functional exercise capacity: 6MWD

• Health‐related quality of life: SGRQ, COPD Assessment Test (CAT) 

• Inspiratory muscle strength: PImax

We used the five GRADE considerations (risk of bias, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of the body of evidence as it relates to the studies that contribute data for the prespecified outcomes. Our time point was the end of the study. We used the methods and recommendations described in the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2021), and the GRADE Handbook, using GRADEpro GDT software. We justified all decisions to downgrade the quality of studies using footnotes and we made comments to aid the reader's understanding of the review where necessary.

Results

Description of studies

Details are available in Characteristics of included studies; Characteristics of excluded studies; Characteristics of studies awaiting classification; Characteristics of ongoing studies tables.

Results of the search

We identified from the literature search 7379 records through database searching and 378 from other sources (International Clinical Trials Registry Platform (ICTRP); Epistemonikos). 

After removing duplicates and screening titles and abstracts, we checked the eligibility of 277 studies through full‐text review (see Figure 1). Searching the reference lists of past published systematic reviews did not reveal further eligible records. In the end, we included 55 studies in this review, of which six RCTs had more than two arms. Of the 55 studies, 53 contributed data to meta‐analyses.

When contact details were available, we contacted the study authors for clarification. We classified 42 studies as 'awaiting classification' because of insufficient data (although we contacted study authors) or because we couldn't find the abstract (only the title was available in Covidence), and 15 as ongoing studies.

One study in Chinese (ZhouL 2016), and one in Spanish (Bustamante 2007), were translated into English by native speakers (see Acknowledgements). One study in Japanese (Okura 2020) was translated into English using Google Translate.

Included studies

Comparison 1: pulmonary rehabilitation plus inspiratory muscle training versus pulmonary rehabilitation

Population

We included 22 RCTs with 1446 participants in this comparison. We classified the COPD stages according to post‐bronchodilator forced expiratory volume in one second (FEV1) (GOLD 2022). They ranged from mild to moderate (Tout 2013), moderate to severe (Abedi Yekta 2019; Berry 1996; Dekhuijzen 1991; Larson 1999; Magadle 2007; Majewska‐Pulsakowska 2016; Schultz 2018; Wang 2017), mild to very severe (Paneroni 2018), moderate to very severe (Beaumont 2015; Charususin 2018; Fanfa Bordin 2020; Mador 2005; Tounsi 2021), and severe to very severe (Beaumont 2018; Dellweg 2017; Weiner 1992; Weiner 2000). Three studies did not report COPD stage.

The number of participants in the intervention group (PR+IMT) was 742. The mean age ranged from 51.33 to 70.8 years,  and the mean body mass index (BMI) ranged from 21.31 to 28.8 kg/m². The number of participants in the control group (PR only) was 704. The mean age ranged from 53.5 to 70.8 years, and the mean BMI ranged from 22.4 to 29.68 Kg/m². In the studies that reported gender, there were around 763 men and 482 women.

Intervention

Pulmonary rehabilitation

The rehabilitation programs varied across the studies. They ranged from training with only a treadmill  (Abedi Yekta 2019), with only a cycle ergometer (Larson 1999), and a combination of training protocols in the remaining studies (training with cycle and treadmill, limb muscles strengthening exercises, stair climbing, and therapeutic patient education programs). Five studies monitored training intensity according to heart rate: 50% (Weiner 1992), 60% (Abedi Yekta 2019; Fanfa Bordin 2020), 80% (Dekhuijzen 1991) and 85% (Larson 1999). One study (Tounsi 2021) individualized the training program based on 60% to 80% of the average speed achieved during the six‐minute walk test.

Inspiratory muscle training

Two studies trained their participants for two days a week (Abedi Yekta 2019; Tout 2013), six studies for three days a week (Fanfa Bordin 2020; Mador 2005; Magadle 2007; Wang 2017; Weiner 1992; Weiner 2000), eight studies reported five days a week (Beaumont 2015; Beaumont 2018; De Farias 2019; Dekhuijzen 1991; Dellweg 2017; Larson 1999; Majewska‐Pulsakowska 2016; Paneroni 2018), and four studies trained their participants for the whole week (Berry 1996; Charususin 2018; Schultz 2018; Tounsi 2021). Two studies (Masanga 2011; Sykes 2005) did not provide details about the training. The number of weeks ranged from two weeks (Paneroni 2018) to 24 weeks (Magadle 2007).

The training was unsupervised in one study (Majewska‐Pulsakowska 2016), partially supervised in three studies (De Farias 2019; Larson 1999; Schultz 2018), and fully supervised in the remaining studies. 

Three trials performed endurance training, using SpiroTiger (De Farias 2019; Paneroni 2018), and normocapnic hyperpnea (Mador 2005), respectively. The other RCTs conducted strength training with Powerbreathe devices, Threshold IMT device, and Respifit S‐Unit. Five studies did not change the training loads during the study,   ranging from 30% to 80% of PImax. In the other RCTs, the training load increased from 15% to 60% of PImax (Weiner 2000), from 30% to 60% of PImax (Abedi Yekta 2019; Larson 1999; Majewska‐Pulsakowska 2016; Schultz 2018; Sykes 2005; Tout 2013), from 50% to 80% of PImax (Berry 1996; Tounsi 2021; Weiner 1992), from 35% to 80% (De Farias 2019), from 50% to 60% (Beaumont 2018), from 50% to 84% of PImax (Charususin 2018), and from 66% to 85% of Maximal Voluntary Ventilation (MVV) (Paneroni 2018).

In Magadle 2007 and Weiner 2000, participants received respectively three months and six months of PR before being allocated to continue with PR alone or to receive IMT.

Comparisons

All the studies focused on our main comparison (PR+IMT vs PR), and there were no indirect comparisons. Three RCTs (Abedi Yekta 2019; Larson 1999; Majewska‐Pulsakowska 2016), had four arms, including both comparisons. One trial had three arms (De Farias 2019). We included the appropriate comparison separately. 

Primary outcomes

Dyspnea: two studies explored dyspnea with the Borg scale at isotime and the Modified Medical Research Council (mMRC) scale, one study (Schultz 2018), with BDI‐TDI, and two studies (Beaumont 2015; Beaumont 2018), reported the Multidimensional Dyspnea Profile (MDP).

Functional exercise capacity: 12 studies measured functional exercise capacity with the 6MWD, three studies used the 12‐minute walk distance (12MWD), three studies used exercise time, and five studies used maximal exercise capacity (Wmax). For the latter measurement, Charususin 2018 calculated Wmax by asking participants to cycle at a load of 20 watts (w) and then increasing it by 10 w/min until symptom limitation. Mador 2005 followed a similar protocol starting at no load until the participant could no longer continue cycling for 30 seconds. Dekhuijzen 1991 increased the load by 10% of the predicted Wmax, which was measured through the following formula: "Wmax predicted = 1.7x weight (kg) + 40x FEV,(L)‐25". Wang 2017 chose an incremental load of 5 w/min or 10 w/min.

Health‐related quality of life (HRQoL): seven studies assessed HRQoL with the SGRQ, three studies used the Chronic Respiratory Disease Questionnaire (CRQ), two studies used the COPD Assessment Test (CAT), and one study used the Clinical COPD Questionnaire (CCQ).

Secondary outcomes

Inspiratory muscle strength (PImax): 17 studies reported PImax. Eight of these studies measured it at residual volume. 

Laboratory exercise test (VO₂peak): five trials measured VO₂peak . All studies except for Berry 1996 reported VO₂peak in L/min. So we used the mean weight of each group in Berry 1996 to convert from mL/kg/min to L/min. 

Respiratory muscle endurance strength (P_thmax): two studies measured P_thmax (Larson 1999; Weiner 1992).

Respiratory muscle endurance time (T_lim) was measured by asking participants to breathe until exhaustion against a load of 30% of PImax (Paneroni 2018), 50% to 60% of PImax (Charususin 2018), 70% of PImax (Dekhuijzen 1991), 70% of MVV (Mador 2005), and 70% to 75% of MVV (Paneroni 2018).

FEV1: six studies reported FEV1 as percentage of predicted and liters. Three studies (Berry 1996; Paneroni 2018; Wang 2017), reported MVV, and one study (Charususin 2018), reported residual volume

Only one abstract (Masanga 2011), reported adverse events.

For both our primary and secondary outcomes, we did not include data from Sykes 2005 in our primary analysis. 

Comparison 2: inspiratory muscle training versus control/sham

Population

We included 37 RCTs with 1021 participants in this comparison. As in comparison 1, we classified COPD stages according to post‐bronchodilator FEV1 (GOLD 2022). They ranged from moderate (Leelarungrayub 2017), severe (Lisboa 1997, Weiner 2003; Weiner 2006), mild to severe (Bavarsad 2015), mild to very severe (Dacha 2019), moderate to severe (Abedi Yekta 2019; Belman 1988; Bustamante 2007; Harver 1989; Hsiao 2003; Koppers 2006; Larson 1999; Majewska‐Pulsakowska 2016; Petrovic 2012; Saka 2021; Sanchez Riera 2001; Scherer 2000; Wu 2017; Xu 2018), moderate to very severe (Berton 2015; Chuang 2017; Heijdra 1996; Langer 2018; Nikoletou 2016; Saher 2021), and severe to very severe (Beckerman 2005; Covey 2001; Hill 2006; Hill 2007; Kim 1993; Larson 1988; Preusser 1994; Ramirez Sarmiento 2002; ZhouL 2016). 

The number of participants in the IMT group was 526. The mean age ranged from 51.8 to 70.4 years, and the mean BMI ranged from 19.25 to 29 kg/m². For studies that reported gender, there were in total 268 men and 129 women. The number of participants in the control group (control/sham) was 495. The mean age ranged from 54.2 to 71.1 years, and the mean BMI ranged from 18.54 to 28.8 kg/m². For studies that reported gender, there were 269 men and 132 women

Intervention

Inspiratory muscle training

Participants trained from two days a week (Abedi Yekta 2019), to the whole week (Beckerman 2005; Berton 2015; Bustamante 2007; Dacha 2019; Harver 1989; Kim 1993; Koppers 2006; Langer 2018; Larson 1988; Petrovic 2012; Xu 2018). The duration of the intervention ranged from two weeks (Saher 2021) to a year (Beckerman 2005), and the total duration of training ranged from four hours (Abedi Yekta 2019) to 144 hours (Beckerman 2005).

Eight studies conducted training with resistive devices. Three studies (Belman 1988; Harver 1989; Wu 2017), used Pflex (Respironics Inc, Pittsburgh, PA, USA) device, one study (Leelarungrayub 2017), used Portex (Smith Medical ASD), one study (Hsiao 2003), used Respirex (Respirex®2, DHD 22‐1000, Diemolding Healthcare Division, Canastota, NY, USA), two studies (Heijdra 1996; Sanchez Riera 2001), used INSPIRx (Intertech Resources Inc; Ft. Myers, FL; Respirecare Medical Inc., The Hague, the Netherlands), and one study (Bavarsad 2015), used Respivol (Medinet, Milano, Italy). 

Participants underwent endurance training with Normocapnic Hyperpnea in two RCTs (Koppers 2006; Scherer 2000), and both endurance and strength training in one RCT (Petrovic 2012), using Respifit S (Mauerbach, Austria). The remaining 25 studies conducted IMT with either Threshold IMT or Powerbreathe devices.

Two trials trained their participants 'as tolerated' (Belman 1988; Bustamante 2007), and one trial (Bavarsad 2015), used an incentive spirometer device at a load equal to or more than the inspiratory volume. The training load increased from 30% to 60% of PImax in six trials (Abedi Yekta 2019; Covey 2001; Larson 1999; Majewska‐Pulsakowska 2016; Nikoletou 2016; Saher 2021), 30% to 45% in one trial (Xu 2018), from 15% to 60% of PImax in two trials (Weiner 2003; Beckerman 2005), from 9% to 100% in one trial (Leelarungrayub 2017), from 50% to 100% of PImax in two trials ( Hill 2006; Langer 2018) and approximately from 50% to 133% in one trial (15 to 40 cmH₂0) (Chuang 2017) The remaining 23 studies chose a fixed load that ranged from 30% to 80% of PImax.

The training was fully supervised in eight studies, partially supervised in two studies, and unsupervised in 20 studies. Six studies did not report details of supervision ( Harver 1989; Hill 2007; Petrovic 2012; Saher 2021; Weiner 2003; Weiner 2006).

Control/sham

Twenty‐two studies used a sham IMT while 15 studies did not provide any intervention to the control group. One study (Cutrim 2019), provided diaphragmatic breathing at a rate of 15 to 20 breaths/min for both the intervention and the control groups. Participants in the control group underwent therapeutic patient education and pursed lips breathing (Covey 2001), and therapeutic patient education (Larson 1999).

Comparison

All the studies focused on our main comparison (IMT versus control/sham), and there were no indirect comparisons. Two studies (Hsiao 2003; Wu 2017), had three arms, including two intervention groups (each group used a different device or protocol) and a control group. When two arms from the same study were included in a forest plot, we halved the number of participants in the control group.

As for comparison 1, we extracted the appropriate arms from Abedi Yekta 2019, Larson 1999 and Majewska‐Pulsakowska 2016.

Primary outcomes

Dyspnea: six studies measured dyspnea with the Borg scale at isotime. Eight studies assessed dyspnea with BDI‐TDI and four studies with the mMRC.

Functional exercise capacity: 16 studies measured functional exercise capacity with the 6MWD, three studies with 12MWD, seven studies with Wmax, five studies with exercise time, and two studies with the shuttle walk test (SWT). Hill 2006 and Koppers 2006 measured Wmax by increasing the work rate by 10% per minute; Larson 1999 asked the participants to warm up by pedalling for 3 minutes at 10 w followed by 2 minutes at 20 w, and then they started the graded cycle at 30 w; and Lisboa 1997 increased the load by 75 kpm every 2 minutes. Sanchez Riera 2001, and Wu 2017 increased the work rate by 10 w/min after one minute of unloaded pedalling.

To measure exercise time, Berton 2015 asked the participants to cycle at 75% of Wmax; Koppers 2006 set the load at 50% of PImax; Scherer 2000) measured it on a treadmill set to 80% of the incline and to 100% of the speed reached at VO₂peak; and Wu 2017 considered it as the time to reach Wmax.

Health‐Related Quality of Life (HRQoL): six studies assessed HRQoL with the SGRQ, five studies used CRQ, two studies used CAT, two studies used SF‐36 (Chuang 2017; Nikoletou 2016), and one study used the CCQ  (Leelarungrayub 2017).

Secondary outcomes

Inspiratory muscle strength (PImax):  32 studies reported PImax. Fourteen studies measured it at residual volume, 10 studies at functional residual capacity, one study reported both measurements, and the remaining studies did not report the measurement method.

Laboratory exercise test (VO₂peak): 11 studies reported VO₂peak. 

Respiratory muscle endurance pressure (P_thmax): five studies (Hill 2006; Koppers 2006; Preusser 1994; Ramirez Sarmiento 2002; Weiner 2003), followed the protocol of Nickerson 1982. One study (Larson 1999), started with an initial load of 30% of PImax and increased by 5.7 cmH₂O until exhaustion.

Respiratory muscle endurance time (T_lim): 10 studies reported T_lim. Langer 2018, Nikoletou 2016 and Petrovic 2012 asked the participants to breathe as long as possible against 50% to 60% of PImax, and Hill 2006 and Ramirez Sarmiento 2002 asked them to breathe against 80% of PImax. Bustamante 2007 set the load at 66% of PImax, Hsiao 2003 at 70%, and Scherer 2000 at 66% or 75% of MVV.

MVV: two studies measured MVV (Belman 1988; Harver 1989)

Residual volume: two studies measured residual volume (Ramirez Sarmiento 2002; Hill 2006).   

Forced expiratory volume at 1 second (FEV1): 10 studies reported FEV1 in %Pred, and 12 studies in litrs.

Adverse events: none of the included studies reported adverse events.

Excluded studies

We excluded 133 studies after the full‐text review. For further details, please refer to  Characteristics of excluded studies.

Risk of bias in included studies

We present the risk of bias assessment for each outcome, including all domain judgments and support for judgments, in a spreadsheet (Ammous 2022). We generated traffic light plots in most forest plots of our primary outcomes and for PImax in the secondary outcomes. 

Comparison 1. Pulmonary rehabilitation plus inspiratory muscle training versus pulmonary rehabilitation

• Dyspnea: one study was at low risk of bias for dyspnea (Borg and mMRC; Charususin 2018), while three others were at high risk of bias. Larson 1999 had issues with intention to treat (ITT) analysis (the number of participants that were not analyzed could impact the results), missing data could depend on its true value and participants were not blinded. Both Beaumont 2018 and  Wang 2017 did not blind participants. 

• Functional exercise capacity: three studies were at low risk of bias for functional exercise capacity (6MWD; Beaumont 2018; Charususin 2018; Dellweg 2017). Most studies that we judged at some concern did not provide sufficient details about allocation concealment, excluding participants from the analysis (less than 5%) and only the journal article was available. We considered one study at high risk of bias because missingness is likely to depend on its true value (Paneroni 2018), and another study because of a lack of details about the randomization process (Tout 2013). 

• Health‐related quality of life: no study was at low risk of bias for the SGRQ and CAT. The main issue across the studies was the lack of blinding. One study was at low risk of bias for CRQ (Charususin 2018), and the two others were at high risk of bias because of issues with ITT analysis (Larson 1999) and lack of blinding (Mador 2005). 

• Inspiratory muscle strength: six studies were at low risk of bias (PImax; Beaumont 2018; Charususin 2018; Dellweg 2017; Fanfa Bordin 2020; Schultz 2018; Tounsi 2021). Most studies were at some concern because of a lack of details about allocation concealment and only the journal article was available. Two studies were considered at high risk of bias (Larson 1999; Paneroni 2018), because missingness is likely to depend on its true value and one study because of a lack of details about the randomization process (Tout 2013).

Comparison 2. Inspiratory muscle training versus control/sham 

• Dyspnea: one study was at low risk of bias for dyspnea (Borg; Langer 2018), two studies were at high risk of bias because participants were not blinded (Larson 1988; Petrovic 2012), and the others were at some concern because lack of details about allocation concealment. Similarly, only Langer 2018 was at low risk of bias for BDI‐TDI. We judged Harver 1989 at high risk of bias because the authors did not mention the reasons behind missing data, Weiner 2003 because the data in the graph were different from the text and Wu 2017 because participants were not blinded. For mMRC, from the four included studies, two studies were at low risk of bias (Langer 2018; Xu 2018), one study at some concern because lack of details about allocation concealment  (Saka 2021), and one study at high risk of bias because participants were not blinded (ZhouL 2016). 

• Functional exercise capacity: two studies were at low risk of bias (6MWD; Cutrim 2019; Xu 2018). Four studies were at high risk of bias because of issues with ITT analysis (Beckerman 2005; Hsiao 2003; Ramirez Sarmiento 2002; Saher 2021), lack of details about missing data (Leelarungrayub 2017), and the data in the graph were different from the text (Weiner 2003). The remaining studies were at some concern because of the lack of details about allocation concealment.  

• Health‐related quality of life: only Xu 2018 was at low risk of bias for the SGRQ and CAT. Most of the studies that reported the SGRQ were at some concern because of the lack of details about allocation concealment (Berton 2015; Saka 2021), lack of details in the trial register (Saka 2021), and no information about whether adjusted analysis was planned in advance (Berton 2015). All the studies that reported CRQ were at high risk of bias (mainly because participants were not blinded) and some concern (lack of details about allocation concealment and issues with the reported results). 

• Inspiratory muscle strength: only three studies were at low risk of bias (PImax; Cutrim 2019; Langer 2018; Xu 2018). All the remaining studies did not provide enough details about allocation concealment and studies at high risk of bias had issues with ITT analysis or missingness, or both.

Overall risk of bias

The main issues we found across the studies were the lack of details about allocation concealment (randomization process), lack of blinding (measurement of the outcome, although we considered using a sham IMT equal to blinding participants), and only journal articles were available (selection of the reported results). 

Effects of interventions

See: Table 1; Table 2

Comparison 1: pulmonary rehabilitation plus inspiratory muscle training versus pulmonary rehabilitation

See: Table 1 

Primary outcomes

Dyspnea

Borg scale

Two studies used the Borg scale at isotime to evaluate dyspnea. Considering a MCID of −1 unit (Ries 2005), there was no improvement in dyspnea with an overall effect estimate (MD 0.19, 95% CI −0.42 to 0.79; I² = 0%; 2 studies, 202 participants;  Analysis 1.1). We judged only one study (Charususin 2018), to be at low risk of bias, which revealed a similar effect estimate. One study (Sykes 2005), reported greater improvement in dyspnea at "heavy load" in the IMT+ exercise group but did not report numerical data.

1.1. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 1: Dyspnea: Borg (at submaximal exercise: 50% to 80% of Wmax)

Using GRADE, we downgraded the certainty of the evidence for Borg by 1 point due to serious concerns regarding imprecision. 

mMRC scale

We did not find any improvement in dyspnea using the mMRC scale (MD −0.12, 95% CI −0.39 to 0.14; I² = 0%; 2 studies, 204 participants;  Analysis 1.2). The MCID is estimated to be between −0.5 to −1 unit (Araújo 2017; Cazzola 2015). 

1.2. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 2: Dyspnea: Modified Medical Research Council (mMRC)

We downgraded the certainty of evidence to very low due to very serious concerns with risk of bias and serious concerns with imprecision.

BDI‐TDI and MDP

There was no significant difference between the two arms in the studies that reported the BDI‐TDI (Schultz 2018), and the MDP (Beaumont 2015; Beaumont 2018).

Functional exercise capacity

6MWD

We pooled the 12 studies that reported the 6MWD test in a meta‐analysis that showed no evidence of a difference between groups (MD 5.95, 95% CI −5.73 to 17.63; I² = 61%; 12 studies, 1199 participants;  Analysis 1.3). The mean and upper bounds of the 95% CI were lower than the MCID of 26 meters indicating that the mean change was not clinically relevant (Puhan 2011).  Sykes 2005 narratively reported a better 6MWD in the intervention group. 

1.3. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 3: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters)

We had very low confidence in the results due to serious concerns with risk of bias and very serious concerns with inconsistency.

We conducted a subgroup analysis according to the duration of the intervention (short, medium and long‐term interventions). The test for subgroup differences was not significant (Chi² = 0.30, df = 2 (P = 0.86), I² = 0%, Analysis 1.4). 

1.4. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 4: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention)

In the following subgroup analysis, we divided the studies according to their baseline PImax and the test for subgroup differences was not significant (Chi² = 1.94, df = 1 (P = 0.16), I² = 48.3%, Analysis 1.5).

1.5. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 5: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness)

One study (Wang 2017), reported a subgroup analysis for intervention group participants with or without respiratory muscle weakness, with greater improvement in 6MWD for the weakened respiratory muscle group.  

In sensitivity analysis, keeping just the studies at low risk of bias (Beaumont 2018; Charususin 2018; Dellweg 2017), increased the mean difference without exceeding the MCID (MD 8.90, 95% CI −11.86 to 29.65; I² = 80%; 4 studies, 379 participants). However, it decreased when switching to the fixed‐effect model (MD 0.73, 95% CI ‐4.80 to 6.26, I² = 61%).

12MWD 

Three studies reported  the 12MWD and showed a larger effect in the intervention group (MD 155.77 meters, 95% CI −84.53 to 396.08; I² =  79%; 80 participants; Analysis 1.6)

1.6. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 6: Functional exercise capacity: 12‐minute walk distance (12MWD) (meters)

Wmax

Five studies reported Wmax and showed no difference between the groups (MD −1.01 watts, 95% CI −6.96 to 4.94;  I² = 25%; 326 participants; Analysis 1.7).

1.7. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 7: Functional exercise capacity: Wmax (watt)

Exercise time (seconds)

Four studies reported exercise time (seconds) and also showed a larger effect in the intervention group (MD 58.62 seconds, 95% CI −25.09 to 142.32; I² = 0%; 192 participants;  Analysis 1.8). We did not include Berry 1996 in the meta‐analysis because they did not specify the level of exercise at which they measured exercise time. However, it is unclear to what extent these changes are clinically relevant.

1.8. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 8: Functional exercise capacity: exercise time (seconds)

Health‐related quality of life (HRQoL)

SGRQ

We pooled seven trials that explored the effect of the intervention on HRQoL using the SGRQ, which showed no difference in total scores (MD 0.13, 95% CI −0.93 to 1.20; I² = 0%; 908 participants; Analysis 1.9). This mean difference did not exceed the MCID threshold of −4 units (Welling 2015). 

1.9. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 9: Health‐related quality of life (HRQoL): St George's Respiratory Questionnaire (SGRQ)

We downgraded the certainty of evidence by two levels due to very serious concerns with risk of bias.

Two studies that reported the three SGRQ domains showed no differences in symptoms (MD −2.33, 95% CI −6.28 to 1.62; Analysis 1.9), activity (MD 0.28, 95% CI −1.65 to 2.20) or impact (MD −1.63, 95% CI −5.38 to 2.11) scores. 

In sensitivity analysis, effect estimates remained unchanged with a fixed‐effect model. We could not explore the summary effect of low risk of bias studies because all the studies were at high risk of bias and some concerns.

CRQ

There were no differences in CRQ domain scores (Analysis 1.10): three RCTs reported 'Dyspnea' (MD −0.30, 95% CI −1.90 to 1.29) and 'Fatigue' (MD 0.28, 95% CI −0.76 to 1.31). Two RCTs reported 'Emotion' (MD −0.63, 95% CI −2.53 to 1.26) and 'Mastery' (MD −0.05, 95% CI −1.18 to 1.08). We used the generic inverse variance method to pool the results because Charususin 2018 showed a different adjusted mean difference compared to non‐adjusted analysis. None of the differences exceeded the MCID threshold of +0.5 units (Alma 2018). One abstract (Sykes 2005), narratively reported an improved  CRQ score in the intervention group.

1.10. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 10: Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ)

CAT

Two studies that reported the CAT scale showed no difference between groups (MD 0.13, 95% CI −0.80 to 1.06; I² = 50%; 657 participants;  Analysis 1.11) and the difference did not exceed the threshold of clinical significance, MCID of −1.6 units (American Thoracic Society). We considered the certainty of evidence to be very low due to very serious concerns with risk of bias and serious concerns with inconsistency.

1.11. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 11: Health‐related quality of life (HRQoL): COPD Assessment Test (CAT)

Secondary outcomes

Inspiratory muscle strength: PImax (cmH₂O)

Aggregated data from 17 studies showed higher inspiratory muscle strength in the intervention group (MD 11.46, 95% CI 7.42 to 15.15; I² = 84%; 1329 participants; Analysis 1.12), but this did not exceed the MCID of 17.2 (Iwakura 2020). The statistical heterogeneity is due to the studies that reported change from baseline and had narrow confidence intervals.

1.12. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 12: Inspiratory muscle strength: PImax (cmH20)

We downgraded the certainty of evidence by one level due to serious concerns with risk of bias.

We performed a subgroup analysis according to the duration of intervention, which did not reveal a difference between subgroups (Chi² = 0.05, df = 2 (P = 0.98), I² = 0%; Analysis 1.13).  

1.13. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 13: Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: duration of the intervention)

We carried out subgroup analysis between studies with participants with and without respiratory muscle weakness. The test for subgroup differences was not significant (Chi² = 1.72, df = 1 (P = 0.19), I² = 41.8%; Analysis 1.14).

1.14. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 14: Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: with or without respiratory muscle weakness)

Wang 2017 divided the participants of the intervention group according to the state of their respiratory muscles, and they reported a larger change in PImax with the normal respiratory muscle group. Sykes 2005 reported a higher PImax narratively in the intervention group.

Sensitivity analysis showed minimal impact of fixed‐effect models on the synthesised results (MD 10.53, 95% CI 9.25 to 11.81, I² = 84%). Retaining studies at low risk of bias (Beaumont 2018; Charususin 2018; Dellweg 2017; Fanfa Bordin 2020; Schultz 2018; Tounsi 2021), also had minimal impact (MD 13.43, 95% CI 11.81 to 15.04;  I² = 90%; 1008 participants). 

Laboratory exercise test: VO₂peak

We combined five studies showing no additional effect of the intervention on VO₂peak (MD −0.01 L/min, 95% CI −0.05 to 0.03; I² = 0%; 313 participants; Analysis 1.15). Sykes 2005 reported no difference between the two groups, but without numerical data. 

1.15. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 15: Laboratory exercise test: VO2peak (L/min)

Respiratory muscle endurance pressure (P_thmax) (cmH₂O)

Two studies reported P_thmax, and we pooled them in an SMD meta‐analysis because they used different techniques to measure the outcome. We got an overall effect estimate (SMD 1.22 cmH₂O, 95% CI −0.18 to 2.66; I² = 80%; 52 participants; Analysis 1.16), which suggests a large effect according to Cohen's d effect size. 

1.16. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 16: Respiratory muscle endurance: respiratory muscle endurance pressure (P_thmax) (cmH₂O)

Respiratory muscle endurance time: T_lim (seconds)

We performed separate analyses according to the nature of the endurance test. We pooled three studies that measured the outcome through sustained ventilation according to %PImax (MD 84.62, 95% CI −50.77 to 220.02; 236 participants; Analysis 1.17), and two studies that measured it according to %MVV (MD 477.69, 95% CI 215.43 to 739.94; 51 participants; Analysis 1.18). For both methods of measurement, we found a larger effect in the intervention group.

1.17. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 17: Respiratory muscle endurance time: Tlim (seconds) (sustained ventilation according to PImax)

1.18. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 18: Respiratory muscle endurance time: Tlim (seconds) (sustained ventilation according to MVV)

Maximal voluntary ventilation (MVV)

We pooled in an SMD meta−analysis two studies that reported MVV in L/min (Berry 1996; Wang 2017), and one study in %Pred (Paneroni 2018). We got an overall effect estimate (SMD 0.40, 95% CI −0.02 to 0.83; I² = 4%; 93 participants; Analysis 1.19), which suggests a moderate effect according to Cohen's d effect size.

1.19. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 19: Maximal voluntary ventilation (MVV)

Respiratory function: forced expiratory volume at 1 second (FEV1)

We combined the studies according to whether they reported FEV1 in %Pred and in liters. We did not find a better effect of PR+IMT, showing an overall effect estimate (MD 0.77, 95% CI −1.72 to 3.26; I² = 0%; 6 studies, 173 participants; Analysis 1.20) and (MD 0.04, 95% CI −0.04 to 0.13; I² = 56%; 6 studies, 889 participants; Analysis 1.21) respectively.

1.20. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 20: Respiratory function: forced expiratory volume at 1 second (FEV1) (%Pred)

1.21. Analysis.

Comparison 1: PR+IMT vs PR, Outcome 21: Respiratory function: forced expiratory volume at 1 second (FEV1) (Liters)

Respiratory function: residual volume

One study reported this outcome (Charususin 2018) and did not show a difference between the two groups.

Adverse events

One abstract (Masanga 2011) reported some adverse effects that were considered "minor and self‐limited": headache (six participants), jaw pain (six participants), neck pain (six participants), back pain (four participants), abdominal pain (two participants), cough (one participant), blood‐streaked sputum (one participant), shoulder pain (one participant) and chest pain (one participant). 

Comparison 2: inspiratory muscle training versus control/sham

See: Table 2

Primary outcomes

Dyspnea

Borg scale

Six studies used the Borg scale to assess dyspnea at isotime, and we entered them into a meta‐analysis. Breathlessness was lower with the intervention and the mean difference was close to the MCID of −1 unit (the lower limit of the 95% CI exceeded it), though results are imprecise (MD −0.94, 95% CI −1.36 to −0.51; I² = 0%; 6 studies, 144 participants; Analysis 2.1). Only one study was at low risk of bias.

2.1. Analysis.

Comparison 2: IMT vs control/sham, Outcome 1: Dyspnea: Borg (at submaximal exercise capacity)

We downgraded the certainty by three levels due to very serious concerns with risk of bias and serious concerns with imprecision.

BDI‐TDI

Eight studies (nine arms) used BDI‐TDI to measure dyspnea (Analysis 2.2). Three studies (four arms) reported the TDI 'Change in Functional impairment' (MD 0.88, 95% CI 0.51 to 1.25), TDI 'Change in Magnitude of task' (MD 0.73, 95% CI 0.35 to 1.12), and TDI 'Change in Magnitude of effort' (MD 0.86, 95% CI 0.42 to 1.30), showing no potential improvement in dyspnea, according to an MCID of +1 unit (Mahler 2005). Eight studies (nine arms) reported the TDI 'Focal score' and revealed a greater effect with IMT (MD 2.98, 95% CI 2.07 to 3.89; I² = 65%; 238 participants). 

2.2. Analysis.

Comparison 2: IMT vs control/sham, Outcome 2: Dyspnea: Baseline and Transition Dyspnea Indexes (BDI‐TDI)

We created a subgroup analysis of the TDI 'Focal score' according to studies judged with or without respiratory muscle weakness. The test for subgroup differences was not significant (Chi² = 2.55, df = 1 (P = 0.11), I² = 60.8%; Analysis 2.3). 

2.3. Analysis.

Comparison 2: IMT vs control/sham, Outcome 3: Dyspnea: Transition Dyspnea Index (TDI): Focal score (subgroup analysis: with or without respiratory muscle weakness)

In sensitivity analysis, the overall effect estimates remained unchanged when switching to the fixed‐effect model in the first three subgroups, while it increased to (MD 4.04, 95% CI 3.70 to 4.39, I² = 65%) with 'Focal score'. This effect mainly resulted from (Sanchez Riera 2001), which got the greatest weight.

Only one study was at low risk of bias (Langer 2018), and we considered the certainty of evidence to be very low due to very serious concerns with risk of bias and serious concerns with imprecision.

mMRC

Four studies reported the mMRC score, and the overall effect estimate revealed a possible improvement in dyspnea in the IMT group (MD −0.59, 95% CI −0.76 to −0.43; I² = 17%; 150 participants; Analysis 2.4), considering an MCID between −0.5 and −1 unit. We considered the certainty of evidence to be low due to serious concerns with risk of bias and imprecision.

2.4. Analysis.

Comparison 2: IMT vs control/sham, Outcome 4: Dyspnea: Modified Medical Research Council (mMRC)

Functional exercise capacity

6MWD

We combined 16 studies (17 arms) that reported the 6MWD, showing a better effect with the IMT compared to control/sham (MD 35.71, 95% CI 25.68 to 45.74; I² = 16%; 501 participants; Analysis 2.5).  

2.5. Analysis.

Comparison 2: IMT vs control/sham, Outcome 5: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters)

We judged the certainty of evidence to be moderate due to serious concerns with risk of bias.

In the first subgroup analysis, we divided the studies according to the duration of the intervention. Only one study (Saher 2021), had a follow‐up of fewer than four weeks. The test for subgroup differences was not significant (Chi² = 0.37, df = 2 (P = 0.83), I² = 0%; Analysis 2.6).

2.6. Analysis.

Comparison 2: IMT vs control/sham, Outcome 6: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention)

In the following subgroup analysis, we divided the studies according to the mean baseline PImax. The test for subgroup differences was not significant (Chi² = 0.17, df = 1 (P = 0.68), I² = 0%; Analysis 2.7). One trial (Xu 2018), performed a subgroup analysis within study data using the same cut‐off. 

2.7. Analysis.

Comparison 2: IMT vs control/sham, Outcome 7: Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness)

In sensitivity analysis, keeping just the studies at low risk of bias (Cutrim 2019; Xu 2018), revealed a larger effect, standing at around twice our MCID (MD 49.13, 95% CI −27.62 to 125.88, I² = 83%). However, we should note that its confidence interval exceeded the line of no effect. The overall effect estimate remained nearly the same when switching to the fixed‐effect model.

12MWD and Wmax

We did not find an additional effect of IMT when we pooled the studies that used the 12MWD and Wmax, revealing, respectively, an overall effect estimate (MD −33.31, 95% CI −158.10 to 91.48; I² = 59%; 3 studies, 101 participants = 101; Analysis 2.8) and (MD 0.66, 95% CI −6.44 to 7.76; I² = 42%; 7 studies, 206 participants; Analysis 2.9). However, for the 12MWD, we noticed large differences in baseline data between the two groups of Preusser 1994, which appeared significant when testing it with the RevMan calculator (P = 0.03). We removed that study in a sensitivity analysis, and we got a positive overall effect (MD 12.76, 95% CI −65.71 to 91.23, I² = 0%).

2.8. Analysis.

Comparison 2: IMT vs control/sham, Outcome 8: Functional exercise capacity: 12‐minute walk distance (12MWD) (meters)

2.9. Analysis.

Comparison 2: IMT vs control/sham, Outcome 9: Functional exercise capacity: Wmax (watt)

Exercise time

Five studies (six arms) reported exercise time, but we did not pool the studies because they used different measurement methods (Analysis 2.10). Globally, their results were consistent and showed a trend of a greater effect in the IMT group compared to the control/sham group.

2.10. Analysis.

Comparison 2: IMT vs control/sham, Outcome 10: Functional exercise capacity: exercise time (seconds)

SWT

Three studies used the SWT, and we included two trials in our primary analysis. We did not find an additional effect in the IMT group (MD −7.45 meters, 95% CI −92.74 to 77.83; Analysis 2.11). 

2.11. Analysis.

Comparison 2: IMT vs control/sham, Outcome 11: Functional exercise capacity: shuttle walk test (SWT) (meters)

Health‐related quality of life (HRQoL)

SGRQ

Two studies (Berton 2015; Saka 2021), reported the items 'Symptoms' (MD −2.10, 95% CI −3.50 to −0.71), 'Activity' (MD −9.86, 95% CI −15.08 to −4.63) and  'Impact' (MD −6.06, 95% CI −13.76 to 1.65). Six studies reported the 'total score' of the SGRQ, showing a larger effect in the IMT group (MD −3.85, 95% CI −8.18 to 0.48; I² = 66%; 182 participants;  Analysis 2.12). The lower limit of the 95% CI exceeded the MCID of −4 units. The statistical heterogeneity is due to the effect of Saka 2021. Only one study was at low risk of bias (Xu 2018), and it has a similar effect estimate to the overall. The overall effect estimate increased to MD −5.11 (95% CI −6.81 to −3.40, I² = 66%), when switching to the fixed‐effect model. 

2.12. Analysis.

Comparison 2: IMT vs control/sham, Outcome 12: Health‐related quality of life (HRQoL): St George Respiratory Questionnaire (SGRQ) 

We judged the certainty of the evidence as 'very low' due to very serious concerns with risk of bias and imprecision. 

CRQ

Four studies (5 arms) reported the four items of the CRQ, while one trial (Larson 1999), reported only dyspnea and fatigue. We divided the items into subgroups. The RCTs showed consistent results across the items, and there was an improvement (considering an MCID of +0.5 unit) in 'Dyspnea' (MD 1.63, 95% CI 0.23 to 3.03; Analysis 2.13), 'Fatigue' (MD 1.32, 95% CI 0.08 to 2.55), 'Emotion' (MD 2.64, 95% CI 0.82 to 4.46), and 'Mastery' (MD 1.57, 95% CI 0.07 to 3.06). None of the included studies was at low risk of bias.

2.13. Analysis.

Comparison 2: IMT vs control/sham, Outcome 13: Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ)

CAT

Two trials reported CAT, revealing an overall effect estimate (MD −2.97, 95% CI −3.85 to −2.10; participants = 86; I² = 0%; Analysis 2.14). We downgraded the certainty of evidence by one level due to serious concerns with imprecision.

2.14. Analysis.

Comparison 2: IMT vs control/sham, Outcome 14: Health‐related quality of life (HRQoL): COPD Assessment Test (CAT)

SF‐36 and CCQ

In the studies that used SF‐36 (Chuang 2017; Nikoletou 2016), and CCQ (Leelarungrayub 2017), there was a better effect in the IMT group in some domains of the scales.

Secondary outcomes

Inspiratory muscle strength: PImax (cmH₂O)

PImax

32 RCTs (34 arms) measured PImax. The upper limit of the 95% CI of summary effect between IMT and control/sham exceeded the MCID  of 17.2 cmH₂O (MD 14.57 cmH₂O, 95% CI 9.85 to 19.29; I² = 89%; 916 participants; Analysis 2.15). 

2.15. Analysis.

Comparison 2: IMT vs control/sham, Outcome 15: Inspiratory muscle strength: PImax (cmH₂O)

We downgraded the certainty of evidence by two levels due to serious concerns with risk of bias and a strongly suspected publication bias.

We divided the studies according to the training duration (Analysis 2.16). The test for subgroup differences was not significant (Chi² = 0.67, df = 2 (P = 0.72), I² = 0%).

2.16. Analysis.

Comparison 2: IMT vs control/sham, Outcome 16: Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: duration of the intervention)

We carried out subgroup analysis between studies with participants with and without respiratory muscle weakness. The test for subgroup differences was not significant (Chi² = 0.34, df = 1 (P = 0.56), I² = 0%; Analysis 2.17).

2.17. Analysis.

Comparison 2: IMT vs control/sham, Outcome 17: Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: with or without respiratory muscle weakness)

We also conducted a subgroup analysis of the measurement method. The test for subgroup differences was not significant (Chi² = 0.93, df = 2 (P = 0.63), I² = 0%; Analysis 2.18) between studies that measured PImax at residual volume, at functional residual capacity, and the studies that did not report the level of measurement.

2.18. Analysis.

Comparison 2: IMT vs control/sham, Outcome 18: Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: method of measurement)

In sensitivity analysis, keeping just the studies at low risk of bias (Cutrim 2019; Langer 2018; Xu 2018), showed a similar result (MD 12.79 cmH₂O, 95% CI 3.63 to 21.95, I² = 55%). Similarly, the overall effect estimate of the random‐effects model did not differ from the fixed‐effect model. 

Laboratory exercise test (VO₂peak)

We combined 11 studies (12 arms) in an SMD meta‐analysis, since they were reported in different units (L/min, mL/min, and mL/kg/min). We got an overall effect estimate suggesting a low to moderate effect according to Cohen's rule of thumb (SMD 0.31, 95% CI 0.05 to 0.57; I² = 14%; 286 participants; Analysis 2.19).

2.19. Analysis.

Comparison 2: IMT vs control/sham, Outcome 19: Laboratory exercise test: VO₂peak

Respiratory muscle endurance pressure (P_thmax) (cmH₂O)

Eight RCTs used P_thmax in their respiratory endurance assessment, and they showed a larger overall effect in the IMT group (MD 9.71, 95% CI 4.93 to 14.50; I² = 53%; 179 participants; Analysis 2.20).

2.20. Analysis.

Comparison 2: IMT vs control/sham, Outcome 20: Respiratory muscle endurance: respiratory muscle endurance pressure (P_thmax) (cmH₂O)

Respiratory muscle endurance time T_lim (seconds)

We pooled 10 studies (11 arms) that reported T_lim, and they showed a better improvement in the outcome in the IMT group (MD 270.57, 95% CI 182.44 to 358.71; I² = 63%; 260 participants; Analysis 2.21). One trial (Heijdra 1996) considered it as the time to reach P_thmax (not included in the analysis).

2.21. Analysis.

Comparison 2: IMT vs control/sham, Outcome 21: Respiratory muscle endurance time: T_lim (seconds)

Maximal voluntary ventilation (MVV)

Two studies reported MVV. We pooled them in an SMD meta‐analysis because Belman 1988 did not report the unit. We got a summary effect suggesting a large effect according to Cohen's d effect size (SMD 0.99, 95% CI 0.28 to 1.69; I² = 0%; 36 participants; Analysis 2.22).

2.22. Analysis.

Comparison 2: IMT vs control/sham, Outcome 22: Maximal voluntary ventilation (MVV)

Respiratory function: forced expiratory volume at 1 second

We ran two analyses according to FEV1 unit without finding a significant difference between IMT and control/sham in either unit. Ten studies (11 arms) reported FEV1 in %Pred showing a difference of (MD 2.62 %predicted, 95% CI 0.20 to 5.04 Analysis 2.23), while 12 studies (13 arms) reported it in liters (MD 0.04 L, 95% CI −0.06 to 0.14 Analysis 2.24).

2.23. Analysis.

Comparison 2: IMT vs control/sham, Outcome 23: Respiratory function: forced expiratory volume at 1 second (FEV1) (%pred)

2.24. Analysis.

Comparison 2: IMT vs control/sham, Outcome 24: Respiratory function: forced expiratory volume at 1 second (FEV1) (Liters)

Respiratory function: residual volume 

Hill 2006 reported residual volume in %Pred and liters, showing a better improvement in the IMT group, while Ramirez Sarmiento 2002 reported it in liters and did not show a difference.

Adverse events

None of the included studies reported a side effect of the intervention.

Discussion

This review has summarised the available evidence of the effect of inspiratory muscle training, used alone or in combination with pulmonary rehabilitation, in people with COPD. We combined the results of the studies in a logical way, so that no data were lost, taking into account the variety of scales, respiratory measurements and training protocols.

Summary of main results

We included 55 RCTs in this review (including trials with more than two arms). Twenty‐two studies (1446 participants) investigated the effect of PR with IMT compared to PR, while 37 studies (1021 participants) focused on the effect of IMT compared to control/sham. Three trials (Abedi Yekta 2019; Larson 1999; Majewska‐Pulsakowska 2016), explored both comparisons (four arms).

Comparison 1: pulmonary rehabilitation plus inspiratory muscle training versus pulmonary rehabilitation

In our first comparison (PR+IMT vs PR), we did not find a significant improvement in the intervention group in dyspnea measured with the Borg scale at isotime, referring to an MCID of −1 unit (moderate‐certainty evidence). We also did not find a potential effect with the mMRC scale, based on an MCID between −0.5 to −1 unit (very low‐certainty evidence.

Studies assessed functional exercise capacity with four measurements: 6MWD, 12MWD, Wmax, and exercise time. We did not find an additional effect of combining PR and IMT with the 6MWD, which we considered our main measurement for this outcome. The overall effect estimate did not reach the MCID of 26 meters (low‐certainty evidence). In subgroup analysis, we divided the studies according to the training duration and mean baseline PImax. The test for subgroup differences was not significant. In sensitivity analysis, three studies at low risk of bias showed a relatively greater treatment effect.

The overall effect estimate of the 12MWD and exercise time showed a larger effect in the intervention group. However, it remains unclear to what extent this difference is clinically relevant because we did not find an MCID for this outcome. The studies that measured Wmax did not reveal a difference between the two groups.

Seven studies investigated the effect of the intervention on HRQoL through the SGRQ. The overall effect estimate showed a positive effect in two domains ('Symptoms' and 'Impact'), without exceeding the MCID of −4 units. There was no difference between the two groups with 'Impact' and 'Total score' (low‐certainty evidence).

In the CRQ, we only found a clinically relevant difference (−0.5 units) in the 'Emotion' domain. The other items ('Dyspnea', 'Fatigue' and 'Mastery') did not show a difference, noting that only three studies reported this scale. 

Two studies reported the CAT scale, and the overall effect estimate failed to reveal a benefit based on an MCID of about −1.6 units (very low‐certainty evidence).

The combination of PR and IMT might have a greater effect on inspiratory muscle strength (PImax) than PR alone, but our treatment effect failed to reach the MCID (moderate‐certainty evidence). In subgroup analysis, we did not find a difference between different training durations nor between studies with participants with or without respiratory muscle weakness.

The overall effect estimate of VO₂peak, MVV, FEV1 (%Pred and L), and residual volume did not reveal a difference between the two interventions. For respiratory muscle endurance tests (T_lim and P_thmax), we discovered a more significant effect in the PR plus IMT. Nonetheless, the clinical relevance of these estimations remained doubtful since we did not find an MCID.

Comparison 2: inspiratory muscle training versus control/sham

In our second comparison, IMT versus control/sham, taking account of the Borg MCID (−1 unit), we found a trend of an improvement in dyspnea in the IMT group (only the lower limit of the 95% CI exceeded the MCID) when the scale was measured at submaximal exercise capacity (very low‐certainty evidence). 

The studies that used BDI−TDI did not reveal a clinically meaningful change (MCID +1 unit) after IMT in the three items of the scale (Functional impairment, Magnitude of task, Magnitude of effort). On the other hand, we found a larger effect with 'Focal score' of the TDI (very low‐certainty evidence). The difference between the items might be explained by the fact that more studies reported Focal score (N = 8) than the other items (N = 3). In subgroup analysis, we did not find a difference between studies with participants judged with and without respiratory muscle weakness. 

The overall effect estimate of the mMRC showed a possible effect with IMT that exceeded the MCID of −0.5 units (low‐certainty evidence).

Turning to functional exercise capacity, we combined 16 studies (17 arms) that used the 6MWD, showing a larger effect exceeding the MCID: 26 meters (moderate‐certainty evidence). Following the same subgroup analysis of our first comparison, we did not find a difference between different training durations and between studies with participants judged with and without respiratory muscle weakness. In sensitivity analysis, studies at low risk of bias (N = 2) showed a greater treatment effect.

Studies also used other measurements to assess exercise capacity, and the overall estimation of the 12MWD and Wmax did not show an advantageous effect of IMT. We did not combine the studies that reported exercise time because the methods of measurement were inconsistent. Still, all the effect estimates were on the right‐hand side of the line of no effect with different degrees of positive effect.

We pooled separately three scales (SGRQ, CRQ and CAT) that assessed the HRQoL in a meta‐analysis. All the scales revealed a better improvement in life quality with IMT, but unlike CRQ and CAT, only the lower limit of the 95% CI of SGRQ exceeded the MCID (−4 units; very low‐certainty evidence). We noticed a larger effect estimate favouring the intervention in some items when other scales were used (CAT, SF‐36, CCQ).

PImax was the main secondary outcome in our second comparison. There was a trend to a greater effect with IMT, considering an MCID of 17.2 cmH₂O (low‐certainty evidence). We did not find a difference between medium‐term and long‐term training nor between studies with participants with or  without respiratory muscle weakness. Measuring the outcome at residual volume or functional residual capacity showed similar results. We got similar results with studies at low risk of bias (N = 3).

We combined the studies that reported VO₂peak in an SMD meta‐analysis. According to Cohen's rule of thumb, we considered the overall effect estimate between low to moderate.

The studies that reported respiratory muscle endurance tests (P_thmax, T_lim, MVV) showed a beneficial effect favouring IMT. However, we could not judge the clinical relevance of this effect since we did not find the MCID of these outcomes. The overall effect estimate of FEV1 (%Pred, L) did not reveal a potential improvement with IMT compared to control/sham. 

Overall completeness and applicability of evidence

This review was conducted on patients with stable COPD, aged 44 years and over. It explored a wide range of participants and included all RCTs regardless of COPD stages and training protocols (duration, load, supervision, devices, follow‐up, measurements). Most studies, except Ahmad 2013 excluded participants who were free from exacerbation for a couple of weeks preceding the trial and were not hospitalised. Therefore, the results of this review may not be applicable to IMT conducted just after an acute exacerbation. We also noticed a shift over the years from using resistive devices to threshold devices. We excluded RCTs that used resistive trainers without controlling the breathing pattern because participants could adapt a non‐fatiguing respiratory pattern without exposing the respiratory muscles to the workload. Currently, with technological development, there has been increasing use of electronic devices that allow remote monitoring and accurate adjustment of the training load.  

Studies that looked at the combined effect of PR plus IMT included people with mild to very severe COPD. One study (Dellweg 2017), performed the intervention on hypercapnic patients who remained dependent on non‐invasive ventilation after prolonged weaning. Two studies (Beaumont 2015; Charususin 2018), specifically trained participants with reduced PImax, and two trials did subgroup analysis within study data according to the state of respiratory muscles (PImax ≤ or > 60 cmH₂O) (Beaumont 2018; Wang 2017). Participants in Magadle 2007 and Weiner 2000 received respectively three months and six weeks of PR before the start of IMT, while most of the other studies excluded this category of patients. These two studies had no impact on the overall effect size when we removed them in a sensitivity analysis. Apart from De Farias 2019, Mador 2005 and Paneroni 2018, which conducted endurance training, all the studies performed strength training with various loads and different training durations.

Furthermore, there was a wide range of PR protocols, starting from just breathing exercises and respiratory drainage to a mix of interventions (e.g. cycling, treadmill, muscle strengthening, therapeutic patient education). To the best of our knowledge, the current guidelines have not yet established a consensus for an optimal duration or components of a rehabilitation program. This might be explained by the multiple factors that could interfere with it, such as participants' motivation and financial and logistic resources. 

In our second comparison, we were also exposed to a diversity of training protocols in the studies that compared IMT with control/sham. More than half of the included RCTs used a sham IMT, and only two (Bavarsad 2015; Dacha 2019), had some participants with mild COPD. One trial (Saher 2021), focused on patients receiving non‐invasive ventilation as part of COPD management. Almost all the trials excluded participants who had exacerbations before starting the trial, and two RCTs worked on patients with limited functional performance (Covey 2001; ZhouL 2016). Two studies had a long intervention duration (Beckerman 2005; Kim 1993), equal to one year and six months respectively; two studies trained their participants with a load up to 100% of their PImax (Chuang 2017; Hill 2006); two trials conducted endurance IMT (Koppers 2006; Scherer 2000), and one trial performed both endurance and strength exercises (Petrovic 2012). The remaining studies had close characteristics (i.e. strength training, similar loads).

This review did not compare threshold IMT with normocapnic hyperpnea training (endurance IMT) nor compare training loads. We thought it would be best to work on these objectives in a separate review.

Overall, this review included all kinds of participants and interventions without preferences. However, our results might not be applicable for hospitalised patients following an exacerbation or patients with severe COPD who require long‐term oxygen therapy. 

Quality of the evidence

Using the GRADE approach, we assessed the certainty of the evidence of five outcomes: dyspnea (Borg, mMRC, BDI‐TDI), functional exercise capacity (6MWD), HRQoL (SGRQ, CAT), and PImax. When the number of RCTs exceeded 10, we created a funnel plot to investigate publication bias. 

For both comparisons, we considered the certainty of evidence to be from very low to moderate across all the outcomes (see Table 1; Table 2). 

Comparison 1: pulmonary rehabilitation plus inspiratory muscle training versus pulmonary rehabilitation

In our first comparison (PR+IMT versus PR), we downgraded the certainty of evidence by one level for the Borg scale measured at submaximal exercise capacity because of serious concerns about risk of bias related to blinding participants.

We downgraded the certainty of evidence of the mMRC to very low due to very serious concerns with risk of bias (all the trials are at high risk of bias) and serious concerns with imprecision (the sample size is less than 400, rule of thumb). 

We considered the certainty of evidence of the 6MWD to be very low due to serious concerns with risk of bias (studies at low risk of bias had different effect estimate from the overall effect estimate) and very serious concerns with inconsistency, although we explained part of the heterogeneity by the difference in training durations. Indeed, there was substantial statistical heterogeneity, and the effect estimates were wide on both sides of the line of no effect, making the benefit of the intervention doubtful. The funnel plot was symmetrical (Figure 2), so publication bias was improbable.

2.

Funnel plot of comparison 7, PR+IMT vs PR, outcome: 7.3 functional exercise capacity: 6‐minute walk distance (6MWD)

We judged the quality of evidence of the SGRQ (total score) as low because of very serious concerns with risk of bias. In fact, none of the studies was at low risk of bias; the main issue was the lack of blinding of participants (outcome measurement).

We considered the certainty of evidence of CAT to be very low due to very serious concerns with risk of bias (most of the evidence is from studies at high risk of bias and some concerns, lack of blinding) and serious concerns with inconsistency (considerable statistical heterogeneity, and confidence intervals do not overlap).

We downgraded the certainty of evidence of PImax by one level due to serious concerns with risk of bias (most of the evidence is from studies at high risk of bias and some concerns). We did not consider statistical heterogeneity because all the effect estimates were on one side showing a benefit). The funnel plot did not raise suspicions of publication bias (Figure 3). As a result, we considered the certainty of the evidence as moderate.

3.

Funnel plot of comparison 7, PR+IMT vs PR, outcome: 7.12 respiratory muscle strength: PImax

Comparison 2: inspiratory muscle training versus control/sham

In our second comparison (IMT versus control/sham), we downgraded the certainty of evidence of Borg measured at submaximal exercise capacity to very low due to very serious concerns with risk of bias (most of the studies were at high risk or some concerns, with different effect estimates compared to low risk of bias studies, issues with blinding) and serious concerns with imprecision (small sample size less than 400 participants, rule of thumb).

For the same reasons explained above, we downgraded the certainty of evidence for dyspnea assessed with BDI‐TDI by three levels. 

We considered the certainty of evidence of mMRC to be low due to serious concerns with risk of bias (most of the evidence is from studies at high risk of bias and some concerns) and imprecision (sample size less than 400 participants, rule of thumb).

We judged the quality of evidence of the 6MWD as moderate due to serious concerns with risk of bias (most of the evidence is from studies at high risk of bias and with some concerns). We did not downgrade inconsistency, despite the substantial statistical heterogeneity, because the effect estimates were on one side of the line of no effect. So we were more confident about the direction of the effect. The funnel plot raised some concerns about publication bias (Figure 4), without downgrading the evidence.

4.

Funnel plot of comparison 8, IMT vs control/sham, outcome: 8.8 functional exercise capacity: 6‐minute walk distance (6MWD)

We downgraded the certainty of evidence of the SGRQ  by three levels because of very serious concerns with risk of bias (studies with greater weight were at high risk of bias and some concerns, showing different effect estimates from studies at low risk of bias) and very serious concerns with imprecision (sample size less than 400 participants, rule of thumb; the 95% CI includes benefit and harm).

We considered the certainty of evidence of CAT to be moderate due to serious concerns with imprecision (sample size less than 400, rule of thumb).

We judged the certainty of evidence of PImax as low due to serious concerns with risk of bias (most of the evidence is from studies at high risk of bias and with some concerns) and a strongly suspected publication bias (Figure 5).

5.

Potential biases in the review process

The review is based on a published protocol (Ammous 2020). We discuss deviations from the protocol in the Differences between protocol and review section.

One of the main issues with the included studies in this review is the risk of bias, using RoB 2, which is outcome‐dependent. Only six out of 22 studies were at low risk of bias for the first comparison, while only three out of 37 studies were at low risk for the second comparison. Across all the outcomes, the major problems were lack of detail about allocation concealment, some issues with missing data, participants being aware of their intervention (for the participant‐reported outcome that involves judgments), and not publishing a protocol or listing the study in a trial register before launching the studies. Sometimes studies at high risk of bias tended to show a larger effect than the others, which reduced the confidence in our results. 

One trial provided IMT for a short period of time (two weeks; Saher 2021), and participants received non‐invasive ventilation throughout the trial.

From all our included studies, only one abstract (Masanga 2011), reported adverse events. Although it is unlikely that IMT may be associated with harm, we have some concerns due to the variety of side effects reported by Masanga 2011, and because the included studies did not discuss them. That is to say, one of the reasons for discontinuing the trial was the inability of participants to continue the intervention. But the studies did not discuss the reasons that made patients take that decision. 

Moreover, when dividing the studies according to mean baseline PImax to classify participants with or without respiratory muscle weakness, we assumed that the proportion of participants not belonging to their assigned subgroup was balanced across the studies. In other words, getting individual data and conducting a subgroup analysis within study data was impossible. So, we assumed this variation was balanced between the studies, considering that PImax follows a normal distribution with fewer outliers. 

Agreements and disagreements with other studies or reviews

Ten published reviews that explored the effect of IMT on COPD have been published over the years. The first one was published in the 90s (Smith 1992). This review included 17 trials in which the participants had chronic airflow limitation without specifying the types of the diseases. They summarised the effect estimate with SMD meta‐analysis and interpreted it according to Cohen’s d effect size. They found a small to moderate effect favouring IMT for MVV, PImax and laboratory exercise tests. However, this review included trials that run resistance training without controlling the breathing pattern.

Ten years later, another systematic review was published (Lötters 2002), including 15 RCTs. Unlike our review, this study pooled altogether in an SMD meta‐analysis RCTs that looked at the effect of IMT as a stand‐alone intervention and when associated with PR. Then, they conducted a subgroup analysis to investigate the additional effect of PR. They found that IMT alone might improve dyspnea and respiratory muscle strength and endurance. When combined with PR, a beneficial effect was seen only in participants with respiratory muscle weakness.

The following review (16 studies) was conducted by Crowe 2005, followed by an update with 18 studies (O'Brien 2008), in which they compared IMT separately with different types of interventions (exercises, breathing techniques, education, PR). The authors concluded, based on a study‐level analysis, that IMT might improve respiratory muscle strength. However, there was less confidence with dyspnea and HRQoL. In our review, we considered the participants who underwent therapeutic patient education or one type of breathing exercises as control, because the aim was not to compare IMT with another intervention but to create a psychological effect for patients not receiving IMT. We excluded the studies that focused on comparing IMT with breathing exercises. Similarly, the same authors worked on another systematic review (Geddes 2005), and then they updated it (Geddes 2008). These reviews compared IMT with no intervention, low versus high IMT, and two different modes of IMT, which is not the purpose of our review.

A narrative review (Shoemakher 2009), reported that IMT, in comparison with sham IMT or no intervention, might improve dyspnea, HRQoL and PImax. Following that, Gosselink 2011 showed similar results to Lötters 2002. The authors included 32 RCTs, and found a possible effect of IMT in improving dyspnea, functional exercise capacity, HRQoL, PImax and respiratory muscle endurance. In subgroup analysis, they reported a benefit of IMT when associated with PR only in participants with respiratory muscle weakness. They also found a larger effect of strength training compared to endurance training.

In a subsequent study, Beaumont 2018a included 37 trials in a meta‐analysis showing an improvement in dyspnea, HRQoL, functional exercise capacity and PImax after threshold IMT. However, they did not find a difference when combining IMT with PR. The last meta‐analysis (Figueiredo 2020), compared IMT alone or associated with other interventions with control, sham, or other interventions. The authors did not find an improvement in dyspnea, HRQoL, or a larger effect in participants with respiratory muscle weakness. They showed a trend of a larger effect with higher training loads.

Our systematic review is the first to have excluded trials that used resistive trainers without controlling the breathing pattern. We are also the first that have excluded the trials that did not measure Borg at isotime. Generally, our results were consistent with past systematic reviews regarding no effect when adding IMT to PR and a possible benefit with IMT as a stand‐alone intervention. However, we could not conclude that there is a better effect in participants with respiratory muscle weakness. A possible explanation for this finding includes the fact that most studies worked on participants without respiratory muscle weakness. 

Authors' conclusions

Implications for practice.

When associated with pulmonary rehabilitation (PR), inspiratory muscle training (IMT) may not have an additional benefit on dyspnea, functional exercise capacity and health‐related quality of life. There was an increase in inspiratory muscle strength and endurance, but this difference was not clinically meaningful. 

IMT may improve dyspnea, functional exercise capacity (mainly the 6MWD) and health‐related quality of life, compared to sham or no intervention. There was also an increase in inspiratory muscle strength and endurance, but judging the clinical significance of these outcomes was challenging due to a non‐standardised minimal clinically important difference.

For both interventions, we could not conclude that there was a possible larger effect in participants with respiratory muscle weakness and with longer durations of training. 

Implications for research.

Overall, we did not detect any potential advantage of combining PR and IMT, which was consistent with past systematic reviews. However, it is still unclear if this intervention is more beneficial in participants with respiratory muscle weakness, and future research may focus more on this group of participants.

We think there is enough evidence for the effect of IMT alone since our results are consistent with past systematic reviews. However, all published trials had a small sample size. So we believe future trials should increase the number of participants, investigate the possible larger effect on participants with respiratory muscle weakness and compare different IMT protocols.

IMT may be a starter intervention for patients unable to undergo PR (e.g. severe COPD, logistical or financial issues). Still, this should be explored further in clinical trials. 

For both comparisons, we highly suggest providing a sham IMT for the control group. Although a sham IMT might not influence the results of an observer‐reported outcome such as the 6‐minute walk distance, we strongly believe that participants who know they are undergoing an intervention may overestimate the effect measured with participant‐reported outcomes that involve judgments. Furthermore, future research may consider training with higher inspiratory flow rate and training at high lung volume (closer to total lung capacity) to improve performance in hyperinflated COPD patients.

What's new

Date	Event	Description
9 January 2023	Amended	Search date corrected in Abstract and PLS to be 20 October 2022

History

Protocol first published: Issue 11, 2020
Review first published: Issue 1, 2023

Risk of bias

Risk of bias for analysis 1.1 Dyspnea: Borg (at submaximal exercise: 50% to 80% of Wmax).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missingness did not affect the outcome.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the protocol and the journal article.

Low risk of bias

No detected issue with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

ITT  was not conducted with a substantial impact on the results.

High risk of bias

Missingness in the outcome is likely to depend on its true value.

High risk of bias

Participants were not blinded and the outcome is a participant‐reported outcome that involves judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. ITT analysis was not implemented, missingness is likely to be dependant on its value and participants were not blinded.

Risk of bias for analysis 1.2 Dyspnea: Modified Medical Research Council (mMRC).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data from two participants out of 149 were missing.

High risk of bias

Participants were not blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

High risk of bias

Participant were not blinded (no sham) and the outcome is a participant‐reported outcome.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only 7% of the data were missing, we considered LOCF imputation a valid method for this kind of intervention.

High risk of bias

Participants were not blinded.

Some concerns

Only the journal article is available.

High risk of bias

Participants were not blinded and only the journal article is available.

Risk of bias for analysis 1.3 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their allocation.

Low risk of bias

No loss of follow up or exclusion.

Low risk of bias

Observer‐reported outcome that do not involve judgements.

Low risk of bias

A trial register is available, and both adjustet and unadjusted analysis were reported.

Some concerns

Lack of information about the randomisation process.

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

"All analyses were performed on an intention‐to‐treat basis".

Low risk of bias

The data from two participants out of 149 were missing.

Low risk of bias

"All the data were collected by a research nurse blinded to treatment allocation".

Low risk of bias

A trial register is available and there is only one way to measure the outcome.

Low risk of bias

No detected issues with the five domains.

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missingness did not bias the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial protocol is available with no differences with the journal article.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed.

Low risk of bias

The data of 2 participants out of 33 were missing (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed.

Low risk of bias

The data of 2 participants out of 33 were missing (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

The reasons for exclusion were not reported, and it is unlikely that a per‐protocol analysis was conducted.

Some concerns

The data of 9 participants out of 38 were missing, and the proportion of missingness was balanced between the two groups.

Low risk of bias

Observer‐reported outcome that do not involve judgements

Some concerns

Only the journal is available, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up were not explained, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

the data of 2 participants out of 31 were missing (6%).

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available, and there is only one way to measure the outcome.

Some concerns

Lack of details about allocation concealement and only the trial register is available.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

11% of the participants discontinued the intervention. The reasons were: intolerance to the intervention and low compliance.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, with no differences with the journal article. No adusted analysis were conducted.

High risk of bias

Missingness is likely to be dependant on its true value.

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balanced between the groups.

Low risk of bias

The outcome was adequately measured, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the abstracts, the methods and the results sections.

Some concerns

Lack of details about allocation concealment, and no issues with the other domains.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

No deviations were reported and all participants were analysed according to their initial allocation.

Low risk of bias

All the data are available.

Low risk of bias

Outcome measurement was appropriate and it is an observer reported outcome that do not involve judgements

Low risk of bias

Only the journal article is available, and no differences between the methods and the results sections.

High risk of bias

Lack of details about the randomisation process and only the journal article is available.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account that only 7% of the data are missing, we considered LOCF imputation a valid method for this kind on intervention.

Low risk of bias

Outcome measurement was appropriate and outcome assessors were blinded.

Some concerns

Only the journal article is available, both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Risk of bias for analysis 1.4 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.4.1 Short‐term (<4 weeks)

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their allocation.

Low risk of bias

No loss of follow up or exclusion.

Low risk of bias

Observer‐reported outcome that do not involve judgements.

Low risk of bias

A trial register is available, and both adjustet and unadjusted analysis were reported.

Some concerns

Lack of information about the randomisation process.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

11% of the participants discontinued the intervention. The reasons were: intolerance to the intervention and low compliance.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, with no differences with the journal article. No adusted analysis were conducted.

High risk of bias

Missingness is likely to be dependant on its true value.

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balanced between the groups.

Low risk of bias

The outcome was adequately measured, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the abstracts, the methods and the results sections.

Some concerns

Lack of details about allocation concealment, and no issues with the other domains.

Subgroup 1.4.2 Medium‐term ( ≥4 weeks and <8 weeks)

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

"All analyses were performed on an intention‐to‐treat basis".

Low risk of bias

The data from two participants out of 149 were missing.

Low risk of bias

"All the data were collected by a research nurse blinded to treatment allocation".

Low risk of bias

A trial register is available and there is only one way to measure the outcome.

Low risk of bias

No detected issues with the five domains.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Subgroup 1.4.3 Long‐term ( ≥8 weeks)

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missingness did not bias the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial protocol is available with no differences with the journal article.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed.

Low risk of bias

2 participants out of 33 were not analysed (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed.

Low risk of bias

2 participants out of 33 were not analysed (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

The reasons for exclusion were not reported, and it is unlikely that a per‐protocol analysis was conducted.

Some concerns

The data of 9 participants out of 38 were missing, and the proportion of missingness was balanced between the two groups.

Low risk of bias

Observer‐reported outcome that do not involve judgements

Some concerns

Only the journal is available, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up were not explained, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

the data of 2 participants out of 31 were missing (6%).

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available, and there is only one way to measure the outcome.

Some concerns

Lack of details about allocation concealement and only the trial register is available.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

No deviations were reported and all participants were analysed according to their initial allocation.

Low risk of bias

All the data are available.

Low risk of bias

Outcome measurement was appropriate and it is an observer reported outcome that do not involve judgements

Low risk of bias

Only the journal article is available, and no differences between the methods and the results sections.

High risk of bias

Lack of details about the randomisation process and only the journal article is available.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account that only 7% of the data are missing, we considered LOCF imputation a valid method for this kind on intervention.

Low risk of bias

Outcome measurement was appropriate and outcome assessors were blinded.

Some concerns

Only the journal article is available, both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Risk of bias for analysis 1.5 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.5.1 With respiratory muscle weakness

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missingness did not bias the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial protocol is available with no differences with the journal article.

Low risk of bias

No detected issues with the five domains.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

No deviations were reported and all participants were analysed according to their initial allocation.

Low risk of bias

All the data are available.

Low risk of bias

Outcome measurement was appropriate and it is an observer reported outcome that do not involve judgements

Low risk of bias

Only the journal article is available, and no differences between the methods and the results sections.

High risk of bias

Lack of details about the randomisation process and only the journal article is available.

Subgroup 1.5.2 Without respiratory muscle weakness

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their allocation.

Low risk of bias

No loss of follow up or exclusion.

Low risk of bias

Observer‐reported outcome that do not involve judgements.

Low risk of bias

A trial register is available, and both adjustet and unadjusted analysis were reported.

Some concerns

Lack of information about the randomisation process.

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

"All analyses were performed on an intention‐to‐treat basis".

Low risk of bias

The data from two participants out of 149 were missing.

Low risk of bias

"All the data were collected by a research nurse blinded to treatment allocation".

Low risk of bias

A trial register is available and there is only one way to measure the outcome.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protocol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protocol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

The reasons for exclusion were not reported, and it is unlikely that a per‐protocol analysis was conducted.

Some concerns

The data of 9 participants out of 38 were missing, and the proportion of missingness was balanced between the two groups.

Low risk of bias

Observer‐reported outcome that do not involve judgements

Some concerns

Only the journal is available, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up were not explained, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

the data of 2 participants out of 31 were missing (6%).

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available, and there is only one way to measure the outcome.

Some concerns

Lack of details about allocation concealement and only the trial register is available.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

11% of the participants discontinued the intervention. The reasons were: intolerance to the intervention and low compliance.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, with no differences with the journal article. No adusted analysis were conducted.

High risk of bias

Missingness is likely to be dependant on its true value.

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balanced between the groups.

Low risk of bias

The outcome was adequately measured, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the abstracts, the methods and the results sections.

Some concerns

Lack of details about allocation concealment, and no issues with the other domains.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was independent on its true value.

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account that only 7% of the data are missing, we considered LOCF imputation a valid method for this kind on intervention.

Low risk of bias

Outcome measurement was appropriate and outcome assessors were blinded.

Some concerns

Only the journal article is available, both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Risk of bias for analysis 1.7 Functional exercise capacity: Wmax (watt).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that the results was not biased by missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the protocol, and the journal article.

Low risk of bias

No detected issues with the five domains.

Dekhuijzen 1991

Some concerns

Only a statements about randomisation was reported, with no information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

The data of 4 week were not reported.

High risk of bias

Lack of details about the randomisation process, allocation concealment, and only the journal article is available. Data of the 4th week were not reported.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

59% of the participants were excluded.

High risk of bias

Missingness is likely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. ITT analysis was not conducted and missingness is likely to be dependant on the outcome.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

Some concern with ITT analysis.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

Low risk of bias

Observer‐reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, and only the journal article is available.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Considering only 6% of the data was missing, we consider LOCF imputation valid for this kind of intervention.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Only the journal article is available.

Risk of bias for analysis 1.9 Health‐related quality of life (HRQoL): St George's Respiratory Questionnaire (SGRQ).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.9.1 Symptoms

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

High risk of bias

Participant‐reported outcome and no sham was used.

Low risk of bias

No differences between the trial register, the conference abstracts, data were reported correctly according to their distribution with no differences with the methods section.

High risk of bias

HRQoF is a participants‐reported outcome and there are no information about blinding or the use of a sham training.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participant reported outcome that involves judgements, and participants were not blinded.

Some concerns

Only the trial register is available.

High risk of bias

Allocation was probably not concealed, participants were not blinded and only the journal article is available.

Subgroup 1.9.2 Activity

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

High risk of bias

Participant‐reported outcome and no sham was used.

Low risk of bias

No differences between the trial register, the conference abstracts, data were reported correctly according to their distribution with no differences with the methods section.

High risk of bias

HRQoF is a participants‐reported outcome and there are no information about blinding or the use of a sham training.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participant reported outcome that involves judgements, and participants were not blinded.

Some concerns

Only the trial register is available.

High risk of bias

Allocation was probably not concealed, participants were not blinded and only the journal article is available.

Subgroup 1.9.3 Impact

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

High risk of bias

Participant‐reported outcome and no sham was used.

Low risk of bias

No differences between the trial register, the conference abstracts, data were reported correctly according to their distribution with no differences with the methods section.

High risk of bias

HRQoF is a participants‐reported outcome and there are no information about blinding or the use of a sham training.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participant reported outcome that involves judgements, and participants were not blinded.

Some concerns

Only the trial register is available.

High risk of bias

Allocation was probably not concealed, participants were not blinded and only the journal article is available.

Subgroup 1.9.4 Total

Abedi Yekta 2019

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to the initial allocation.

Low risk of bias

Missing data is balanced between the the four groups.

High risk of bias

HRQoF is a participant‐reported outcome that involves judgements.

Low risk of bias

A trial register is available, and no adjusted analysis were conducted.

High risk of bias

Lack of details about the randomisation process and participants were not blinded.

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

High risk of bias

Participant‐reported outcome and no sham was used.

Low risk of bias

No differences between the trial register, the conference abstracts, data were reported correctly according to their distribution with no differences with the methods section.

High risk of bias

HRQoF is a participants‐reported outcome and there are no information about blinding or the use of a sham training.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of 2 participants out of 31 were missing.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the trial register is available.

Majewska‐Pulsakowska 2016

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

The scale is validated, participants were not blinded, and it is a participant reported‐outcome that do not involve judgements.

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available and participants were not blinded.

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balanced between the two groups.

Low risk of bias

Participant‐reported outcome that involves judgements, and a sham was used.

Low risk of bias

No differences between the trial register, the methods and the results sections.

Some concerns

Lack of details about allocation concealment.

Tout 2013

High risk of bias

No details about randomisation in the trial (we only discovered that it was a RCT in a personal communication).

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participant reported outcome that involves judgements, and participants were not blinded.

Some concerns

Only the trial register is available.

High risk of bias

Allocation was probably not concealed, participants were not blinded and only the journal article is available.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account only 7% of the data were missing, we considered LOCF a valid imputation method for this kind of intervention.

High risk of bias

Participants were not blinded, and it is an observer reported outcome that involves judgements.

Some concerns

Only the journal article is available, and both adjusted and unadjusted analysis were reported.

High risk of bias

Participants were not blinded and only the journal article is available.

Risk of bias for analysis 1.10 Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.10.1 Dyspnea

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that the results were not biased by missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the protocol, and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

59% of the participants were excluded.

High risk of bias

Misssingness is likely to be dependant on its true value.

Some concerns

Issues with blinding.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and there are still issues with blinding. There is some concern with ITT, missingness is likely to be dependant on the outcome and only the journal article is available.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Sone concern with ITT analysis.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

High risk of bias

No sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, participants were not blinded, and only the journal article is available.

Subgroup 1.10.2 Fatigue

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that the results were not biased by missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the protocol, and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

59% of the participants were excluded.

High risk of bias

Misssingness is likely to be dependant on its true value.

Some concerns

Issues with blinding.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and there are still issues with blinding. There is some concern with ITT, missingness is likely to be dependant on the outcome and only the journal article is available.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Sone concern with ITT analysis.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

High risk of bias

No sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, participants were not blinded, and only the journal article is available.

Subgroup 1.10.3 Emotion

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that the results were not biased by missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the protocol, and the journal article.

Low risk of bias

No detected issues with the five domains.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Sone concern with ITT analysis.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

High risk of bias

No sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, participants were not blinded, and only the journal article is available.

Subgroup 1.10.4 Mastery

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that the results were not biased by missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the protocol, and the journal article.

Low risk of bias

No detected issues with the five domains.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Sone concern with ITT analysis.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

High risk of bias

No sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, participants were not blinded, and only the journal article is available.

Risk of bias for analysis 1.11 Health‐related quality of life (HRQoL): COPD Assessment Test (CAT).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT was balanced between the groups

Low risk of bias

Missingness was balanced between the groups

Low risk of bias

Participant‐reported outcome that involves judgements, and a sham was used

Low risk of bias

No differences between the trial register, the methods and the results sections.

Some concerns

Lack of details about allocation concealment.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted

Low risk of bias

Taking account only 7% of the data were missing, we considered LOCF a valid imputation method for this kind of intervention.

High risk of bias

Participants were not blinded, and it is an observer reported outcome that involves judgements.

Some concerns

Only the journal article is available, and both adjusted and unadjusted analysis were reported.

High risk of bias

Participants were not blinded and only the journal article is available.

Risk of bias for analysis 1.12 Inspiratory muscle strength: PImax (cmH20).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was appropriate and it is an observer‐reported outcome that do not involve judgement.

Low risk of bias

A trial register is available, and both adjusted and unadjusted analysis were reported.

Some concerns

No information about allocation concealment.

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conferences abstract and the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Berry 1996

Some concerns

Only a statement about randomisation was reported. Baseline data were not reported.

Low risk of bias

No deviations were reported, and only one participant was excluded because of exacerbation (non related to the intervention).

Low risk of bias

One participant was missing out of 16 (6%)

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article and adjusted endpoint scores are available.

Some concerns

No information about allocation concealment, only the journal article is available and only endpoint scores were reported

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missing data did not affect the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A protocol is available.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protocol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protocol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Dekhuijzen 1991

Some concerns

Only a statements about randomisation was reported, with no information about allocation concealment.

Low risk of bias

"All participants were analysed according to their initial allocation"

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is appropriate and it is an observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available, and no mention of adjusted analysis.

Some concerns

Lack of details about the randomisation process, allocation concealment, and only the journal article is available.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and all‐time points were reported. No differences between the trial register, the methods and results sections.

Low risk of bias

No detected issues with the five domains.

Fanfa Bordin 2020

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness in the outcome did not depend on its true value.

Low risk of bias

The method of measurement is appropriate, and outcome assessors were blinded.

Low risk of bias

No differences between the trial register and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if some excluded participant could have been included.

High risk of bias

Missingness is likely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available and only significance was reported

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

No deviations were reported, and the reason for exclusion were not explained.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, no differences between the methods and the results section, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only the data of 2 participants out of 31 were missing (6%).

Low risk of bias

The method of measuring the outcome is appropriate, and it is an Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the trial register is available.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

Missingness in the outcome is likely to be dependant on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the methods and the results sections.

High risk of bias

Missingness is likely to be dependant on its true value

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balnced between the groups.

Low risk of bias

The method of measurement is adequate, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the methods and the results sections.

Low risk of bias

Lack of details about allocation concealement.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account only 7% of the data were missing, we consider LOCF imputation a valid technique for this kind of intervention.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available, and both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Weiner 1992

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is valid, and the outcome is observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Weiner 2000

High risk of bias

Clear imbalance in the group sizes.

Low risk of bias

No deviations or exclusions.

Low risk of bias

The drop out is balanced between the two groups.

Low risk of bias

The method of measured is valid, and it is observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, and no adjusted analysis were reported.

High risk of bias

Substantial differences between group sizes and only the journal article is available.

Risk of bias for analysis 1.13 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: duration of the intervention).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.13.1 Short‐term (<4 weeks)

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was appropriate and it is an observer‐reported outcome that do not involve judgement.

Low risk of bias

A trial register is available, and both adjusted and unadjusted analysis were reported.

Some concerns

No information about allocation concealment.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and all‐time points were reported. No differences between the trial register, the methods and results sections.

Low risk of bias

No detected issues with the five domains.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

Missingness in the outcome is likely to be dependant on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the methods and the results sections.

High risk of bias

Missingness is likely to be dependant on its true value

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balnced between the groups.

Low risk of bias

The method of measurement is adequate, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the methods and the results sections.

Low risk of bias

Lack of details about allocation concealement.

Subgroup 1.13.2 Medium‐term ( ≥ 4 weeks and <8 weeks)

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conferences abstract and the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

Dekhuijzen 1991

Some concerns

Only a statements about randomisation was reported, with no information about allocation concealment.

Low risk of bias

"All participants were analysed according to their initial allocation"

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is appropriate and it is an observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available, and no mention of adjusted analysis.

Some concerns

Lack of details about the randomisation process, allocation concealment, and only the journal article is available.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and all‐time points were reported. No differences between the trial register, the methods and results sections.

Low risk of bias

No detected issues with the five domains.

Weiner 2000

High risk of bias

Clear imbalance in the group sizes.

Low risk of bias

No deviations or exclusions.

Low risk of bias

The drop out is balanced between the two groups.

Low risk of bias

The method of measured is valid, and it is observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, and no adjusted analysis were reported.

High risk of bias

Substantial differences between group sizes and only the journal article is available.

Subgroup 1.13.3 Long‐term ( ≥ 8 weeks)

Berry 1996

Some concerns

Only a statement about randomisation was reported. Baseline data were not reported.

Low risk of bias

No deviations were reported, and only one participant was excluded because of exacerbation (non related to the intervention).

Low risk of bias

One participant was missing out of 16 (6%)

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article and adjusted endpoint scores are available.

Some concerns

No information about allocation concealment, only the journal article is available and only endpoint scores were reported.

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missing data did not affect the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A protocol is available.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protcol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Low risk of bias

The data of 2 participants out of 33 are missing (6%)

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the abstract and the trial protcol are available

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Dekhuijzen 1991

Some concerns

Only a statements about randomisation was reported, with no information about allocation concealment.

Low risk of bias

"All participants were analysed according to their initial allocation"

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is appropriate and it is an observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available, and no mention of adjusted analysis.

Some concerns

Lack of details about the randomisation process, allocation concealment, and only the journal article is available.

Fanfa Bordin 2020

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness in the outcome did not depend on its true value.

Low risk of bias

The method of measurement is appropriate, and outcome assessors were blinded.

Low risk of bias

No differences between the trial register and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if some excluded participant could have been included.

High risk of bias

Missingness is likely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available and only significance was reported

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

No deviations were reported, and the reason for exclusion were not explained.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, no differences between the methods and the results section, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only the data of 2 participants out of 31 were missing (6%).

Low risk of bias

The method of measuring the outcome is appropriate, and it is an Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the trial register is available.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account only 7% of the data were missing, we consider LOCF imputation a valid technique for this kind of intervention.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available, and both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Weiner 1992

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is valid, and the outcome is observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Risk of bias for analysis 1.14 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: with or without respiratory muscle weakness).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 1.14.1 With respiratory muscle weakness

Charususin 2018

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

There is evidence that missing data did not affect the results.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A protocol is available.

Low risk of bias

No detected issues with the five domains.

Dekhuijzen 1991

Some concerns

Only a statements about randomisation was reported, with no information about allocation concealment.

Low risk of bias

"All participants were analysed according to their initial allocation"

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is appropriate and it is an observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available, and no mention of adjusted analysis.

Some concerns

Lack of details about the randomisation process, allocation concealment, and only the journal article is available.

Dellweg 2017

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

A trial register is available, and all‐time points were reported. No differences between the trial register, the methods and results sections.

Low risk of bias

No detected issues with the five domains.

Weiner 1992

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement is valid, and the outcome is observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Weiner 2000

High risk of bias

Clear imbalance in the group sizes.

Low risk of bias

No deviations or exclusions.

Low risk of bias

The drop out is balanced between the two groups.

Low risk of bias

The method of measured is valid, and it is observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, and no adjusted analysis were reported.

High risk of bias

Substantial differences between group sizes and only the journal article is available.

Subgroup 1.14.2 Without respiratory muscle weakness

Beaumont 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was appropriate and it is an observer‐reported outcome that do not involve judgement.

Low risk of bias

A trial register is available, and both adjusted and unadjusted analysis were reported.

Some concerns

No information about allocation concealment.

Beaumont 2018

Low risk of bias

Allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Only the data of two participants out of 149 were missing.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conferences abstract and the methods and the results sections.

Low risk of bias

No detected issues with the five domains.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Some concerns

The data of 2 participants out of 33 are missing

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

De Farias 2019

Some concerns

The allocation sequence was random, lack of details about allocation concealment, and no baseline differences.

Some concerns

2 participants out of 33 were not analysed

Some concerns

The data of 2 participants out of 33 are missing

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial protocol are available.

Some concerns

Lack of details about the randomisation process, reason for missingness, and only the abstract et the protocol are available.

Fanfa Bordin 2020

Low risk of bias

The sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness in the outcome did not depend on its true value.

Low risk of bias

The method of measurement is appropriate, and outcome assessors were blinded.

Low risk of bias

No differences between the trial register and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if some excluded participant could have been included.

High risk of bias

Missingness is likely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available and only significance was reported

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Mador 2005

Some concerns

Randomisation was reported only as a sentence.

Some concerns

No deviations were reported, and the reason for exclusion were not explained.

Some concerns

The reasons of missingness were not reported, but its proportion was balanced between the groups.

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Only the journal article is available, no differences between the methods and the results section, and no adjusted analysis were conducted.

Some concerns

Lack of details about the randomisation process, reasons for exclusion and/or loss of follow up, and only the journal article is available.

Magadle 2007

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only the data of 2 participants out of 31 were missing (6%).

Low risk of bias

The method of measuring the outcome is appropriate, and it is an Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the trial register is available.

Paneroni 2018

Low risk of bias

Allocation sequence was concealed.

Low risk of bias

No participants were excluded.

High risk of bias

Missingness in the outcome is likely to be dependant on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the methods and the results sections.

High risk of bias

Missingness is likely to be dependant on its true value

Schultz 2018

Some concerns

No information about allocation concealment.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness was balnced between the groups.

Low risk of bias

The method of measurement is adequate, and it is an observer reported outcome that do not involve judgements.

Low risk of bias

No differences between the trial register, the methods and the results sections.

Low risk of bias

Lack of details about allocation concealement.

Tounsi 2021

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

The outcome was measured according to the guidelines and it is an observer‐reported outcome that does not involve judgement.

Low risk of bias

No differences between the journal article and the trial register.

Low risk of bias

No detected issues with the five domains.

Wang 2017

Low risk of bias

The allocation sequence was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Taking account only 7% of the data were missing, we consider LOCF imputation a valid technique for this kind of intervention.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available, and both adjusted and unadjusted analysis were reported.

Some concerns

Only the journal article is available.

Risk of bias for analysis 2.1 Dyspnea: Borg (at submaximal exercise capacity).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Dacha 2019

Low risk of bias

From personal communication with the trialist, the suequence was random, concealed, and the groups had similar baseline characteristics.

Low risk of bias

Form personal communication with the trialist, ITT was conducted.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded, and it is a participant‐reported outcome that involves judgement.

Some concerns

Only the abstract and the trial register are available.

Some concerns

Only the abstract and the trial register are available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

No deviations were detected, and all participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) were missing.

Low risk of bias

The scale is validated and participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39)

Low risk of bias

The scale is validated, and a sham was used.

Some concerns

No differences between the methods and the results sections, and only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

A trial register is available, with no differences with the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

77 participants were disqualified from participation (out of 130), and it is unclear if the reasons might have an impact on the outcome.

High risk of bias

Around half of the data were missing, and missingness is likely to be dependant on its true value.

High risk of bias

Participants were not blinded, and it is a participant‐reported outcome that involves judgements.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only journal article is available. There is some concern with ITT analysis, missingness is likely to be dependant on its value and participants were not blinded.

Petrovic 2012

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

High risk of bias

No mention of blinding, and it is a participant‐reported outcome that involves judgements.

Some concerns

The trial register did not provide precise information.

High risk of bias

Lack of details about the randomisation process, participants were not blinded, and the trial register does not provide supplementary information.

Risk of bias for analysis 2.2 Dyspnea: Baseline and Transition Dyspnea Indexes (BDI‐TDI).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.2.1 Functional impairment

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

missingness in each group was not mentioned.

Low risk of bias

A sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Subgroup 2.2.2 Magnitude of task

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

missingness in each group was not mentioned.

Low risk of bias

A sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Subgroup 2.2.3 Magnitude of effort

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

missingness in each group was not mentioned.

Low risk of bias

A sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Subgroup 2.2.4 Focal score

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded based on the intervention received.

High risk of bias

Missingness is likely to be dependant on its true value.

High risk of bias

Participants were not blinded.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, participants were not blinded and only the journal article is available.

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

missingness in each group was not mentioned.

Low risk of bias

A sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issues with the five domains.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

Adjusted analysis were conducted only for the training group.

High risk of bias

Lack of details about the randomisation process and adjusted analysis were conducted only for the training group.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Low risk of bias

missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Risk of bias for analysis 2.3 Dyspnea: Transition Dyspnea Index (TDI): Focal score (subgroup analysis: with or without respiratory muscle weakness).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.3.1 With respiratory muscle weakness

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded based on the intervention received.

High risk of bias

Missingness is likely to be dependant on its true value.

High risk of bias

Participants were not blinded.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, participants were not blinded and only the journal article is available.

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

missingness in each group was not mentioned.

Low risk of bias

A sham was used.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

High risk of bias

Adjusted analysis were conducted only for the training group.

Some concerns

Lack of details about the randomisation process and adjusted analysis were conducted only for the training group.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

serious concern with blinding

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

serious concern with blinding

Some concerns

There are some differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article.

Subgroup 2.3.2 Without respiratory weakness

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issues with the five domains.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Low risk of bias

missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Risk of bias for analysis 2.4 Dyspnea: Modified Medical Research Council (mMRC).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

ITT analysis was conducted

Low risk of bias

No missing data

Low risk of bias

Participants were blinded

Low risk of bias

No differences between the trial register, the abstract, and the article

Low risk of bias

No issues with the five domains

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is unlikely to be dependent on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted

Low risk of bias

Missingness is independent on ites true value

Low risk of bias

A sham IMT was used

Low risk of bias

No differences between the abstracts, the trial register and the article

Low risk of bias

No issues with the five domains

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

Three participants were excluded (6%)

Some concerns

No details of the reasons of missingness, but its proportion is small (6%)

High risk of bias

Participants were not blinded, and it is a participant reported outcome that involve judgements.

Some concerns

Only the journal article is available

High risk of bias

Some issues with ITT analysis, missingness could be dependant on its true value, participants were not blinded and only the journal article is available.

Risk of bias for analysis 2.5 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Bavarsad 2015

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Only two participants out of 40 were excluded (no potential impact on the results).

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer‐reported outcome that do not involve judgement, and the method of measurement was appropriate.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation technique, allocation concealment, and only the journal article is available.

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reasons for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment and only the journal article is available. 26% of the participants dropped out of the study with no reason reported for 11%.

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal articile is available.

Cutrim 2019

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was valid, and it is observer reported outcome that do not involve judgements.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issue with the five domains.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. ITT was not conducted and missingness could be dependant on its true value.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. ITT was not conducted and missingness could be dependant on its true value.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

The method of measurement was valid and it is an observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

Two participants were excluded (12%).

Low risk of bias

Missingness did not depend on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted.

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Oberserver reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be depandant on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the protocol, trial register, and the journal article.

Low risk of bias

No detected issues with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded.

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

No ITT analysis was conducted, missingness could be dependant on its true value, and only the journal article is available.

Risk of bias for analysis 2.6 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.6.1 Short‐term (<4 weeks)

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Subgroup 2.6.2 Medium‐term (≥4 weeks and <8 weeks)

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal articile is available.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

The method of measurement was valid and it is an observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Subgroup 2.6.3 Long‐term (≥8 weeks)

Bavarsad 2015

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Only two participants out of 40 were excluded (no potential impact on the results).

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer‐reported outcome that do not involve judgement, and the method of measurement was appropriate.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation technique, allocation concealment, and only the journal article is available.

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reasons for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

No information about allocation concealment and only the journal article is available. 26% of the participants dropped out of the study with no reason reported for 11%.

Cutrim 2019

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was valid, and it is observer reported outcome that do not involve judgements.

Low risk of bias

A trial register is available, and no differences between the methods and the results sections.

Low risk of bias

No detected issue with the five domains.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. ITT was not conducted and missingness could be dependant on its true value.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. ITT was not conducted and missingness could be dependant on its true value.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

Two participants were excluded (12%).

Low risk of bias

Missingness did not depend on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

Observer reported outcome that do not invovle judgement

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be depandant on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the protocol, trial register, and the journal article.

Low risk of bias

No detected issues with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded.

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

No ITT analysis was conducted, missingness could be dependant on its true value, and only the journal article is available.

Risk of bias for analysis 2.7 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.7.1 With respiratory muscle weakness

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal articile is available.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the protocol, trial register, and the journal article.

Low risk of bias

No detected issues with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded.

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

No ITT analysis was conducted, missingness could be dependant on its true value, and only the journal article is available.

Subgroup 2.7.2 Without respiratory muscle weakness

Bavarsad 2015

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Only two participants out of 40 were excluded (no potential impact on the results).

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer‐reported outcome that do not involve judgement, and the method of measurement was appropriate.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation technique, allocation concealment, and only the journal article is available.

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reasons for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

No information about allocation concealment and only the journal article is available. 26% of the participants dropped out of the study with no reason reported for 11%.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. ITT was not conducted and missingness could be dependant on its true value.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

The method of measurement was valid and it is an observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

Two participants were excluded (12%).

Low risk of bias

Missingness did not depend on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

Observer reported outcome that do not invovle judgement

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be depandant on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the protocol, trial register, and the journal article.

Low risk of bias

No detected issues with the five domains.

Risk of bias for analysis 2.9 Functional exercise capacity: Wmax (watt).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) were missing.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

Only the journal article is avaible.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

Observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

59% of the participants were excluded.

High risk of bias

Missingness is likely to be dependent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

serious concern with ITT analysis.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conducted appropriately.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Some differences between the trial register and the journal article in blinding.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involves judgements.

Some concerns

Some differences between the trial register and the journal article in blinding.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Risk of bias for analysis 2.12 Health‐related quality of life (HRQoL): St George Respiratory Questionnaire (SGRQ) .

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.12.1 Symptoms

Berton 2015

Some concerns

No information about allocation concealment.

Some concerns

One excluded participant (4%) would unlikely affect the results.

Some concerns

The proportion of missingness was balanced between the intervention and the control group

Low risk of bias

The scale is validated and a sham was used.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment, one participant was excluded because of non adherence, missingness maybe dependant on its true value, and no details if adjusted analysis were planned in advance.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independant on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Subgroup 2.12.2 Activity

Berton 2015

Some concerns

No information about allocation concealment.

Some concerns

One excluded participant (4%) would unlikely affect the results.

Some concerns

The proportion of missingness was balanced between the intervention and the control group

Low risk of bias

The scale is validated and a sham was used.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment, one participant was excluded because of non adherence, missingness maybe dependant on its true value, and no details if adjusted analysis were planned in advance.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independant on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Subgroup 2.12.3 Impact

Berton 2015

Some concerns

No information about allocation concealment.

Some concerns

One excluded participant (4%) would unlikely affect the results.

Some concerns

The proportion of missingness was balanced between the intervention and the control group

Low risk of bias

The scale is validated and a sham was used.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment, one participant was excluded because of non adherence, missingness maybe dependant on its true value, and no details if adjusted analysis were planned in advance.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independant on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Subgroup 2.12.4 Total

Abedi Yekta 2019

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to the initial allocation.

Some concerns

Missing data is balanced between the the four groups.

High risk of bias

Participants were not blinded, and it is a participant‐reported outcome that involve judgements.

Low risk of bias

No differences between the trial register and the journal article.

High risk of bias

Lack of details about the randomisation process, missingness could be dependant on its true value, and participants were not blinded.

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reason for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment and only the journal article is available. 26% of the participants dropped out of the study with no reason reported for 11%.

Berton 2015

Some concerns

No information about allocation concealment.

Some concerns

One excluded participant (4%) would unlikely affect the results.

Some concerns

The proportion of missingness was balanced between the intervention and the control group

Low risk of bias

The scale is validated and a sham was used.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment, one participant was excluded because of non adherence, missingness maybe dependant on its true value, and no details if adjusted analysis were planned in advance.

Majewska‐Pulsakowska 2016

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participant reported outcome that involves judgements and participants were not blinded.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and participants were not blinded.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independant on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness did not depend on its true value.

Low risk of bias

A sham was used, and the scale is validated.

Low risk of bias

No differences between the protocol, the trial register and the journal article.

Low risk of bias

No detected issues with the five domains.

Risk of bias for analysis 2.13 Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.13.1 Dyspnea

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

A sham IMT was used, and the scale is validated.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only 5% of the data were missing.

Low risk of bias

Participants were blinded, and it is a participant reported outcome that involves judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

No ITT analysis

High risk of bias

Missingness is likely to be dependant on its true value.

Some concerns

Some concern with blinding.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and there are still issues with blinding. There is some concern with ITT, missingness is likely to be dependant on the outcome and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Some of the participants were excluded, but still less than 5%.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with ITT analysis, important amount of missing data and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Subgroup 2.13.2 Fatigue

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

A sham IMT was used, and the scale is validated.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only 5% of the data were missing.

Low risk of bias

Participants were blinded, and it is a participant reported outcome that involves judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

High risk of bias

No ITT analysis

High risk of bias

Missingness is likely to be dependant on its true value.

Some concerns

Some concern with blinding.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and there are still issues with blinding. There is some concern with ITT, missingness is likely to be dependant on the outcome and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Some of the participants were excluded, but still less than 5%.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with ITT analysis, important amount of missing data and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Subgroup 2.13.3 Emotion

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

A sham IMT was used, and the scale is validated.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only 5% of the data were missing.

Low risk of bias

Participants were blinded, and it is a participant reported outcome that involves judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Some of the participants were excluded, but still less than 5%.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with ITT analysis, important amount of missing data and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Subgroup 2.13.4 Mastery

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

A sham IMT was used, and the scale is validated.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Only 5% of the data were missing.

Low risk of bias

Participants were blinded, and it is a participant reported outcome that involves judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

Some of the participants were excluded, but still less than 5%.

Some concerns

Missingness was balanced between the two groups.

Low risk of bias

Participants were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with ITT analysis, important amount of missing data and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

High risk of bias

Participants were not blinded.

Some concerns

Insufficient details in the trial register.

High risk of bias

Lack of details about the randomisation process,probably participant were not blinded and there are differences in reporting blinding between the trial register and journal article

Risk of bias for analysis 2.14 Health‐related quality of life (HRQoL): COPD Assessment Test (CAT).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

ITT analysis was conducted

Low risk of bias

Missingness was unlikely to be dependent on its true value

Low risk of bias

A sham IMT was used

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted

Low risk of bias

The reasons for missingness are independent on the intervention

Low risk of bias

A sham IMT was used

Low risk of bias

No differences between the protocol, the register and the journal article

Low risk of bias

No detected issues in the five domains

Risk of bias for analysis 2.15 Inspiratory muscle strength: PImax (cmH₂O).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reason for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment or a trial register. 26% of the participants dropped out of the study with no reason reported for 11%.

Belman 1988

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

missing data are due to intercurrent illnessess and pneumothorax.

Low risk of bias

Observed‐reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Berton 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment and if adjusted analysis were planned in advance.

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

No details about allocation concealment, randomisation and only the journal article is available.

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The reason of missingness was loss of contact.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Covey 2001

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Some concerns

No ITT analysis were conducted, with no impact on the results.

Some concerns

Missingness is balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, no ITT was conducted, missingness could depend on its true value, and only the journal article is available.

Cutrim 2019

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involves judgements, and the measurement was valid.

Low risk of bias

No differences between the trial register and the journal article

Low risk of bias

No detected issues with the five domains.

Dacha 2019

Low risk of bias

From personal communication with the trialist, the suequence was random, concealed, and the groups had similar baseline characteristics.

Low risk of bias

From personal communication, ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial register are available (study ongoing)

Some concerns

Only the abstract and the trial register are available (ongoing study).

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

Missingness in each group was not explained.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Heijdra 1996

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was correct, and it is an observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randmisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Kim 1993

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could depend on the outcome.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

Observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded, and the method of measurement was valid.

Low risk of bias

No differences between the trial register, the abstracts, and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1988

Low risk of bias

Allocation sequence was concealed, and both groups had similar baseline characteristics.

High risk of bias

4 participants out of 45 were discarded (around 8%), and also two participants were not reported although they completed the training.

Some concerns

Reasons for drop out were not reported, but the proportion of missing data is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Some participants were excluded with no provided reasons, no ITT was conducted, missingness is unlikely to affect the outcome, and only the trial register is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if the reasons of exclusion might have an impact on the results.

High risk of bias

some of the reasons of missingness are: lack of interest, inability to perform training, poor adherence.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

3 participants (5%) were excluded due to poor adherence.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about the randomisation process, no ITT analysis, important amount of missing data and only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with  ITT analysis, important amount of missing data and only the journal article is available.

Petrovic 2012

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement is valid.

Some concerns

No additional information were provided in the trial register.

Some concerns

Lack of details about the randomisation process and the trial register does not provide supplementary information.

Preusser 1994

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

12% of the participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted appropriately.

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

The outcome was measured according to international guidelines

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Weiner 2006

Some concerns

Randomisation was reported only as a sentence, and both groups had similar PImax level.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issue with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded (6%)

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Some issues with ITT analysis, missingness could be dependant on its true value, and only the journal article is available.

Risk of bias for analysis 2.16 Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: duration of the intervention).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.16.1 Short‐term (<4 weeks)

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not invovle judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Subgroup 2.16.2 Medium‐term (≥4 weeks and <8 weeks)

Belman 1988

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

missing data are due to intercurrent illnessess and pneumothorax.

Low risk of bias

Observed‐reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

No details about allocation concealment, randomisation and only the journal article is available.

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The reason of missingness was loss of contact.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

Observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Larson 1988

Low risk of bias

Allocation sequence was concealed, and both groups had similar baseline characteristics.

High risk of bias

No ITT analysis.

Some concerns

Reasons for drop out were not reported, but the proportion of missing data is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Some participants were excluded with no reasons provided, no ITT was conducted missingness is unlikely to affect the outcome, and only the trial register is available.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

3 participants (5%) were excluded due to poor adherence.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about the randomisation process, no ITT analysis, important amount of missing data and only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with  ITT analysis, important amount of missing data and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

12% of the participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted appropriately.

Subgroup 2.16.3 Long‐term (≥8 weeks)

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reason for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment or a trial register. 26% of the participants dropped out of the study with no reason reported for 11%.

Berton 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment and if adjusted analysis were planned in advance.

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The reason of missingness was loss of contact.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Covey 2001

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Some concerns

No ITT analysis were conducted, with no impact on the results.

Some concerns

Missingness is balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, no ITT was conducted, missingness could depend on its true value, and only the journal article is available.

Cutrim 2019

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involves judgements, and the measurement was valid.

Low risk of bias

No differences between the trial register and the journal article

Low risk of bias

No detected issues with the five domains.

Dacha 2019

Low risk of bias

From personal communication with the trialist, the suequence was random, concealed, and the groups had similar baseline characteristics.

Low risk of bias

From personal communication with the trialist, ITT was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial register are available (study ongoing)

Some concerns

Only the abstract and the trial register are available (ongoing study).

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

Missingness in each group was not explained.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Heijdra 1996

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was correct and it is an observer‐reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Kim 1993

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could depend on the outcome.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded, and the method of measurement was valid.

Low risk of bias

No differences between the trial register, the abstracts, and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1988

Low risk of bias

Allocation sequence was concealed, and both groups had similar baseline characteristics.

High risk of bias

No ITT analysis.

Some concerns

Reasons for drop out were not reported, but the proportion of missing data is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Some participants were excluded with no reasons provided, no ITT was conducted missingness is unlikely to affect the outcome, and only the trial register is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if the reasons of exclusion might have an impact on the results.

High risk of bias

some of the reasons of missingness are: lack of interest, inability to perform training, poor adherence.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Petrovic 2012

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement is valid.

Some concerns

No additional information were provided in the trial register.

Some concerns

Lack of details about the randomisation process and the trial register does not provide supplementary information.

Preusser 1994

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

12% of the participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted appropriately.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independant on its true value

Low risk of bias

The outcome was measured according to international guidelines

Some concerns

The trial register did not provide sufficient information

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Weiner 2006

Some concerns

Randomisation was reported only as a sentence, and both groups had similar PImax level.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issue with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded (6%)

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Some issues with ITT analysis, missingness could be dependant on its true value, and only the journal article is available.

Risk of bias for analysis 2.17 Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: with or without respiratory muscle weakness).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.17.1 With respiratory muscle weakness

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The reason of missingness was loss of contact.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

Missingness in each group was not explained.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Preusser 1994

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Weiner 2006

Some concerns

Randomisation was reported only as a sentence, and both groups had similar PImax level.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issue with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded (6%)

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Some issues with ITT analysis, missingness could be dependant on its true value, and only the journal article is available.

Subgroup 2.17.2 Without respiratory muscle weakness

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reason for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment or a trial register. 26% of the participants dropped out of the study with no reason reported for 11%.

Belman 1988

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

missing data are due to intercurrent illnessess and pneumothorax.

Low risk of bias

Observed‐reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Berton 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment and if adjusted analysis were planned in advance.

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

No details about allocation concealment, randomisation and only the journal article is available.

Covey 2001

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Some concerns

No ITT analysis were conducted, with no impact on the results.

Some concerns

Missingness is balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, no ITT was conducted, missingness could depend on its true value, and only the journal article is available.

Dacha 2019

Low risk of bias

From personal communication with the trialist, the suequence was random, concealed, and the groups had similar baseline characteristics.

Low risk of bias

From personal communication with the trialist, ITT was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial register are available (study ongoing)

Some concerns

Only the abstract and the trial register are available (ongoing study).

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Kim 1993

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could depend on the outcome.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

Observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded, and the method of measurement was valid.

Low risk of bias

No differences between the trial register, the abstracts, and the journal article.

Low risk of bias

No detected issues with the five domains.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if the reasons of exclusion might have an impact on the results.

High risk of bias

some of the reasons of missingness are: lack of interest, inability to perform training, poor adherence.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

3 participants (5%) were excluded due to poor adherence.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about the randomisation process, no ITT analysis, important amount of missing data and only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with  ITT analysis, important amount of missing data and only the journal article is available.

Petrovic 2012

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement is valid.

Some concerns

No additional information were provided in the trial register.

Some concerns

Lack of details about the randomisation process and the trial register does not provide supplementary information.

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

12% of the participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted appropriately.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their initial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

The outcome was measured according to international guidelines

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issue with the five domains.

Risk of bias for analysis 2.18 Inspiratory muscle strength: PImax (cmH₂O) (subgroup analysis: method of measurement).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 2.18.1 Residual Volume (RV)

Beckerman 2005

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

High risk of bias

No reason for drop out were reported for five participants (11%).

High risk of bias

11 participants dropped out (6 died). However, No reasons for drop out were reported for the remaining participants.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

High risk of bias

No information about allocation concealment or a trial register. 26% of the participants dropped out of the study with no reason reported for 11%.

Covey 2001

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Some concerns

No ITT analysis were conducted, with no impact on the results.

Some concerns

Missingness is balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, no ITT was conducted, missingness could depend on its true value, and only the journal article is available.

Cutrim 2019

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involves judgements, and the measurement was valid.

Low risk of bias

No differences between the trial register and the journal article

Low risk of bias

No detected issues with the five domains.

Heijdra 1996

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

Participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

The method of measurement was correct, and it is an observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randmisation process, and only the journal article is available.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Hsiao 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

High risk of bias

10% of participants were excluded (per protocol analysis).

Some concerns

Missingness is unlikely to be dependant on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. There are serious issues with ITT and missingness is likely to be dependant on the outcome.

Kim 1993

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

Participants were analysed according to their initial allocation.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could depend on the outcome.

Koppers 2006

Some concerns

Randomisation was reported only as a sentence, the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

7% of the data are missing (3 participants out of 39).

Low risk of bias

Observer reported outcome that do not involve judgements.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Larson 1988

Low risk of bias

Allocation sequence was concealed, and both groups had similar baseline characteristics.

High risk of bias

4 participants out of 45 were discarded (around 8%), and also two participants were not reported although they completed the training.

Some concerns

Reasons for drop out were not reported, but the proportion of missing data is balanced between the two groups.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Some participants were excluded with no provided reasons, no ITT was conducted, missingness is unlikely to affect the outcome, and only the trial register is available.

Larson 1999

Some concerns

Randomisation was reported only as a sentence.

Some concerns

It is unclear if the reasons of exclusion might have an impact on the results.

High risk of bias

some of the reasons of missingness are: lack of interest, inability to perform training, poor adherence.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, and only the trial register is available. There is some concern with ITT and missingness is likely to be dependant on the outcome.

Leelarungrayub 2017

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

High risk of bias

6 participants discontinued training, but there is no mention of the proportion of missingness in each group.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process and only the journal article is available. Missingness may be dependant on its true value.

Saher 2021

Some concerns

Randomisation was reported only as a sentence. Authors used concealed envelopes without mentioning if they were opaque

High risk of bias

ITT analysis was not applied

Some concerns

Missingness could depend on its true value

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available

High risk of bias

Lack of details about the randomisation process, only the journal article is available, missingness could depend on its true value, and ITT analysis was not conducted.

Scherer 2000

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

One participant excluded due to rip fracture.

Some concerns

Missingness was balanced between the two groups

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, only the journal article is available and missingness could be dependent on its true value.

Weiner 2003

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

High risk of bias

data in the graph are different from what was reported in the text.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and data in the graphs are different from the text.

Weiner 2006

Some concerns

Randomisation was reported only as a sentence, and both groups had similar PImax level.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Wu 2017

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

There were some differences between the trial register and the journal article, without biasing the results.

Some concerns

Lack of details about the randomisation process and there are differences in reporting blinding between the trial register and journal article.

Subgroup 2.18.2 Functional Residual Capacity (FRC)

Belman 1988

Some concerns

Randomisation was reported only as a sentence.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

missing data are due to intercurrent illnessess and pneumothorax.

Low risk of bias

Observed‐reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Harver 1989

Some concerns

Randomisation was reported only as a sentence, and both groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

High risk of bias

Missingness in each group was not explained.

Low risk of bias

Observer reported outcome that do not involve judgement, and the method of measurement was valid.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, missingness, and only the journal article is available.

Hill 2006

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The data of only two participants (5%) are missing.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment, and only the journal article is available.

Langer 2018

Low risk of bias

Allocation was random, concealed, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded, and the method of measurement was valid.

Low risk of bias

No differences between the trial register, the abstracts, and the journal article.

Low risk of bias

No detected issues with the five domains.

Lisboa 1997

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about allocation concealment and only the journal article is available.

Nikoletou 2016

Some concerns

No information about allocation concealment, and the two groups had similar baseline characteristics.

Some concerns

3 participants (5%) were excluded due to poor adherence.

Some concerns

Missingness is balanced between the two groups.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Lack of details about the randomisation process, no ITT analysis, important amount of missing data and only the journal article is available.

Some concerns

Lack of details about the randomisation process, some issues with  ITT analysis, important amount of missing data and only the journal article is available.

Petrovic 2012

Some concerns

Randomisation was reported only as a sentence, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Observer reported outcome that do not involve judgements, and the method of measurement is valid.

Some concerns

No additional information were provided in the trial register.

Some concerns

Lack of details about the randomisation process and the trial register does not provide supplementary information.

Preusser 1994

Some concerns

No information about allocation concealment, and both groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Sanchez Riera 2001

Some concerns

Randomisation was reported only as a sentence, and baseline characteristics were similar between the groups.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process and only the journal article is available.

Subgroup 2.18.3 Not reported

Berton 2015

Some concerns

No information about allocation concealment.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Some concerns

No information if adjusted analysis were planned in advance.

Some concerns

No information about allocation concealment and if adjusted analysis were planned in advance.

Bustamante 2007

Some concerns

Only a statement about randomisation was reported, the two groups had similar baseline characteristics.

Low risk of bias

No participants were excluded.

Low risk of bias

No missing data.

Low risk of bias

Observer‐reported outcome that do not involve judgements, and the method of measurement was valid.

Some concerns

Only the journal article is available.

Some concerns

No details about allocation concealment, randomisation and only the journal article is available.

Chuang 2017

Some concerns

Only a statement about randomisation was reported, and the two groups had similar baseline characteristics.

Low risk of bias

All participants were analysed according to their initial allocation.

Low risk of bias

The reason of missingness was loss of contact.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

Some concerns

Lack of details about the randomisation process, and only the journal article is available.

Dacha 2019

Low risk of bias

From personal communication with the trialist, the suequence was random, concealed, and the groups had similar baseline characteristics.

Low risk of bias

From personal communication, ITT analysis was conducted.

Low risk of bias

No missing data.

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the abstract and the trial register are available (study ongoing)

Some concerns

Only the abstract and the trial register are available (ongoing study).

Ramirez Sarmiento 2002

Some concerns

Randomisation was reported only as a sentence, and no differences between the groups at visual inspection.

High risk of bias

12% of the participants were excluded.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Observer reported outcome that do not involve judgement.

Some concerns

Only the journal article is available.

High risk of bias

Lack of details about the randomisation process, only the journal article is available and ITT analysis was not conduted appropriately.

Saka 2021

Some concerns

Lack of details about allocation concealment

Low risk of bias

All participants were analysed according to their intial allocation

Low risk of bias

Missingness is independent on its true value

Low risk of bias

The outcome was measured according to international guidelines

Some concerns

Lack of details in the trial register

Some concerns

Lack of details about the randomisation process, and the trial register did not provide sufficient information.

Xu 2018

Low risk of bias

The sequence was random, concealed, and the two groups had similar baseline characteristics.

Low risk of bias

ITT analysis was conducted.

Low risk of bias

Missingness is independent on its true value.

Low risk of bias

Outcome assessors were blinded.

Low risk of bias

No differences between the trial register, the conference abstracts and the journal article.

Low risk of bias

No detected issue with the five domains.

ZhouL 2016

Low risk of bias

Allocation was concealed.

Some concerns

3 participants were excluded (6%)

Some concerns

No details of the reasons of missingness, but its proportion is small (6%).

Low risk of bias

Outcome assessors were blinded.

Some concerns

Only the journal article is available.

Some concerns

Some issues with ITT analysis, missingness could be dependant on its true value, and only the journal article is available.

Appendices

Appendix 1. Database search strategies

Database/search platform/date of last search	Search strategy	Results
Airways Register (via Cochrane Register of Studies) Date of most recent search: 19 October 2022	1 MeSH DESCRIPTOR Pulmonary Disease, Chronic Obstructive Explode All AND INSEGMENT 2 MeSH DESCRIPTOR Bronchitis, Chronic AND INSEGMENT 3 (obstruct*) near3 (pulmonary or lung* or airway* or airflow* or bronch* or respirat*) AND INSEGMENT 4 COPD:MISC1 AND INSEGME 5 (COPD OR AECOPD):TI,AB,KW AND INSEGMENT 6 #1 OR #2 OR #3 OR #4 OR #5 AND INSEGMENT 7 MESH DESCRIPTOR Breathing Exercises AND INSEGMENT 8 MESH DESCRIPTOR Respiratory Muscles EXPLODE ALL AND INSEGMENT 9 MESH DESCRIPTOR Exercise Tolerance AND INSEGMENT 10 (IMT or RMT):ti,ab AND INSEGMENT 11 ((inspiratory or ventilat* or respiratory) NEAR3 (muscle or resistance) NEAR3 (train* or strength* or endur*)) AND INSEGMENT 12 (threshold NEAR2 (load or device*)) AND INSEGMENT 13 (resist* NEAR2 device*) AND INSEGMENT 14 isocapnic hyperpnea AND INSEGMENT 15 #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #14 OR #13 16 #6 AND #15 17 INREGISTER 18 #16 AND #17	Oct 2020=937   Aug 2021=44   Oct 2022=265 ( together with CENTRAL)
CENTRAL (via Cochrane Register of Studies) Date of most recent search: 19 October 2022	1    MeSH DESCRIPTOR Pulmonary Disease, Chronic Obstructive Explode All AND CENTRAL:TARGET 2    MeSH DESCRIPTOR Bronchitis, Chronic AND CENTRAL:TARGET 3    (obstruct*) near3 (pulmonary or lung* or airway* or airflow* or bronch* or respirat*) AND CENTRAL:TARGET 4    COPD:MISC1 AND CENTRAL:TARGET 5    (COPD OR AECOPD):TI,AB,KW AND CENTRAL:TARGET 6    #1 OR #2 OR #3 OR #4 OR #5 AND CENTRAL:TARGET 7    MESH DESCRIPTOR Breathing Exercises AND CENTRAL:TARGET 8    MESH DESCRIPTOR Respiratory Muscles EXPLODE ALL AND CENTRAL:TARGET 9    MESH DESCRIPTOR Exercise Tolerance AND CENTRAL:TARGET 10    (IMT or RMT):ti,ab AND CENTRAL:TARGET 11    ((inspiratory or ventilat* or respiratory) NEAR3 (muscle or resistance) NEAR3 (train* or strength* or endur*)) AND CENTRAL:TARGET 12    (threshold NEAR2 (load or device*)) AND CENTRAL:TARGET 13    (resist* NEAR2 device*) AND CENTRAL:TARGET 14    isocapnic hyperpnea AND CENTRAL:TARGET 15    #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #14 OR #13 AND CENTRAL:TARGET 16    #6 AND #15 AND CENTRAL:TARGET 17    CENTRAL:TARGET 18    #16 AND #17 AND CENTRAL:TARGET	Oct 2020=1239   Aug 2021=69   Oct 2022=265 ( together with the Airways Register)
MEDLINE (Ovid) ALL Date of most recent search: 20 October 2022	1    Lung Diseases, Obstructive/ 2    exp Pulmonary Disease, Chronic Obstructive/ 3    emphysema$.tw. 4    (chronic$ adj3 bronchiti$).tw. 5    (obstruct$ adj3 (pulmonary or lung$ or airway$ or airflow$ or bronch$ or respirat$)).tw. 6    (COPD or AECOPD or AECB).ti,ab. 7    or/1‐6 8    Breathing Exercises/ 9    exp Respiratory Muscles/ 10    Exercise Tolerance/ 11    Muscle Strength/ 12    (IMT or RMT).ti,ab. 13    ((inspiratory or ventilat$ or respiratory) adj3 (muscle or resistance) adj3 (train$ or strength$ or endur$)).tw. 14    (threshold adj2 (load or device$)).tw. 15    (resist$ adj2 device$).tw. 16    isocapnic hyperpnea.tw. 17    or/8‐16 18    (controlled clinical trial or randomized controlled trial).pt. 19    (randomized or randomised).ab,ti. 20    placebo.ab,ti. 21    randomly.ab,ti. 22    trial.ab,ti. 23    groups.ab,ti. 24    or/18‐23 25    Animals/ 26    Humans/ 27    25 not (25 and 26) 28    24 not 27 29    7 and 17 and 28	Oct 2020=1609   Aug 2021=86   Oct 2022=172
Embase (Ovid) Date of most recent search: 20 October 2022	1    Chronic Obstructive Lung Disease/ 2    Obstructive Airway Disease/ 3    Chronic Bronchitis/ 4    Lung Emphysema/ 5    emphysema$.tw. 6    (chronic$ adj3 bronchiti$).tw. 7    (obstruct$ adj3 (pulmonary or lung$ or airway$ or airflow$ or bronch$ or respirat$)).tw. 8    (COPD or AECOPD or AECB).ti,ab. 9    or/1‐8 10    breathing exercise/ 11    breathing muscle/ 12    exercise tolerance/ 13    muscle strength/ 14    (IMT or RMT).ti,ab. 15    ((inspiratory or ventilat$ or respiratory) adj3 (muscle or resistance) adj3 (train$ or strength$ or endur$)).tw. 16    (threshold adj2 (load or device$)).tw. 17    (resist$ adj2 device$).tw. 18    isocapnic hyperpnea.tw. 19    or/10‐18 20    Randomized Controlled Trial/ 21    randomization/ 22    controlled clinical trial/ 23    Double Blind Procedure/ 24    Single Blind Procedure/ 25    Crossover Procedure/ 26    (clinica$ adj3 trial$).tw. 27    ((singl$ or doubl$ or trebl$ or tripl$) adj3 (mask$ or blind$ or method$)).tw. 28    exp Placebo/ 29    placebo$.ti,ab. 30    random$.ti,ab. 31    ((control$ or prospectiv$) adj3 (trial$ or method$ or stud$)).tw. 32    (crossover$ or cross‐over$).ti,ab. 33    or/20‐32 34    exp animals/ or exp invertebrate/ or animal experiment/ or animal model/ or animal tissue/ or animal cell/ or nonhuman/ 35    human/ or normal human/ or human cell/ 36    34 and 35 37    34 not 36 38    33 not 37 39    9 and 19 and 38	Oct 2020=2093   Aug 2021=146   Oct 2022=251
CINHAL (EBSCO)  Date of most recent search: 19 October 2022	S41    S7 AND S17 AND S40 S40    S39 NOT S38 S39    S18 OR S19 OR S20 OR S21 OR S22 OR S23 OR S24 OR S25 OR S26 OR S27 OR S28 OR S29 OR S30 OR S31 OR S32 S38    S36 NOT S37 S37    (MH "Human") S36    S33 OR S34 OR S35 S35    TI (animal model*) S34    (MH "Animal Studies") S33    (MH "Animals+") S32    AB (cluster W3 RCT) S31    MH (crossover design) OR MH (comparative studies) S30    AB (control W5 group) S29    PT (Randomized Controlled Trial) S28    (MH "Placebos") S27    MH ("sample size") AND AB (assigned OR allocated OR control) S26    TI (trial) S25    AB (random*)   S24    TI (randomised OR randomized) S23    (MH "Cluster Sample") S22    (MH "Pretest‐Posttest Design") S21    (MH "Random Assignment") S20    (MH "Single‐Blind Studies") S19    (MH "Double‐Blind Studies") S18    (MH "Randomized Controlled Trials") S17    S8 OR S9 OR S10 OR S11 OR S12 OR S13 OR S14 OR S15 OR S16 S16    isocapnic hyperpnea S15    resist* N2 device* S14    threshold* N2 (load or device*) S13    (inspiratory or ventilat* or respiratory) N3 (muscle or resistance) N3 (train* or strength* or endur*) S12    AB (IMT or RMT) or TI (IMT or RMT) S11    (MH "Muscle Strength") S10    (MH "Exercise Tolerance" S9    (MH "Respiratory Muscles+") S8    (MH "Breathing Exercises") S7    S1 or S2 or S3 or S4 or S5 or S6 S6    COPD or COAD or COBD or AECB S5    (pulmonary or lung* or airway* or airflow* or bronch* or respirat*) and (obstruct*) and (disease*) S4    chronic bronchitis S3    "emphysema*" S2    (MH "Lung Diseases, Obstructive") S1    (MH "Pulmonary Disease, Chronic Obstructive+")	Oct 2020=182   Aug 2021=22   Oct 2022=76
PsycINFO (Ovid) Date of most recent search: 19 October 2022	1    exp Chronic Obstructive Pulmonary Disease/ 2    emphysema$.tw. 3    (chronic$ adj3 bronchiti$).tw. 4    (obstruct$ adj3 (pulmonary or lung$ or airway$ or airflow$ or bronch$ or respirat$)).tw. 5    (COPD or AECOPD or AECB).ti,ab. 6    or/1‐5 7    ((inspiratory or ventilat$ or respiratory) adj3 (muscle or resistance) adj3 (train$ or strength$ or endur$)).tw. 8    (IMT or RMT).ti,ab. 9    (threshold adj2 (load or device$)).tw. 10    (resist$ adj2 device$).tw. 11    isocapnic hyperpnea.tw. 12    or/7‐11 13    random$.tw. 14    (clinical adj5 trial$).tw. 15    (control$ adj5 trial$).tw. 16    ((clinical or control$ or comparativ$) adj5 (study or studies)).tw. 17    placebo$.tw. 18    (single blind$ or single‐blind$).tw. 19    (double blind$ or double‐blind$).tw. 20    (triple blind$ or triple‐blind$).tw. 21    or/13‐20 22    6 and 12 and 21	Oct 2020=10   August 2021=1   Oct 2022=3
PEDro  Date of most recent search: 20 October 2022	Abstract & title: inspiratory muscle copd random* Topic: chronic respiratory disease Method: Clinical trial When searching: Match all search terms (AND)	Oct 2020=81   Aug 2021=1   Oct 2022=5
ClinicalTrials.gov Date of most recent search: 20 October 2022	Study type:    Interventional Condition:    COPD Intervention:    inspiratory muscle training OR threshold load OR threshold device OR resistive device	Oct 2020=58   Aug 2021=6   Oct 2022=12
WHO trials portal Date of most recent search: 20 October 2022	Condition:    COPD Intervention:    inspiratory muscle training OR threshold load OR threshold device OR resistive device	Oct 2020=347   Aug 2021=2   Oct 2022=9

Data and analyses

Comparison 1. PR+IMT vs PR.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Dyspnea: Borg (at submaximal exercise: 50% to 80% of Wmax)	2	202	Mean Difference (IV, Random, 95% CI)	0.19 [‐0.42, 0.79]
1.2 Dyspnea: Modified Medical Research Council (mMRC)	2	204	Mean Difference (IV, Random, 95% CI)	‐0.12 [‐0.39, 0.14]
1.3 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters)	12	1199	Mean Difference (IV, Random, 95% CI)	5.95 [‐5.73, 17.63]
1.4 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention)	12		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.4.1 Short‐term (<4 weeks)	4	687	Mean Difference (IV, Random, 95% CI)	10.25 [‐24.98, 45.49]
1.4.2 Medium‐term ( ≥4 weeks and <8 weeks)	2	178	Mean Difference (IV, Random, 95% CI)	36.72 [‐67.67, 141.11]
1.4.3 Long‐term ( ≥8 weeks)	7	363	Mean Difference (IV, Random, 95% CI)	7.82 [‐6.90, 22.54]
1.5 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness)	12		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.5.1 With respiratory muscle weakness	3	219	Mean Difference (IV, Random, 95% CI)	31.17 [‐10.50, 72.84]
1.5.2 Without respiratory muscle weakness	9	981	Mean Difference (IV, Random, 95% CI)	0.38 [‐11.65, 12.42]
1.6 Functional exercise capacity: 12‐minute walk distance (12MWD) (meters)	3	80	Mean Difference (IV, Random, 95% CI)	155.77 [‐84.53, 396.08]
1.7 Functional exercise capacity: Wmax (watt)	5	326	Mean Difference (IV, Random, 95% CI)	‐1.01 [‐6.96, 4.94]
1.8 Functional exercise capacity: exercise time (seconds)	3	192	Mean Difference (IV, Random, 95% CI)	58.62 [‐25.09, 142.32]
1.9 Health‐related quality of life (HRQoL): St George's Respiratory Questionnaire (SGRQ)	7		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.9.1 Symptoms	2	169	Mean Difference (IV, Random, 95% CI)	‐2.33 [‐6.28, 1.62]
1.9.2 Activity	2	169	Mean Difference (IV, Random, 95% CI)	0.28 [‐1.65, 2.20]
1.9.3 Impact	2	169	Mean Difference (IV, Random, 95% CI)	‐1.63 [‐5.38, 2.11]
1.9.4 Total	7	908	Mean Difference (IV, Random, 95% CI)	0.13 [‐0.93, 1.20]
1.10 Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ)	3		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.10.1 Dyspnea	3		Mean Difference (IV, Random, 95% CI)	‐0.30 [‐1.90, 1.29]
1.10.2 Fatigue	3		Mean Difference (IV, Random, 95% CI)	0.28 [‐0.76, 1.31]
1.10.3 Emotion	2		Mean Difference (IV, Random, 95% CI)	‐0.63 [‐2.53, 1.26]
1.10.4 Mastery	2		Mean Difference (IV, Random, 95% CI)	‐0.05 [‐1.18, 1.08]
1.11 Health‐related quality of life (HRQoL): COPD Assessment Test (CAT)	2	657	Mean Difference (IV, Random, 95% CI)	0.13 [‐0.80, 1.06]
1.12 Inspiratory muscle strength: PImax (cmH20)	17	1329	Mean Difference (IV, Random, 95% CI)	11.46 [7.42, 15.50]
1.13 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: duration of the intervention)	17		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.13.1 Short‐term (<4 weeks)	4	687	Mean Difference (IV, Random, 95% CI)	12.63 [4.14, 21.11]
1.13.2 Medium‐term ( ≥ 4 weeks and <8 weeks)	4	233	Mean Difference (IV, Random, 95% CI)	12.27 [3.75, 20.79]
1.13.3 Long‐term ( ≥ 8 weeks)	11	478	Mean Difference (IV, Random, 95% CI)	11.52 [5.50, 17.53]
1.14 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: with or without respiratory muscle weakness)	16		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.14.1 With respiratory muscle weakness	5	282	Mean Difference (IV, Random, 95% CI)	14.84 [11.35, 18.34]
1.14.2 Without respiratory muscle weakness	11	1031	Mean Difference (IV, Random, 95% CI)	10.57 [5.23, 15.91]
1.15 Laboratory exercise test: VO2peak (L/min)	5	313	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.05, 0.03]
1.16 Respiratory muscle endurance: respiratory muscle endurance pressure (Pthmax) (cmH2O)	2	52	Std. Mean Difference (IV, Random, 95% CI)	1.22 [‐0.18, 2.62]
1.17 Respiratory muscle endurance time: Tlim (seconds) (sustained ventilation according to PImax)	3	236	Mean Difference (IV, Random, 95% CI)	84.63 [‐50.77, 220.02]
1.18 Respiratory muscle endurance time: Tlim (seconds) (sustained ventilation according to MVV)	2	51	Mean Difference (IV, Random, 95% CI)	477.69 [215.43, 739.94]
1.19 Maximal voluntary ventilation (MVV)	3	93	Std. Mean Difference (IV, Random, 95% CI)	0.40 [‐0.02, 0.83]
1.20 Respiratory function: forced expiratory volume at 1 second (FEV1) (%Pred)	6	173	Mean Difference (IV, Random, 95% CI)	0.77 [‐1.72, 3.26]
1.21 Respiratory function: forced expiratory volume at 1 second (FEV1) (Liters)	6	889	Mean Difference (IV, Random, 95% CI)	0.04 [‐0.04, 0.13]

Comparison 2. IMT vs control/sham.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Dyspnea: Borg (at submaximal exercise capacity)	6	144	Mean Difference (IV, Random, 95% CI)	‐0.94 [‐1.36, ‐0.51]
2.2 Dyspnea: Baseline and Transition Dyspnea Indexes (BDI‐TDI)	8		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.2.1 Functional impairment	3	95	Mean Difference (IV, Random, 95% CI)	0.88 [0.51, 1.25]
2.2.2 Magnitude of task	3	95	Mean Difference (IV, Random, 95% CI)	0.73 [0.35, 1.12]
2.2.3 Magnitude of effort	3	95	Mean Difference (IV, Random, 95% CI)	0.86 [0.42, 1.30]
2.2.4 Focal score	8	238	Mean Difference (IV, Random, 95% CI)	2.98 [2.07, 3.89]
2.3 Dyspnea: Transition Dyspnea Index (TDI): Focal score (subgroup analysis: with or without respiratory muscle weakness)	7		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.3.1 With respiratory muscle weakness	4	152	Mean Difference (IV, Random, 95% CI)	3.52 [2.55, 4.49]
2.3.2 Without respiratory weakness	3	70	Mean Difference (IV, Random, 95% CI)	2.28 [1.10, 3.46]
2.4 Dyspnea: Modified Medical Research Council (mMRC)	4	150	Mean Difference (IV, Fixed, 95% CI)	‐0.59 [‐0.76, ‐0.43]
2.5 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters)	16	501	Mean Difference (IV, Random, 95% CI)	35.71 [25.68, 45.74]
2.6 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: duration of the intervention)	16		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.6.1 Short‐term (<4 weeks)	1	34	Mean Difference (IV, Random, 95% CI)	33.06 [23.05, 43.07]
2.6.2 Medium‐term (≥4 weeks and <8 weeks)	4	131	Mean Difference (IV, Random, 95% CI)	31.15 [1.50, 60.81]
2.6.3 Long‐term (≥8 weeks)	11	336	Mean Difference (IV, Random, 95% CI)	38.47 [22.75, 54.20]
2.7 Functional exercise capacity: 6‐minute walk distance (6MWD) (meters) (subgroup analysis: with or without respiratory muscle weakness)	15		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.7.1 With respiratory muscle weakness	5	178	Mean Difference (IV, Random, 95% CI)	33.74 [25.08, 42.40]
2.7.2 Without respiratory muscle weakness	11	291	Mean Difference (IV, Random, 95% CI)	29.80 [12.86, 46.73]
2.8 Functional exercise capacity: 12‐minute walk distance (12MWD) (meters)	3	101	Mean Difference (IV, Random, 95% CI)	‐33.31 [‐158.10, 91.48]
2.9 Functional exercise capacity: Wmax (watt)	7	206	Mean Difference (IV, Random, 95% CI)	0.66 [‐6.44, 7.76]
2.10 Functional exercise capacity: exercise time (seconds)	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.11 Functional exercise capacity: shuttle walk test (SWT) (meters)	2	57	Mean Difference (IV, Random, 95% CI)	‐7.45 [‐92.74, 77.83]
2.12 Health‐related quality of life (HRQoL): St George Respiratory Questionnaire (SGRQ)	6		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.12.1 Symptoms	2	53	Mean Difference (IV, Random, 95% CI)	‐2.10 [‐3.50, ‐0.71]
2.12.2 Activity	2	53	Mean Difference (IV, Random, 95% CI)	‐9.86 [‐15.08, ‐4.63]
2.12.3 Impact	2	53	Mean Difference (IV, Random, 95% CI)	‐6.06 [‐13.76, 1.65]
2.12.4 Total	6	182	Mean Difference (IV, Random, 95% CI)	‐3.85 [‐8.18, 0.48]
2.13 Health‐related quality of life (HRQoL): Chronic Respiratory Disease Questionnaire (CRQ)	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.13.1 Dyspnea	5	178	Mean Difference (IV, Random, 95% CI)	1.63 [0.23, 3.03]
2.13.2 Fatigue	5	178	Mean Difference (IV, Random, 95% CI)	1.32 [0.08, 2.55]
2.13.3 Emotion	4	163	Mean Difference (IV, Random, 95% CI)	2.64 [0.82, 4.46]
2.13.4 Mastery	4	154	Mean Difference (IV, Random, 95% CI)	1.57 [0.07, 3.06]
2.14 Health‐related quality of life (HRQoL): COPD Assessment Test (CAT)	2	86	Mean Difference (IV, Random, 95% CI)	‐2.97 [‐3.85, ‐2.10]
2.15 Inspiratory muscle strength: PImax (cmH2O)	32	916	Mean Difference (IV, Random, 95% CI)	14.57 [9.85, 19.29]
2.16 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: duration of the intervention)	32		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.16.1 Short‐term (<4 weeks)	1	34	Mean Difference (IV, Random, 95% CI)	12.60 [6.94, 18.26]
2.16.2 Medium‐term (≥4 weeks and <8 weeks)	8	223	Mean Difference (IV, Random, 95% CI)	11.89 [6.76, 17.02]
2.16.3 Long‐term (≥8 weeks)	26	748	Mean Difference (IV, Random, 95% CI)	15.02 [9.26, 20.78]
2.17 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: with or without respiratory muscle weakness)	28		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.17.1 With respiratory muscle weakness	10	323	Mean Difference (IV, Random, 95% CI)	11.08 [7.51, 14.64]
2.17.2 Without respiratory muscle weakness	19	499	Mean Difference (IV, Random, 95% CI)	13.82 [5.36, 22.29]
2.18 Inspiratory muscle strength: PImax (cmH2O) (subgroup analysis: method of measurement)	32		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.18.1 Residual Volume (RV)	15	466	Mean Difference (IV, Random, 95% CI)	12.63 [8.46, 16.81]
2.18.2 Functional Residual Capacity (FRC)	9	206	Mean Difference (IV, Random, 95% CI)	11.87 [7.33, 16.41]
2.18.3 Not reported	8	244	Mean Difference (IV, Random, 95% CI)	19.52 [4.63, 34.41]
2.19 Laboratory exercise test: VO2peak	11	286	Std. Mean Difference (IV, Random, 95% CI)	0.31 [0.05, 0.57]
2.20 Respiratory muscle endurance: respiratory muscle endurance pressure (Pthmax) (cmH2O)	8	179	Mean Difference (IV, Random, 95% CI)	9.71 [4.93, 14.50]
2.21 Respiratory muscle endurance time: Tlim (seconds)	10	260	Mean Difference (IV, Random, 95% CI)	270.57 [182.44, 358.71]
2.22 Maximal voluntary ventilation (MVV)	2	36	Std. Mean Difference (IV, Random, 95% CI)	0.99 [0.28, 1.69]
2.23 Respiratory function: forced expiratory volume at 1 second (FEV1) (%pred)	10	314	Mean Difference (IV, Random, 95% CI)	2.62 [0.20, 5.04]
2.24 Respiratory function: forced expiratory volume at 1 second (FEV1) (Liters)	12	362	Mean Difference (IV, Random, 95% CI)	0.04 [‐0.06, 0.14]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Abedi Yekta 2019.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 17/15 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 51.33 (10.4) Gender (M/F): 9/6 BMI, mean (SD), kg/m²: 27.6 (3.7) PR N (randomized/analyzed): 17/14 Loss to follow‐up or excluded: 3 Age, mean (SD) in years: 53.5 (10.37) Gender (M/F): 8/6 BMI, mean (SD), kg/m²: 27.39 (5.11) IMT N (randomized/analyzed): 17/16 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 51.88 (9.05) Gender (M/F): 7/9 BMI, mean (SD), kg/m²: 27.05 (4.53) Control/sham N (randomized/analyzed): 17/15 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 55.67 (11.08) Gender (M/F): 7/8 BMI, mean (SD), kg/m²: 25.98 (4.1) Overall N (randomized/analyzed): 68/60 Loss to follow‐up or excluded: 8 Age, mean (SD) in years: 53.07 (10.11) Range age (min, max): 32,70 Gender (M/F): 31/29 COPD stage (GOLD): moderate to severe Included criteria Stage 2 or 3 COPD according to the GOLD criteria; Age range 30‐70 years; No previous known diseases, such as heart disease (e.g. congestive heart failure and coronary artery disease), renal disease (e.g. end‐stage renal disease and chronic renal failure), or liver disease (e.g. hepatic cirrhosis and hepatic cancer); no history of known pulmonary diseases, such as lung cancer and pleural disease; no history of musculoskeletal diseases (e.g. myasthenia gravis) or restrictive deformities of the lungs; Lack of severe limitations in the limbs inhibiting aerobic exercise; no pulmonary surgery in the past 12 months; No recent fracture of the ribs in the past 6 months; No history of psychotropic diseases; and no use of drugs, alcohol, or psychiatric drugs. Excluded criteria Exacerbation of the disease during the study; Need for long‐term oxygen therapy for more than 15 h/d; The occurrence of complications, such as pneumothorax or diseases exacerbating and disrupting the treatment.
Interventions	Intervention characteristics PR: consisted of aerobic exercise performed on a treadmill ergometer, at 40%‐60% of the heart rate, 2 d/week, for 8 weeks and 40 min/session. IMT: conducted 2 d/week for 8 weeks, 15 min/session with Powerbreathe KH4. Each session lasted 15 min, all sessions were supervised, and the training load was set at 40%‐60% of S‐Index. PR+IMT: consisted of a combination of PR and IMT protocols. Control: no intervention received by this group.
Outcomes	HRQoL: SGRQ (Total)
Identification	Sponsorship source: Shahid Beheshti University of Medical Sciences Country: Iran Setting: Hospital Hussein Department of Physiotherapy Author's name: Saeed Rezaei Institution: Department of Sports Medicine, School of Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran Email: saeedrezaee1394@yahoo.com Clinical trial register: https://www.irct.ir/trial/29724
Notes	A change of 4 units in SGRQ score was considered significant.

Bavarsad 2015.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 20/15 Loss to follow‐up or excluded: 5 Age mean (SD) in years: 58.8 (6.82) Gender (M/F): 13/2 BMI mean (SD), kg/m²: 24.88 (5.02) Control/Sham N (randomized/analyzed): 20/15 Loss to follow‐up or excluded: 5 Age mean (SD) in years: 54.2 (8.09) Gender (M/F): 14/1 BMI mean (SD), kg/m²: 24.97 (4.72) Overall N (randomized/analyzed): 40/30 Loss to follow‐up or excluded: 10 Age range (min, max): 45, 65 Gender (M/F): 27/3 COPD stage (GOLD): mild to severe Included criteria Patients with spirometric evidence of significant chronic airflow limitation(FEV1/FVC < 70%pred) with mild to very severe COPD diagnosis according to the GOLD criteria; Age range 45‑65 years; No history of PR; Established treatment plan 1 month prior to the study. Excluded criteria BMI > 35; Comorbid conditions such as diabetes, musculoskeletal disorders, cardiovascular diseases; and neurological diseases that can affect the results of 6MWD; Having exacerbation for 1 month prior to the study; A history of long‑term oxygen therapy; A history of spontaneous pneumothorax and rib fracture.
Interventions	Intervention characteristics IMT The training consisted of unsupervised sessions, 15 min/d, 6 d/week for 8 weeks. The device used was an incentive spirometer (Respivol), at a load equal to or more than the initial inspiratory volume. The researchers were informed of the training sessions through phone calls during the 8 weeks. A checklist, which was designed to be completed on weekdays, was prepared for the participants so that they could mark the relevant day after a training session Control No intervention was received by this group.
Outcomes	Dyspnea: Borg Notes: dyspnea was assessed at the beginning and the end of the 6MWD Functional exercise capacity: 6MWD Respiratory function: FEV1 (%pred) Respiratory function: FEV1 (L)
Identification	Sponsorship source: The Deputy of Research Affairs at the Ahvaz Jundishapur University of Medical Sciences Country: Iran Setting: Specialized Pulmonary Clinic of Ahvaz Author's name: Esmaeil Eidani Institution: Pulmonary Unit, Department of Medicine, Ahvaz Jundishapur University of Medical Sciences Email: esmaileidani@gmail.com Address: Ahzav, Iran
Notes

Beaumont 2015.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group Subgroup analysis: FEV1 > or ≤ 50% of the predicted value
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 16/16 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 62 (10) Gender (M/F): 4/12 BMI mean (SD), kg/m²: 27.3 (4.2) PR N (randomized/analyzed): 18/16 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 61 (8) Gender (M/F): 7/11 BMI mean (SD), kg/m²: 26.8 (6.0) Overall N (randomized/analyzed): 34/32 Loss to follow‐up or excluded: 0 Gender (M/F): 11/23 COPD stage (GOLD): moderate to very severe Included criteria COPD diagnosed according to American Thoracic Society/European Respiratory Society criteria; PImax > 60 cmH2O at admission. Excluded criteria Previous pneumonectomy or lobectomy in the past 6 months; Impossibility to measure IC at the end of the 6MWD; The incapacity to follow the standard rehabilitation program; The absence of written informed consent. Pretreatment: FEV 1 was lower in the IMT group; the Borg scale was higher in the control group.
Interventions	Intervention characteristics PR+IMT PR: conducted for 3 weeks, 5 d/week, and included aerobic exercise on a cycle ergometer and a treadmill (30 min/d each), strengthening of lower and upper limb muscle groups, therapeutic educational program, aerobic gymnastics in groups, smoking cessation program, and socio‐psychological and dietary advice. IMT: performed daily in 2 sessions of 15 min each, 5 times/week, for 3 weeks, supervised by a physiotherapist. The participants had to breathe slowly with an increased tidal volume. A threshold inspiratory muscle trainer (Threshold IMT1) was used at 40% of PImax. The intensity was not modified during the program. PR Participants in this group received only the standardised PR program.
Outcomes	Dyspnea: Borg Dyspnea: MDP Unpleasantness Sensory intensity Muscle work/effort Not enough air/smother/air hunger Mental effort/concentrate Tight/constricted Breathing a lot (rapid, deep, and heavy) Crush Depression Satisfaction Anxiety Frustration Anger  Happiness Fear Functional exercise capacity: 6MWD Respiratory muscle strength: PImax
Identification	Sponsorship source: EA3878 (G.E.T.B.O.), CIC INSERM 0502, University Hospital of Brest Country: France Setting: Pulmonary Rehabilitation Unit, Morlaix Hospital Centre, European University of Occidental Brittany Author's name: Marc Beaumont Institution: Pulmonary Rehabilitation Unit, Morlaix HospitalCentre, European University of Occidental Brittany Email: marc.beaumont@univ‐brest.fr Address: Morlaix29672, Cedex, France Clinical trial register: NCT01545011
Notes	P‐value of the change from baseline is from the adjusted analysis.

Beaumont 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group Subgroup analysis: PImax (> or ≤ 60 cmH2O)
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 74/74 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 62.2 (8.0) Gender (M/F): 44/30 BMI, mean (SD), kg/m²: 26.2 (5.9) PR N (randomized/analyzed): 75/75 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 65.9 (8.9) Gender (M/F): 50/25 BMI, mean (SD), kg/m²: 24.7 (5.9) Overall N (randomized/analyzed): 149 Loss to follow‐up excluded: 2 Gender (M/F): 94/55 COPD stage (GOLD): severe to very severe Included criteria Severe or very severe COPD diagnosed according to ATS/ERS criteria at admission(FEV1 < 50%pred Excluded criteria Previous pneumonectomy or lobectomy in the past 6 months;  Spontaneous risk of pneumothorax or rib fracture;  Incapacity to follow a standard rehabilitation program (locomotor deficits, acute cardiac failure and acute exacerbation of COPD at the beginning of the program); The absence of written informed consent. Pretreatment: 6 patients with lobectomy or pneumonectomy in the control group were included.
Interventions	Intervention characteristics PR+IMT PR: conducted over 4 weeks, 5 d/week and included aerobic exercise on a cycle ergometer and a treadmill (each for 30 min/day), strengthening of lower and upper limb muscle groups, a therapeutic, educational program, aerobic gymnastics in groups, a smoking cessation program and socio‐psychological and dietary advice. IMT: consisted of 2 sessions of 15 min each, supervised by a physiotherapist, 5 times/week, over 4 weeks. The patients had to breathe slowly with an increased tidal volume; after 10 inspirations, they could have a break by breathing at rest for a short time. The cycle of 10 inspirations was repeated 15 times. The inspiratory muscle training was performed using a threshold inspiratory muscle trainer (PowerBreathe Medic; PowerBreathe, Southam, UK) at a resistance of 50% of PImax. The intensity was increased (+10%) after 10 days of training to reach 60% of the initial PImax. PR: participants in this group received only the standardised PR program.
Outcomes	Dyspnea: Borg Dyspnea: mMRC Dyspnea: MDP: Unpleasantness Sensory intensity Muscle work/effort Not enough air/smother/air hunger Tight/constricted Mental effort/ concentrate Breathing a lot (rapid, deep, and heavy) Depression Anxiety Frustration Anger  Fear Functional exercise capacity: 6MWD HRQoL: SGRQ  Symptoms Activity Impact Total Respiratory muscle strength: PImax
Identification	Country: France Setting: Rehabilitation programme unit of Centre Hospitalier des Pays de Morlaix Author's name: Marc Beaumont Institution: Pulmonary Rehabilitation Unit, Morlaix HospitalCentre, European University of Occidental Brittany Email: marc.beaumont@univ‐brest.fr Address: 29672 Morlaix CEDEX, France Clinical trial register: NCT02074813
Notes	Borg and MDP scales were conducted at the end of 6MWD. Subgroup analysis according to PImax (> or ≤ 60 cmH2O) was conducted. See supplementary table.

Beckerman 2005.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 21/17 Loss to follow‐up or excluded: 4 Age, mean (SD), in years: 67.7 (16.49) Gender (M/F): 17/4 Control/sham N (randomized/analyzed): 21/14 Loss to follow‐up or excluded: 7 Age, mean (SD), in years: 66.9 (15.12) Gender (M/F): 15/6 Overall N (randomized/analyzed): 42/31 Loss to follow‐up or excluded: 11 Gender (M/F): 32/10 COPD stage: severe to very severe Included criteria Spirometric evidence of significant chronic airflow limitation (FEV1 < 50% of predicted, FEV1/FVC < 70% of predicted) with a diagnosis of COPD according to the criteria of the ATS were recruited from the community. The patients were all new to an IMT program, and none were receiving additional regular exercise or dietary supplements. Excluded criteria Patients with cardiac disease and poor compliance and needing supplemental oxygen
Interventions	Intervention characteristics IMT: participants trained daily in 2 sessions of 15 min each, 6 times/week for 12 months. The training was performed using a threshold inspiratory muscle trainer (POWERbreathe; Gaiam Ltd; Southam, Warwickshire, UK). The participants started breathing at a resistance that required generation of 15% of Pimax for 1 week. The load was then increased incrementally, 5% to 10% each session, to reach 60% of Pimax at the end of the first month. IMT was then continued at 60% of the Pimax adjusted monthly to the new Pimax achieved. The training was conducted in a rehabilitation center for 1 month under the supervision of a respiratory therapist followed by home training, verified by a respiratory therapist daily by phone and once weekly by a personal visit, for the next 11 months. Control/sham: this group trained with the same protocol at a load equal to 7 cm H2O.
Outcomes	Dyspnea: Borg Functional exercise capacity: 6MWD HRQoL: SGRQ Respiratory muscle strength: PImax
Identification	Country: Israel Setting: Home Author's name: Paltiel Weiner Institution: Department of Medicine A, Hillel Yaffe Medical Center Email: weiner@hillel‐yaffe.health.gov.il Address: Hadera, Israel 38100
Notes	The study reported data as mean and SE, we computed SD for the baseline characteristics. The SE reported in the text of the trial is less than shown in the graph.

Belman 1988.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/8 Age, mean (SD), in years: 64 (8.4) Gender (M/F): 4/4 Control/sham N (randomized/analyzed): not reported/9 Age, mean (SD), in years: 64 (9.0) Gender (M/F): 6/3 Overall N (randomized/analyzed): 20/17 Loss to follow‐up or excluded: 3 Gender (M/F): 10/7 COPD stage: moderate to severe Included criteria Presence of COPD as defined by the ATS, and improvement in FEV1 < 20% after inhaled isoproterenol Excluded criteria Evidence of coronary artery disease, cardiac arrhythmias, congestive heart failure, and orthopedic problems such as shoulder girdle and spinal abnormalities, which would interfere with the performance of the breathing manoeuvres
Interventions	Intervention characteristics IMT: the training protocol consisted of unsupervised sessions, 2 sessions/d, 7/week for 6 weeks. Each session lasted 15 min, using a Pflex device (the breathing pattern was controlled), and the training load was as tolerated. Control/sham: the control group received similar training at a load of around 7.5‐10 cmH2O.
Outcomes	Respiratory muscle strength: PImax (FRC) Respiratory muscle endurance:  MVV Pthmax Respiratory function: FEV1 (L)
Identification	Country: USA Setting: hospital Author's name: Michael J Belman Institution: Division of Pulmonary Medicine, Cedars‐Sinai Medical Center, and The University of California Address: Los Angeles, California 90048
Notes

Berry 1996.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 8/7 Loss to follow‐up or excluded: 1 Age, mean (SD), in years: 67.0 (3.39) Gender (M/F): 4/4 PR N (randomized/analyzed): 9/9 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 70.8 (4.8) Gender (M/F): 6/3 Overall N (randomized/analyzed): 17/16 Loss to follow‐up or excluded: 1 Gender (M/F): 10/7 COPD stage (GOLD): moderate to severe Included criteria FEV1/FVC < 0.65; Dyspnea on exertion experienced during activities of daily living; A history of cigarette or tobacco smoke exposure > 20 pack‐years; The ability to self ambulate; Age > 60 years; Willingness to undergo testing and intervention procedures; COPD under appropriate medical management. Excluded criteria Significant cardiac disease; Orthopedic or neurologic impairment, serious renal, liver, or gastrointestinal disorders; current psychiatric illness or substance dependence; uncontrolled diabetes or hypertension; Current or previous (within 6 months) enrollment in a rehabilitation or exercise program;   SaoO2 < 90% during exercise at a heart rate > 50% of age‐predicted maximum.
Interventions	Intervention characteristics PR+IMT PR: the rehabilitation program involved walking, upper extremity strength training, and progressive IMT. Walking intensity was set at 50%‐75% of the participant's heart rate reserve. This prescription was based on the heart rate response from the participant's initial graded exercise test. The duration of walking was increased progressively throughout the intervention to a maximum of 20 min. Participants were taught to monitor their heart rate during walking by palpating their radial artery and to adjust their exercise intensity as needed. Heart rates were taken midway through and at the end of the walking sessions. Upper extremity weight‐training consisted of 5 different exercises. Participants performed 2 sets of 12 repetitions for each exercise with the weight being progressively increased as their strength increased. All walking and strength training was performed at the exercise center. IMT: was performed twice daily for a 15‐min period, 7 d/week, for 12 weeks, using threshold IMT device (Healthscan Products, Cedar Grove, NJ). The initial pressure setting was set at 15% of the participant's PImax for 2 weeks. During Weeks 3 and 4, the threshold pressure was increased to 30% of the participant's PImax. During weeks 5 and 6, the threshold pressure was increased to 60% of the participant's PImax. During Weeks 7 through 12, the threshold pressure was set at 80% of the participants PImax. PR + sham IMT: this group underwent PR as described above, and IMT at 15% of PImax for the duration of the 12‐week study period.
Outcomes	Dyspnea: Borg  Functional exercise capacity:  exercise time (treadmill) 12MWD Respiratory muscle strength: PImax Laboratory exercise test: VO2peak (mL/kg/min) Respiratory muscle endurance: MVV Respiratory function:  FEV1 (%pred) FEV1 (L)
Identification	Sponsorship source: Country: USA Setting: Wake Forest University Author's name: MichaelJ.Berry Institution: Department of Health and Sport Science Address: P.O. Box 7234, Wake ForestUniversity, Winston‐Salem, NC 27109
Notes	All values are adjusted means ± SEM. Values were adjusted using pre‐intervention scores as the covariate.

Berton 2015.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): /7 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 65.3 Gender (M/F): 3 BMI, mean (SD), kg/m²: 25.8 Control/sham N (randomized/analyzed): /6 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 66.3 (9.2) Gender (M/F): 1/5 BMI, mean (SD), kg/m²: 24.5 (5.0) Overall N (randomized/analyzed): 24/13 Loss to follow‐up or excluded: 9 Age, mean (SD) in years: 66.8 (9.1) Gender (M/F): 4/9 COPD stage: moderate to very severe Included criteria Patients with spirometric evidence of significant chronic airflow limitation (FEV1 < 70%pred, FEV1/FVC < 0.7) according to the criteria of Global Strategy for Diagnosis, Management, and Prevention of COPD; Patients with a long history of smoking (> 20 pack‐years) were invited to participate from the tertiary clinic care center. Participants were receiving continuously formoterol/budesonide (12/400 μg) twice a day (dry powder inhaler), short‐acting bronchodilators as rescue medications, and had not participated in PR in the previous 24 months. Excluded criteria Exacerbation of COPD in the previous 3 months or during the study; Cardiac disease (acute coronary syndrome in the previous 3 months or cardiac ejection fraction < 50%); long term oxygen therapy or arterial oxygen saturation < 90% at rest, Neuromuscular disease, peripheral arterial disease, cancer; Physically unable to move.
Interventions	Intervention characteristics IMT: the training was unsupervised, 30 min/d, 7 d/week, for 8 weeks. The device used was Powerbreathe(Southam, UK) set at 30% of PImax. Control/sham: this group underwent similar training at no load.
Outcomes	Dyspnea: Borg Functional exercise capacity: exercise time HRQoL: SGRQ Symptoms Activity Impact Total Inspiratory muscle strength (PImax)
Identification	Country: Brazil Author's name: Danilo C. Berton Institution: Graduation Program in Pulmonology, Federal University of Rio Grande do Sul (UFRGS), Brazil Email: dberton@hcpa.edu.br Address: Rua Ramiro Barcelos, 2350, Room 2050. Postal Code: 90035‐003, Porto Alegre, RS, Brazil Clinical trial register: NCT 01945398
Notes	Participants were instructed to maintain diaphragmatic breathing, with a breathing rate of 15‐20 breaths/min.

Bustamante 2007.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 12/12 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 62 (13.7) BMI, mean (SD), kg/m²: 26.03 (3.46) Control/sham N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 61.5 (8.6) BMI, mean (SD), kg/m²: 26.9 (4.41) Overall N (randomized/analyzed): 22/22 Loss to follow‐up excluded: 0 COPD stage (GOLD): moderate to severe Included criteria Patients with a clinically stable COPD for a month before selection. Excluded criteria Hypoxemia (PaO2 < 60 mmHg), asthma, coronary heart disease, metabolic diseases, orthopedic diseases, history of recent thoracic or abdominal surgery, treatment with corticosteroid, hormones and chemotherapy.
Interventions	Intervention characteristics IMT: the training was unsupervised, conducted twice a day, 15 min/session, 7 d/week, for 6 weeks. Threshold IMT device was used at the maximum tolerated load. Control/sham: this group underwent a similar training at a load equal to 7cmH20.
Outcomes	HRQoL: CRQ Dyspnea Fatigue Emotion Mastery Respiratory muscle strength: PImax Respiratory muscle endurance time: Tlim (Threshold device) Notes: sustained time with a threshold of 66% of PImax
Identification	Country: Spain Author's name: Víctor Bustamante Madariaga Institution: Servicio de Neumología. Hospital de Basurto. Osakidetza. Vizcaya. España Email: VICTOR.BUSTAMANTEMADARIAGA@hbas.osakidetza.net
Notes

Charususin 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 110/89 Loss to follow‐up or excluded: 26 Age, mean (SD) in years: 66 (8) Gender (M/F): 52/58 BMI, mean (SD), kg/m² : 25 (6) PR N (randomized/analyzed): 109/85 Loss to follow‐up or excluded: 29 Age, mean (SD) in years: 65 (7) Gender (M/F): 43/66 BMI, mean (SD), kg/m² : 24 (6) Overall N (randomized/analyzed): 219/174 Loss to follow‐up or excluded: 55 Gender (M/F): 95/124 COPD stage (GOLD): moderate to very severe Included criteria Clinically stable COPD patients with reduced maximal inspiratory mouth pressure (PImax < 60 cm H2O or < 50% of the predicted normal value) participated in the study between February 2012 and October 2016. Excluded criteria Diagnosed psychiatric or cognitive disorders Progressive neurological or neuromuscular disorders Severe orthopaedic problems having a major impact on daily activities Previous inclusion in a rehabilitation program (< 1 year) Pretreatment: "One of the centres offering a 36 session program (32% of total inclusions) consistently exceeded between‐group differences in the centre offering 20 sessions (36% of total inclusions). In the other centres offering 36 sessions (combined 32% of total inclusions), between‐group differences in these outcomes were consistently smaller than in the centre offering a lower training volume (see online supplementary table E4)"
Interventions	Intervention characteristics PR+IMT PR: consisted of cycling, treadmill walking, stair climbing, arm ergometry and resistance training of arm as well as leg muscles training volume, and ranged from 20 sessions (Germany) to 36 sessions (other centers). Training frequency ranged from 3‐5 sessions/week. Duration of training sessions was around 60 min. Participants performed endurance training or interval training at moderate to high intensities. During PR, training intensities were progressively increased according to a Borg CR‐10 scale ratings of 4–6 on dyspnea sensation. IMT: performed daily using the PowerBreathe KHP2 device (POWERbreatheKHP2, HaB International, Southam, UK) according to previously described methods: 2‐3 sessions/d, 7 min each, 7 d/week, for 12 weeks. Training intensity in the intervention group was set initially at a load of approximately 50% of participants’ maximal inspiratory mouth pressure (PImax). This initial load was then continuously and gradually increased to the highest tolerable intensity during each of the supervised sessions. PR +sham IMT: this group underwent the same training as described above with IMT load at 10% of PImax. The load was not modified throughout the intervention period.
Outcomes	Dyspnea: Borg post‐6MWD Incremental cycle ergometer test Constant cycle ergometer test Functional exercise capacity:  6MWD Wmax  Exercise time HRQoL: CRQ Dyspnea Fatigue Emotion Mastery Total Respiratory muscle strength: PImax Laboratory exercise test: VO2peak Respiratory muscle endurance: MVV Respiratory muscle endurance time: Tlim Respiratory function:  FEV1 (L) RV
Identification	Sponsorship source: DL and HD are postdoctoral fellows of Research Foundation Flanders. HaB International (Southam, UK) and McRoberts (The Hague, The Netherlands) provided equipment for testing and training in this study on loan. This study was further supported by local funds throughout the participating centers. The following specific funding sources were reported: University Hospital Leuven, Belgium (FWO grant GOA4516N en KU Leuven grant C22/15/035); Ghent University Hospital, Belgium (UZ Gent grant FS/LGZ/994); Institut Universitaire de Cardiologie et de Pneumologie de Québec, Université Laval, Quebec, Canada (Ordre professionnel de la physiothérapie du Québec). Country: Belgium, The Netherlands, Germany, Canada Setting: multicenter Author's name: Daniel Langer Institution: Department of Rehabilitation Sciences, KU Leuven Email: daniel. langer@ kuleuven. be Address: Leuven 3001, Belgium Clinical trial register: NCT01397396
Notes	Adjusted difference (95% CI) at post‐training with its P value were reported.

Chuang 2017.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 30/27 Loss to follow‐up or excluded: 3 Age, mean (SD), in years: 66.22 (12.76) Gender (M/F): 17/10 Control/sham N (randomized/analyzed): 30/28 Loss to follow‐up or excluded: 2 Age, mean (SD), in years: 66.04 (10.99) Gender (M/F): 19/9 Overall N (randomized/analyzed): 60/55 Loss to follow‐up or excluded: 5 Gender (M/F): 36/19 COPD stage (GOLD): moderate to very severe Included criteria Outpatients with a stable condition Showed airflow limitation on pulmonary function test, that is FEV1 < 80% and FEV1/FVC < 70% Participants were conscious and able to express themselves orally, were self‐walkers and co‐operated with the intervention. Excluded criteria Cardiovascular diseases Severe unstable diseases, such as pulmonary heart disease or cancer Ongoing oxygen treatment Muscle power rating < 5 ADL scale score < 80 Severe mental illness
Interventions	Intervention characteristics IMT: the training was conducted 5 d/week, for 8 weeks. Each session lasted for 21–30 min. It consisted of a cycle of 2 min inspiratory training with a pressure threshold loading device and 1 min of rest, and then repeated 7 cycles. The training load ranged from 15 cmH20 to 40 cmH2O. Each participant was followed up once a day by phone from Monday–Friday, and visits were made by research assistants every 2 weeks to increase the pressure threshold loading progressively after assessing the participant’s condition. Control: no intervention was received by this group.
Outcomes	Dyspnea: BDI Functional exercise capacity: 6MWD HRQoL: SF‐36 Questionnaire Physiological functions Mental functions Respiratory muscle strength: PImax
Identification	Country: Taiwan Author's name: Hsiao‐Yun Chang Institution: Department of Nursing, Fooyin University, Kaohsiung City Email: chang369@gmail.com Address: Taiwan
Notes

Covey 2001.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	IMT N (randomized/analyzed): 19/12 Loss to follow‐up or excluded: 10 Age mean (SD) in years: 65 (6) BMI, mean (SD), kg/m²: 26 (4.8) Control/sham N (randomized/analyzed): 18/15 Loss to follow‐up or excluded: 3 Age mean (SD) in years: 67 (10) BMI, mean (SD), kg/m²: 28.8 (6) Overall N (randomized/analyzed): 37/27 Loss to follow‐up or excluded: 10 Gender (M/F): 18/9 COPD stage: severe to very severe Included criteria Severe to very severe airflow obstruction (FEV1 < 50%pred) Age: 45‐75 Severely limited functional performance Excluded criteria  History of asthma or > 25% increase in FEV1 after bronchodilator History of a major exacerbation within the past 2 months Current oral corticosteroid use (> 10 mg prednisone per day) Other health problems that would inhibit their ability to participate
Interventions	Intervention characteristics IMT: participants performed IMT at home 5 d/week, 30 min/d for 16 weeks (for a total of 80 training sessions) using Threshold IMT device. Starting training loads were 30% of PImax as tolerated. Participants were visited weekly at home by a nurse who supervised training, evaluated training loads, and progressively increased the training load as tolerated with a goal of achieving 60% of PImax. Home visits generally lasted approximately 30 min. An interval training protocol was used with participants performing 6 work sets of 5 mins’ duration separated by rest intervals lasting 1‐3 min. Control/sham: participants were visited by a nurse every 2 weeks for 16 weeks for a structured program of health education. Each home‐based session lasted approximately 1‐1.5 h and covered such topics as nutrition, relaxation techniques, pursed lip breathing, respiratory medications, respiratory infection, energy conservation techniques, oxygen therapy, and smoking cessation.
Outcomes	Dyspnea: Borg HRQoL: CRQ Respiratory muscle strength (PImax) Respiratory muscle endurance pressure: Pthmax
Identification	Sponsorship: This study was supported by a grant from the National Institutes of Nursing Research, grant number NRO1428; and was conducted at the University of Illinois at Chicago, College of Nursing, Chicago, Ill and Hines VA Hospital, Section of Critical Care and Pulmonary Medicine, Hines, Ill. Country: USA Setting: home‐based Author's name: Margaret K. Covey Institution: University of Illinois at Chicago, College of Nursing Email: mkcovey@uic.edu Address: 845 South Damen Avenue, Chicago, IL 60612
Notes

Cutrim 2019.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 11/11 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 66 (8.5) Gender (M/F): 8/3 BMI, mean (SD), kg/m² : 25 (4) Control/sham N (randomized/analyzed): 11/11 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 70 (8.0) Gender (M/F): 9/2 BMI, mean (SD), kg/m² : 25 (4) Overall N (randomized/analyzed): 22/22 Gender (M/F): 17/5 Included criteria Patients with COPD without heart failure or pulmonary hypertension With inspiratory muscle weakness (PImax < 70% of predicted) and with stable pharmacological treatment, i.e. no drug change at least 1 month before the start of IMT Excluded criteria Functional limiting factors that would interfere with the performance of IMT and/or the exercise capacity test, such as acute myocardial infarction 3 months before inclusion in the study, unstable angina or unstable ventricular arrhythmia or in the last 3 months prior to initiation, acute respiratory disease, rheumatic diseases, degenerative diseases, neurological sequelae, cognitive deficit etc
Interventions	Intervention characteristics IMT: the session of IMT consisted of 30 min (in a clinical setting) 3 times/week, using the Threshold Inspiratory Muscle Training device (POWERbreathe Medic+Plus, NCS, Barueri, SP, Brazil). The inspiratory load was set at 30% of PImax, for 12 weeks. During exercise, participants were instructed to maintain diaphragmatic breathing at a rate of 15–20 breaths/min. Control/sham: no intervention received by this group (except diaphragmatic breathing).
Outcomes	Functional exercise capacity: 6MWD Respiratory muscle strength: PImax
Identification	Sponsorship source: Fundação de Amparo à Pesquisa do Estado do Maranhão (FAPEMA); Hospital Universitário Presidente Dutra; Cristiano Mostarda received grants from CNPq (Universal 442374/2014‐3) and FAPEMA (Bolsa Produtividade and Universal 00358/15). Country: Brazil Setting: Hospital Universitário Presidente Dutra Author's name: Cristiano Teixeira Mostardaa Institution: Universidade Federal do Maranhão, São Luís, Brazil Email: cristiano.mostarda@gmail.com Address: Av. dos Portugueses, 1966, Cidade Universitária Dom Delgado, São Luís, MA, Brazil Clinical trial register: RBR‐4mz6w9
Notes	Participants were instructed to maintain diaphragmatic breathing at a rate of 15–20 breaths/min. Adjusted analysis for age, weight and baseline were reported.

Dacha 2019.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/6 N (analyzed): 6 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 65 (4) Gender (M/F): 3/3 BMI, mean (SD), kg/m² : 29 (8) COPD stage (GOLD): mild to very severe Control/sham N (randomized/analyzed): not reported/4 N (analyzed): 4 Loss to follow‐up excluded: 0 Age, mean (SD), in years: 68 (9) Gender (M/F): 1/3 BMI, mean (SD), kg/m²: 22 (2) COPD stage (GOLD): moderate to very severe Overall N (randomized/analyzed): 24/10 N (analyzed): 10 Loss to follow‐up excluded: 0 Gender (M/F): 4/6 COPD stage (GOLD): mild to very severe Included criteria Clinical Diagnosis of COPD Inspiratory muscle weakness (PImax < 70%pred or < 60 cmH2O) BDI < 7  Peripheral muscle fatigue present after CPET Excluded criteria Major cardiovascular limiting exercise capacity more than pulmonary function impairment Severe orthopedic with a major impact on ADL Psychiatric or cognitive disorders Progressive neurological or neuromuscular disorders Long term O2 therapy Previous inclusion in a rehabilitation program (< 1 year)
Interventions	Intervention characteristics IMT: the training was conducted twice a day, 7 d/week, for 8 weeks. All the sessions were supervised, 5 min each, using Powerbreathe device set at a 50% of PImax. Control/sham: participants in this group received a similar protocol with a training load at 10% of PImax.
Outcomes	Dyspnea: Borg Respiratory muscle strength: PImax Laboratory exercise test: VO2peak
Identification	Sponsorship source: KU Leuven Country: Belgium Author's name: Sauwaluk Dacha Institution: Faculty of Kinesiology and Rehabilitation Sciences, Department of Rehabilitation Sciences, Research Group for Rehabilitation in Internal Disorders Email: sauwaluk.dacha@kuleuven.be Address: KU Leuven, Leuven, Belgium Clinical trial register: NCT03240640
Notes	This study is still ongoing.

De Farias 2019.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Overall N (randomized/analyzed): 33/31 Loss to follow‐up or excluded: 2  Age, mean (SD) in years:  66.2 (4.9) BMI: 28 (4.3) Inclusion criteria Individuals with a clinical diagnosis of COPD according to GOLD Being treated at the Ambulatory Pulmonology department of University Hospital Onofre Lopes (HUOL)/Empresa Brasileira de Serviços Hospitalares (EBSERH) Aged 40‐80 years Living in the city of Natal, RN/Brazil Not using oxygen therapy or presenting disease exacerbation in the last 3 months Not practising regular physical activity in the last 6 months Exclusion criteria Musculoskeletal comorbidities that impair gait SpO2 <  90% during 6MWD Hypertensive without control medication as well as those presenting with a hypertensive peak (> 140/90 mmHg) for > 3 consecutive days An intellectual understanding impairment that interferes with the evaluation tests Those who stop the therapeutic program, miss activity for > 1 week, or miss reevaluation
Interventions	Intervention characteristics PR+ sham IMT PR: performed 3 times/week for 10 weeks. All individuals were instructed to perform the rehabilitation program 5 days/week — 3 days with supervision and 2 without supervision — for at least 1 h/d. The program consisted of health education, treadmill aerobic training with 70% of max incremental shuttle walk test speed and peripheral muscle strength training. Sham IMT: performed with Powerbreathe device at no load for 3 cycles of 12 repetitions. PR+ threshold IMT PR: the same protocol as described above Threshold IMT: participants trained using POWERbreathe KH1 (POWERbreathe International Ltd.) at 35% of MIP), increasing 5% every week until reaching 80% of MIP at the 10th week, which was maintained until the end of the protocol. The MIP was assessed weekly to adjust the training load percentage. Participants performed 3 cycles of 12 repetitions. PR + isocapnic hyperpnea PR: the same protocol as described above. Isocapnic hyperpnea: participants received training with a duration of up to 20 min, 1 min of training and 1 min of rest, where they were encouraged with instructions such as “breathe faster.” The training respiratory rate chosen was calculated as 35‐fold FEV1, so that ventilation corresponded to 50%–60% of MVV. The device used was STMedical device (SpiroTiger, Chamonix Mont Blanc, France).
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD Respiratory muscle strength: PImax
Identification	Sponsorship source: no funding Country:  Brazil Author's name: Guilherme Augusto de Freitas Institution: PneumoCardioVascular Lab/HUOL, Empresa Brasileira de Serviços Hospitalares ‐ EBSERH), Universidade Federal do Rio Grande do Norte (UFRN), Email: fregonezi.guilherme@gmail.com Address: Natal, Rio Grande do Norte, Brazil
Notes

Dekhuijzen 1991.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 58/8 Gender (M/F): 14/6 PR N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 60/7 Gender (M/F): 16/4 Overall N (randomized/analyzed): 40/40 Loss to follow‐up or excluded: 0 Gender (M/F): 30/10 COPD stage (GOLD): moderate to severe Included criteria: patients with functional limitations due to COPD Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT PR: consisted of exercise training, i.e. cycling, walking, and training of back, shoulder, and abdominal muscles. The intensity of the exercise training was determined by the symptoms of the patients. Moreover, the heart rate during these exercises did not exceed 80% of the maximal heart rate reached during the maximal bicycle ergometer test. Other parts of the PR program were callisthenics, conventional physiotherapy (breathing retraining, relaxation exercises), and education about pulmonary disease and the purpose and use of the medications. The training was conducted 2 h every day, 5 d/week, for 10 weeks. IMT: was conducted along with PR, 15 min twice a day, supervised by the physiotherapist, using incentive spirometer at 70% of PImax. PR: this group received only the PR described above.
Outcomes	HRQoL: ADL Functional exercise capacity:  12MWD Wmax (incremental cycle ergometer test) Respiratory muscle strength: PImax Laboratory exercise test: VO2peak Respiratory muscle endurance time: Tlim
Identification	Country: The Netherlands Setting: Medical Centre Dekkerswald (outpatient clinic) Author's name: P. N. Richard Dekhuijzen Institution: the University of Nijmegen, Department of Pulmonary Diseases, Medical Centre Dekkerswald Address: Groesheek, the Netherlands
Notes	We used convertunits.com to convert from Kpa to cmH2O.

Dellweg 2017.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 15/15 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 66 (8) Gender (M/F): 7/8 BMI, mean (SD), kg/m²: 24.5 (6.4) PR N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Age, mean (SD), in years: 66 (7.5) Gender (M/F): 9/5 BMI, mean (SD), kg/m²: 23.9 (4) Overall N (randomized/analyzed): 29/29 Loss to follow‐up or excluded: 0 Gender (M/F): 16/13 COPD stage (GOLD): severe to very severe Included criteria Patients with persistent hypercapnic respiratory failure who required non‐invasive ventilation after prolonged weaning according to the criteria of Boles et al (Boles 2007) Main diagnosis of COPD Participants had to be admitted to our in‐patient post‐weaning rehabilitation unit, had to be ambulatory, co‐operative, and physically able to participate in twice‐daily physiotherapeutic sessions Patients had to have confirmed COPD stage 3 or 4 by a lung function test (FeV1 < 50%, Fev1/FVC < 70%) Had to be free of exacerbation Excluded criteria Renal impairment (serum creatinine levels < 2 mg/dL) Severe cardiac impairment (ejection fraction < 40%)
Interventions	Intervention characteristics PR+IMT: PR: consisted of twice daily, hour‐long physiotherapy group sessions in the rehabilitation gym. Physiotherapy sessions included training with arm and leg ergometers, as well as weight training. Additionally, every patient participated in daily ergotherapy group sessions lasting 1 h each to improve fine motor skills. IMT: consisted of supervised sessions, once daily during weekdays, for 4 weeks. Participants underwent strength training at 80% of PImax and endurance training at 60% of PImax using Respifit S Trainer (Biegler, Mauerbach, Austria) PR+ (sham IMT): participants in this group received a similar protocol as described above with IMT training load fixed at 5 cm H2O, and using Threshold IMT (Philips‐Respironics, Pittsburgh, PA, USA).
Outcomes	Functional exercise capacity: 6MWD Respiratory muscle strength: PImax (RV) Respiratory function: FEV1
Identification	Sponsorship source: no funding Country: Germany Author's name: Dominic Dellweg Institution: Department for Pulmonology, Intensive Care and Rehabilitation Email: d.dellweg@fkkg.de Address: 1, 57392 Schmallenberg, Germany Clinical trial register: NCT00291460
Notes

Fanfa Bordin 2020.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized or analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 63.2 (5.7) Gender (M/F): 5/5 BMI, mean (SD), kg/m² : 25.3 (5.1) PR N (randomized/analyzed): 12/10 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 66.2 (9.2) Gender (M/F): 4/6 BMI, mean (SD), kg/m² : 25.4 (5.8) Overall N (randomized/analyzed): 22/20 Loss to follow‐up or excluded: 2 Gender (M/F): 9/11 COPD stage (GOLD): moderate to very severe Included criteria COPD patients at the stages 2‐4, according to the GOLD classification Participating in a lung rehabilitation program for at least 2 months Clinically stable Signed the informed consent Excluded criteria Individuals with asthma, and/or cardiovascular disease and individuals with cognitive and/or behavioral impairments
Interventions	Intervention characteristics PR+IMT PR: the lung rehabilitation program included 30 min of cycling exercise on a vertical cycle ergometer for lower limbs (Movement, BM 2700, Brazil) set at 60% of the maximal heart rate determined using the Karvonen method modified for the reserve heart rate (Meyer 2013). Participants also performed strengthening exercises for the upper and lower limbs' major muscles, with the intensity of 50%‐80% of the 1 repetition maximum test (1RM). IMT: involved supervised sessions, 3 times/week for 20 min over 4 weeks, which was increased to 25 min on weeks 5 and 6, and to 30 min on weeks 7 and 8. Participants used Threshold IMT set at 50% of PImax. PR: this group received only the PR described above.
Outcomes	Respiratory muscle strength: PImax
Identification	Country: Brazil Setting: Santa Cruz do Sul Hospital, Brazil Author's name: Diogo Fanfa Bordin Institution: Universidade de Santa Cruz do Sul (Unisc) – Santa Cruz do Sul (RS), Brazil Email: diogo.fanfa@hotmail.com Clinical trial register: NCT02014155
Notes

Harver 1989.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 61.1 (9.5) Gender (M/F): 8/2 Control/sham N (randomized/analyzed): not reported/9 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 64.8 (8.4) Gender (M/F): 8/1 Overall N (randomized/analyzed): 23/19 Loss to follow‐up or excluded: 4 Gender (M/F): 16/3 COPD stage (GOLD): moderate to severe Included criteria Participants for the study were recruited from the clinical practice of pulmonary physicians at the institution Participants were stable both clinically and functionally, and medications were not changed during the study. Excluded criteria: not reported
Interventions	Intervention characteristics IMT: participants trained twice a day, 15 min each session, 7 d/week, for 8 weeks. The training device was Pflex (with a controlled breathing pattern), and the training load ranged from 5 cmH2O to 35 cmH2O (≈30% PImax FRC) Control/sham: this group received a similar IMT protocol with a training load set at 5cmH2O
Outcomes	Dyspnea: BDI‐TDI:  Functional impairment  Magnitude of task  Magnitude of effort Focal score Respiratory muscle strength: PImax  RV FRC Respiratory function: FEV1 L Respiratory muscle endurance: MVV
Identification	Sponsorship source: The American Lung Association of New Hampshire and by grants HL07449 and HL29068 from the National Heart, Lung, and Blood Institute Country: USA Setting: Outpatient pulmonary clinic and pulmonary function laboratory Author's name: Andrew Harver Institution: Department of Psychology, SUNY Stony Brook Address: Stony Brook, NY 11794
Notes

Heijdra 1996.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 62.4 (8.8) Gender (M/F): 7/3 BMI, mean (SD), kg/m²: 23.7 (3.3) Control/sham N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years:  61.8 (7.3) Gender (M/F): 8/2 BMI, mean (SD), kg/m²:  23 (3.2) Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Gender (M/F): 15/5 COPD stage (GOLD): moderate to very severe Included criteria Mean nocturnal arterial oxygen saturation was < 92% All participants were in a stable condition at the time of study as defined by a fluctuation in FEV1 of < 10% in the preceding 6 months Excluded criteria Patients with other pulmonary diseases, chest wall deformations, a previous thoracotomy, diabetes mellitus, neuromuscular diseases, obstructive sleep apnea syndrome, or an overlap syndrome
Interventions	Intervention characteristics IMT: participants trained at home, daily, 2 sessions of 15 min/day for 10 weeks. They used incentive flowmeter (INSPIRx; Resprecare Medical Inc., the Hague, the Netherlands) set at 60% of PImax Control/sham: participants received a similar training protocol and the resistance was set at 10% of PImax
Outcomes	Inspiratory muscle strength (PImax) Respiratory muscle endurance time (Tlim)
Identification	Sponsorship source: Dutch Asthma Foundation (No. 90‐27) Country: The Netherlands Author's name: Yvonne F. Heijdra, Department of Pulmonary Diseases  Institution: University of Nijmegen, Medical Centre Dekkerswald Address: P.O. Box 9001, 6560 GB Groesbeek, the Netherlands
Notes

Hill 2006.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 18/16 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 69.4 (7.2) Gender (M/F): 11/5 BMI, mean (SD), kg/m²: 24.9 (4.3) Control/sham N (randomized/analyzed): 17/17 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 66.6 (9.8) Gender (M/F): 11/6 BMI, mean (SD), kg/m²: 24.1 (3.7) Overall N (randomized/analyzed): 35/33 Loss to follow‐up or excluded: 2 Gender (M/F): 22/11 COPD stage (GOLD): severe to very severe Included criteria Participants who had a diagnosis of COPD Smoking history of 10 pack‐years FEV1 ranging 15%–70% of predicted normal Excluded criteria Comorbid conditions likely to reduce exercise capacity (e.g. symptomatic ischemic heart disease, BMI > 35 kg.m²) Previous lung surgery The use of long‐term oxygen therapy Weaning doses of oral corticosteroids
Interventions	Intervention characteristics IMT: participants attended supervised training sessions 3 times/week for 8 weeks. Each session lasted 21 min and comprised 7 cycles of 2 min of breathing on an inspiratory threshold loading device (Threshold IMT; Respironics, Cedar Grove, NJ, USA) followed by 1 minute of rest. The training load was set at a range of around 45%‐101% of PImax Control/sham: this group received a similar IMT protocol with a training load set at 10% of PImax
Outcomes	Dyspnea: Borg Post 6MWD Wmax Functional exercise capacity:  6MWD Wmax  HRQoL: CRQ  Dyspnea Fatigue Emotion Control Total Respiratory muscle strength (PImax) Respiratory muscle endurance: Respiratory muscle endurance pressure (Pthmax) Laboratory exercise test: VO2peak (mL/kg/min) Respiratory function: FEV1 %pred L Respiratory function: RV %pred L
Identification	Sponsorship source: The National Health and Medica Research Council (Canberra, Australia) grant number 212016 Country: Australia Author's name: P.R. Eastwood Institution: Dept of Pulmonary Physiology, Sir Charles Gairdner Hospital Email: peter.eastwood@health.wa.gov.au Address: Hospital Avenue Nedlands Western Australia, Australia 6009
Notes

Hill 2007.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 18/16 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 69.4 (7.2) Gender (M/F): 11/5 BMI, mean (SD), kg/m²: 24.9 (4.3) Control/sham N (randomized/analyzed): 17/17 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 66.6 (9.8) Gender (M/F): 11/6 BMI, mean (SD), kg/m²: 24.1 (3.7) Overall N (randomized/analyzed): 35/33 Loss to follow‐up or excluded: 2 Gender (M/F): 22/11 COPD stage (GOLD): severe to very severe Included criteria Participants who had a diagnosis of COPD Smoking history of 10 pack‐years  FEV1 ranging 15%–70% of predicted normal Excluded criteria Comorbid conditions likely to reduce exercise capacity (e.g. symptomatic ischemic heart disease, BMI > 35 kg.m²) Previous lung surgery The use of long‐term oxygen therapy Weaning doses of oral corticosteroids
Interventions	Intervention characteristics IMT: participants attended supervised training sessions 3 times/week for 8 weeks. Each session lasted 21 min and comprised 7 cycles of 2 min of breathing on an inspiratory threshold loading device (Threshold IMT; Respironics, Cedar Grove, NJ, USA) followed by 1 minute of rest. The training load was set at a range of around 45%‐101% of PImax Control/sham: this group received a similar IMT protocol with a training load set at 10% of PImax
Outcomes	Respiratory muscle endurance time: Tlim Respiratory muscle endurance pressure (Pthmax) Note: the same as Hill 2006
Identification	Sponsorship source: The National Health and Medica Research Council (Canberra, Australia) grant number 212016 Country: Australia Author's name: P.R. Eastwood Institution: Dept of Pulmonary Physiology, Sir Charles Gairdner Hospital Email: peter.eastwood@health.wa.gov.au Address: Hospital Avenue Nedlands Western Australia, Australia 6009
Notes

Hsiao 2003.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT (Threshold device) N (randomized/analyzed): not reported/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 68.2 (6.5) Gender (M/F): 10/0 IMT (Incentive spirometer) N (randomized/analyzed): not reported/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 70.4 (5.3) Gender (M/F): 8/2 Control/sham N (randomized/analyzed): not reported/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 71.1 (3.9) Gender (M/F): 8/2 Overall N (randomized/analyzed): 42/30 Loss to follow‐up or excluded: 12 Gender (M/F): 26/4 COPD stage (GOLD): moderate to severe Included criteria Patient with moderate to severe COPD FEV1 < 2L and FEV1/FVC <  60% Excluded criteria Restrictive lung disease History of cardiovascular disease or musculoskeletal conditions that could interfere with the training or the testing maneuvers
Interventions	Intervention characteristics IMT: participants trained twice a day, 15 min/session, 5days/week for 8 weeks. The sessions were unsupervised. Participants were divided into two groups using either Threshold IMT or Respirex, which were set at 50% of PImax Control/sham: this group did not receive any intervention
Outcomes	Functional exercise capacity: 6MWD Respiratory muscle strength: PImax (RV) Respiratory muscle endurance time: Tlim (Threshold device) Notes: measured while breathing against 70% of PImax
Identification	Country: Taiwan Setting: Pulmonary clinic of 1 university hospital Author's name: Ying Tai Wu Institution: School and graduate institute of physical therapy, college of medicine, National Taiwan University Hospital Address: 7 Chung Shan South Road, Taipei 100, Taiwan
Notes

Kim 1993.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/41 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 66 (7) BMI, mean (SD), kg/m²: 25 (4) Control/sham N (randomized/analyzed): not reported/26 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 63 (8) BMI, mean (SD), kg/m²: 25 (4) Overall N (randomized/analyzed): 129/67 Loss to follow‐up or excluded: 62 Gender (M/F): 51/16 BMI, mean (SD), kg/m²: COPD stage (GOLD): severe to very severe Included criteria No history of asthma or restrictive lung disease Free of respiratory tract infection for at least 2 months prior to enrollment and free of other chronic health problems that would interfere with their ability to participate in the study Excluded criteria > 10 mg of prednisone daily
Interventions	Intervention characteristics IMT: participants trained 7 days/week, 30 min/session, for 6 months. Training sessions were unsupervised, and the device used was a Threshold IMT device set at 30% of PImax Control/sham: this group received IMT at no load
Outcomes	Functional exercise capacity: 12MWD Respiratory muscle endurance time: Tlim Notes: measured while breathing at 66% of PImax. Respiratory muscle strength: PImax (RV)
Identification	Sponsorship source: supported by a grant from the National Center for Nursing Research, grant number NRO1428 Country: USA Author's name: Mija Kim Institution: College of Nursing, University of Illinois at Chicago
Notes

Koppers 2006.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group Subgroup analysis: PImax < 75%pred
Participants	IMT N (randomized/analyzed): 19/18 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 54.4 (7.7) Gender (M/F): 8/10 BMI, mean (SD), kg/m²: 26.7 (5.0) Control/sham N (randomized/analyzed): 20/18 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 57.0 (8.5) Gender (M/F): 9/9 BMI, mean (SD), kg/m²: 27.5 (3.3) Overall N (randomized/analyzed): 39/36 Loss to follow‐up or excluded: 3 Range age (min, max): 38,73 Gender (M/F): 17/19 COPD stage (GOLD): moderate to severe Included criteria Chronic airflow obstruction, defined as an FEV1/FVC ratio 70%, FEV1of 30%‐80%pred, after bronchodilation Stable clinical condition for at least 6 weeks Excluded criteria Hypoxemia at rest or during exercise Cardiac orthopaedic disease BMI > 30 kg/m2
Interventions	Intervention characteristics IMT: respiratory muscle endurance training was performed by means of tube breathing. A tube (internal diameter, 3 cm) connected to a mouthpiece was added to the respiratory system to rebreathe exhaled carbon dioxide (normocapnic hyperpnea). The maximum ventilatory capacity that can be sustained for 15 min is approximately 60% of MVV. Therefore, the aimed level of ventilation during training was set at 60% of MVV, which was calculated from 35 times FEV1 (60% MVV = 0.6 * 35 * FEV1). The dead space was adjusted to 60% of the participant's IVC plus the resting tidal volume. Participants trained twice daily, for 15 min, 7 d/week for 5 weeks. Control/sham: participant breathed 6‐7 times/min through an incentive flowmeter (Inspirx; Resprecare Medical; the Hague, the Netherlands). Airflow resistance was set at 5% of PImax.
Outcomes	Dyspnea: Borg Notes: Borg was measured at isocapnic time during constant‐load exercise training on a cycle ergometer Functional exercise capacity:  6MWD Wmax Exercise time (Constant cycle ergometer test) HRQoL: CRQ (Total) Respiratory muscle strength: PImax (RV) Laboratory exercise test: VO2peak (mL/kg/min) Respiratory muscle endurance: Respiratory muscle endurance pressure (Pthmax) Respiratory function: FEV1 (L)
Identification	Sponsorship source: Country: The Netherlands Setting: Department of Pulmonology Dekkerswald, University Medical Center Nijmegen Author's name: Ralph J. H. Koppers Institution: MedicalCenter Leeuwarden, Leeuwarden; and Department of Pulmonology (Drs. Vos, Boot, and Folgering), Dekkerswald, University Medical Center Nijmegen Email: R.J.H.Koppers@ZNB.nl Address: Nijmegen, the Netherlands
Notes

Langer 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 73 (4) Gender (M/F): 4/6 BMI, mean (SD), kg/m²: 24.1 (4.6) Control/sham N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 67 (8) Gender (M/F): 3/7 BMI, mean (SD), kg/m²: 25.1 (6.7) Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 70 (7) Gender (M/F): 7/13 BMI, mean (SD), kg/m²: 24.6 (5.6) COPD stage (GOLD): moderate to very severe Included criteria Participants were clinically stable COPD patients with reduced inspiratory muscle strength (Pimax < 70 cmH2O measured at plethysmographic FRC) Persistent activity‐related dyspnea (BDI < 9) despite optimal medical therapy Excluded criteria Inability to perform physiological testing Active cardiovascular comorbidity (i.e.severe heart failure with reduced left ventricular ejection fraction, cardiomyopathy, recent acute myocardial infarction, cardiac arrhythmias, or stroke), or other conditions that could impact dyspnea or exercise capacity
Interventions	Intervention characteristics IMT: the training was performed with a home‐based protocol using an electronic device: Powerbreathe KH2 (HAB International, Southam, UK). Participants trained 2‐3 daily sessions of 30 breaths (4‐5 min/session) performed 7 days/week for 8 weeks. The training load started at around 40% of PImax and it was increased weekly until the highest tolerable intensity Control/sham: this group performed IMT at a load of < 10% of PImax
Outcomes	Dyspnea: Borg isotime cycle ergometer test) peak exercise cycle ergometer test) Notes: Borg measured at isotime: cycle ergometer test Dyspnea: BDI‐TDI: Total Dyspnea: mMRC Functional exercise capacity: Exercise time (constant cycle ergometer test) Respiratory muscle strength: PImax  RV FRC Respiratory muscle endurance time (Tlim) Notes: measured through breathing against 50%‐60% of PImax Laboratory exercise test: VO2peak (Constant cycle ergometer test) (L/min) Respiratory function: FEV1 (L)
Identification	Sponsorship source: this work was supported by the Ontario Thoracic Society, Spear/StartEndowment Fund, Queen’s University. D. Langer received a postdoctoral fellowship and travel grant from the Research Foundation Flanders; thePowerBreathe devices used in the study were provided by HaB International Country: Canada Setting: Queen’s University Health Sciences and Affiliated Teaching Hospitals Author's name: D. E. O’Donnell Institution: Respiratory Investigation Unit, Queen’s University and Kingston Health Sciences Centre Email: odonnell@queensu.ca Address: 102 Stuart St., Kingston, ON, Canada, K7L 2V6 Clinical trial register: NCT01900873
Notes	webplotdigitizer used to extract IMT training load intensity. Adjusted analysis were provided.

Larson 1988.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 60 (6) Control/sham N (randomized/analyzed): not reported/12 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 68 (3) Overall N (randomized/analyzed): 45/22 Loss to follow‐up or excluded: 23 Gender (M/F): 20/2 COPD stage (GOLD): severe to very severe Included criteria FEV1 < 65%pred Stable condition Patients with a recent history of exacerbation were allowed to enter the study only after they reported that their ability to function and day‐to‐day symptoms had returned to baseline All patients were enrolled > 2 months after recovery from an exacerbation except 1 who enrolled 1 month after recovery from an exacerbation because he insisted that his breathing would not improve any further Patients were advised against participation if they indicated that the 12MWD was too strenuous. Excluded criteria Evidence of restrictive lung disease based on lung volumes A history of asthma, and if they had a history of cardiovascular disease or musculoskeletal conditions that could interfere with either the training or testing maneuvers Taking psychotropic drugs or abusing alcohol Participants were dropped from the study if they: experienced an exacerbation, required a change in their pharmacologic regimen, or reported < 80% compliance with the training protocol.
Interventions	Intervention characteristics IMT: participants trained 7 d/week for 8 weeks using the Threshold IMT device at 30% of PImax. They initiated the training for 15 min/d during the first week and gradually increased the duration to 30 min/day for the remaining 7 weeks. Control/sham: this group received the same IMT protocol with the training load set at 15% of PImax
Outcomes	Respiratory muscle strength: PImax (RV) Functional exercise capacity: 12MWD Respiratory muscle endurance time: Tlim Note: measured through breathing against 66% of PImax
Identification	Sponsorship source: supported in part by Grant No. HL‐31558 from the National Institutes of Health and by a grant from Sigma Theta Tau, Psi Chapter Country: USA Author's name: Janet L. Larson Institution: College of Nursing, University of Illinois at Chicago  Address: 845 S.Damen Avenue, Chicago,IL 60612
Notes

Larson 1999.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): not reported/14 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 68 (6) BMI, mean (SD), kg/m²: 27 (4) PR N (randomized/analyzed): not reported/14 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 66 (6) BMI, mean (SD), kg/m²: 26 (4) IMT N (randomized/analyzed): not reported/13 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 66 (5) BMI, mean (SD), kg/m²: 28 (4) Control/sham N (randomized/analyzed): /12 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 62 (7) BMI, mean (SD), kg/m²: 26 (5) Overall N (randomized/analyzed): 130/53 Loss to follow‐up or excluded: 83 COPD stage (GOLD): moderate to severe Included criteria Patients between 45 and 75 years of age;  Moderate to severe airflow obstruction (FEV1 < 65%pred and FEV1/FVC < 70%) Complaints of dyspnea on exertion Clinically stable condition No participation in a pulmonary rehabilitation program in the last year Excluded criteria A history of asthma Experienced a major exacerbation in the 2 months before enrollment Took > 10 mg of prednisone/d Required home oxygen therapy or experienced oxyhemoglobin desaturation < 85% with exercise, and/or had other health problems that would interfere with exercise
Interventions	Intervention characteristics PR:  participants performed cycle ergometer training at home on a calibrated stationary cycle ergometer (BodyGuard 990; BodyGuard, Sandnes, Norway), 20 min/d, 5 d/week for 2 months. An interval training protocol was used with participants performing 4 work sets, 5 min in duration, separated by rest intervals (2–4 min) of unloaded cycling. The training was initiated at 50% of the peak work rate, taken from the best baseline graded exercise test, and evaluated weekly with progressive increases as tolerated. Patients were instructed to pedal at a rate of 60 revolutions/minute (rpm), and they were encouraged to push themselves to the limits of their dyspnea, without exceeding a heart rate equal to 85% of the predicted maximal heart rate. IMT: the participants trained with Threshold IMT (HealthScan, Cedar Grove, NJ), 30 min/d, 5 d/week for 2 months. the training was initiated at 30% of PImax with progressive increases up to 60% of PImax PR+IMT: participants received both PR and IMT as described above. Control/sham: this group participated in a structured program of health education
Outcomes	Dyspnea: Borg Pthmax  Wmax Submaximal exercise at 50% of Wmax HRQoL: CRQ Dyspnea Fatigue Respiratory muscle strength: PImax (RV) Respiratory muscle endurance: respiratory muscle endurance pressure (Pthmax) Functional exercise capacity: Wmax Laboratory exercise test: VO2peak (L/min)
Identification	Sponsorship source: funded by a research grant from the National Institute of Nursing Research, National Institutes of Health, RO1‐NR01428 Country: USA Author's name: Janet L. Larson Institution: University of Illinois at Chicago Email: LLarson@uic.ed Address: 845 S. Damen, Chicago, IL60612
Notes

Leelarungrayub 2017.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 12/10 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 63.25 (1.49) Gender (M/F): 4/6 BMI, mean (SD), kg/m²: 20.6 (1.9) Control/sham N (randomized/analyzed): 12/10 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 68.75 (2.39) Gender (M/F): 6/4 BMI, mean (SD), kg/m²: 22.1 (2.4) Overall N (randomized/analyzed): 24/20 Loss to follow‐up or excluded: 2 Gender (M/F): 10/10 COPD stage (GOLD): moderate Included criteria All the participants were ex‐smokers They had a stable clinical condition during the experiments Excluded criteria Uncontrolled hypertension, unstable cardiac disease, recurrent symptoms of pneumothorax, thoracic or chest pain including neuromuscular disorders, liver diseases or endocrine abnormalities Participants taking supplements or any nutrients such as vitamins or N‐acetylcysteine compounds during this study
Interventions	Intervention characteristics IMT: participants trained with Portex (Smith Medical ASD), 20‐30 min/session, 7 d/week for 6 weeks. They started breathing through a 6 mm hole once daily for the first 2 weeks, before changing to 4 mm and 2 mm holes in the 2nd and 4th week, respectively. 30 slowly repeated inspirations passed through the device, with a 3‐min interval of rest in each of 4 training sessions. Control: this group did not receive any intervention
Outcomes	Functional exercise capacity: 6MWD HRQoL: CCQ Symptom score Mental score  Function score Total score Respiratory muscle strength: PImax (RV) Respiratory function: FEV1 L %pred
Identification	Sponsorship source: Country: Thailand Setting: Sansai hospital, Sansai, Chiang Mai Province, Thailand Author's name: Jirakrit leelarungrayub Institution: Department of Physical Therapy, Faculty of associated Medical sciences, Chiang Mai University Email: donrawee.leela@cmu.ac.th Address: Intawaroroj road, Sripoom, Chiang Mai 50200, Thailand
Notes	We calculated the SD manually from the SE (SD = SE*√N). This study also used a prototype that was excluded from our analysis because it hasn't been validated yet. We contacted the Portex manufacturer to ask for further information about the device.

Lisboa 1997.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 61 (6.32) Age (SD): 6.32 Gender (M/F): 6/4 Control/Sham N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 64 (6.32) Age (SD): 6.32 Gender (M/F): 7/3 Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Gender (M/F): 13/7 COPD stage (GOLD): severe Included criteria A stable period of their disease Dyspnea during ADL FEV1/FVC < 60% The absence of cardiac or any other disease that could interfere with exercise performance Excluded criteria: not reported
Interventions	Intervention characteristics IMT: participants trained at home using a Threshold IMT device (HealthScan Products Inc., NJ, USA), for 30 min/d, 6 d/week for 10 weeks at 30% of PImax Control/sham: this group received a similar IMT protocol and trained at 10% of PImax
Outcomes	Dyspnea: BDI‐TDI (Total) Dyspnea: Borg Notes: It was evaluated under basal conditions Functional exercise capacity: 6MWD Respiratory muscle strength: PImax (FRC) Respiratory function: FEV1 (L) Functional exercise capacity: Wmax Notes: we considered 1W = 6.11 kpm/min Laboratory exercise test: VO2peak (mL/min)
Identification	Country: Chile Author's name: C. Lisboa Institution: Dept of respiratory disease, Catholic University of Chile Address: Santiago, Chile
Notes	The authors mentioned COPD in the discussion section.

Mador 2005.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 19/15 Loss to follow‐up or excluded: 4 Age mean (SD) in years: 69.7 (7.7) BMI, mean (SD), kg/m²: 28.3 (8.9) PR N (randomized/analyzed): 19/14 Loss to follow‐up or excluded: 5 Age mean (SD) in years: 70.9 (7.4) BMI, mean (SD), kg/m²: 26.9 (3.3) Overall N (randomized/analyzed): 38/29 Loss to follow‐up or excluded: 9 COPD stage: moderate to very severe Included criteria Confirmed diagnosis of COPD
Interventions	Intervention characteristics PR+IMT PR: participants trained on a cycle ergometer, initially at 50% of Wmax and keeping Borg score ≤ 5 during exercise. They also trained on a treadmill at a speed ranging from 1.1‐2.0 miles per hour (1.7‐3.2 km/h) at 0% elevation based on the participant’s functional capacity (i.e. on 6MWD results). When the participants could exercise for 20 min without intolerable dyspnea or leg fatigue, the speed and/or elevation was increased. IMT: participants trained 15‐20 min/d, 3 d/week for 8 weeks, and breathed from a rebreathing bag while obtaining fresh air through a side port. V˙E and PetCO2 were continuously recorded using a metabolic cart (Medgraphics; St. Paul, MN), and oxygen saturation was measured by pulse oximetry. Rebreathing bags of 1.5‐2.0 L, depending on the participant’s vital capacity, were used, and the size of the bag was additionally adjusted with a clamp until stable normocapnia as estimated by the PetCO2 was obtained during preliminary trials. PR: this group received only the PR program described above
Outcomes	Functional exercise capacity:  6MWD Exercise time  Wmax  HRQoL: CRQ Respiratory muscle strength: PImax Respiratory muscle endurance time: Tlim
Identification	Country: USA Author's name: M. Jeffery Mador Institution: Division of Pulmonary, Critical Care, and Sleep Medicine, Section 111S Email: mador@acsu.buffalo.edu Address: State University of New York at Buffalo, Veterans Administration Medical Center, 3495 Bailey Ave, Buffalo, NY 14215
Notes

Magadle 2007.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 16/14 Loss to follow‐up or excluded: 2 Age (mean): 65.2 (13.6) PR N (randomized/analyzed): 15/13 Loss to follow‐up or excluded: 2 Age (mean): 66.1 (12.39) Overall N (randomized/analyzed): 31/29 Loss to follow‐up or excluded: 2 COPD stage (GOLD): moderate to severe Included criteria Patients with spirometric evidence of significant chronic airflow limitation (FEV1 < 50%pred, FEV1/FVC < 70%pred) and were diagnosed as having COPD according to ATC criteria Excluded criteria Patients with cardiac disease, poor compliance, or requirement of supplemental oxygen
Interventions	Intervention characteristics PR+IMT PR: this program included lower extremity endurance exercise (walking or cycling), upper extremity exercise and strength training with free weights. This phase included 36 sessions of 90 min duration (3 times/week for 12 weeks). In the 2nd phase, participants trained 1 h 3 times/week for 6 months. IMT: started in the 2nd phase of the trial, and consisted of training 30 min/session, 3 times/week for 6 months. They used the Threshold IMT device at a training load ranged from 15%‐ 60% of PImax. PR (+sham IMT): this group received a similar training protocol and participants performed IMT at no load.
Outcomes	Functional exercise capacity: 6MWD HRQoL: SGRQ (Total) Respiratory muscle strength: PImax (RV) Respiratory function: FEV1 (%pred)
Identification	Country: Israel Setting: community‐based rehabilitation center Author's name: Paltiel Weiner Institution: Department of Medicine A, Hillel Yaffe Medical Center Email: weiner@hillel‐yaffe.health.gov.il Address: Hadera 38100, Israel
Notes	Participants were enrolled in a 1st phase that consisted of 36 sessions of PR for 3 months (90 min/session).

Majewska‐Pulsakowska 2016.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 13/13 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 61.5 (6.1) Gender (M/F): 10/3 BMI, mean (SD), kg/m²: 28.8 (6.2) PR N (randomized/analyzed): 9/9 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 62.3 (5.2) Gender (M/F): 6/3 BMI, mean (SD), kg/m²: 28.2 (6.2) IMT N (randomized/analyzed): 8/8 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 63.4 (9.8) Gender (M/F): 2/6 BMI, mean (SD), kg/m²: 26.1 (5.9) Control/sham N (randomized/analyzed): 13/13 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 65.5 (7.0) Gender (M/F): 7/6 BMI, mean (SD), kg/m²: 27.8 (4.9) Overall N (randomized/analyzed): 43/43 Loss to follow‐up or excluded: 0 Gender (M/F): 25/18 COPD stage (GOLD): moderate to severe Included criteria COPD treated for at least 1 year, stage 2 and 3 according to GOLD (2014) 50–70 years old A stable clinical condition with no exacerbations over the period of 4 weeks before the study Excluded criteria Participation in PR in the year preceding the study Diagnosed bronchial asthma Long‐term home oxygen therapy Clinically significant diseases of the cardiovascular system Any uncontrolled chronic disease; muscle and nervous disorders reducing the patient's mobility Mental disorders preventing contact and co‐operation with the patient
Interventions	Intervention characteristics PR: consisted of training on a cycle ergometer, 3 times/week for 8 weeks in an ambulatory setting under the supervision of a cardiologist. Each training session began (warm‐up) and finished (relaxation) with a pedalling load of 10 W for 3 min. The duration of a training session was initially 23 min, and then it was gradually increased up to 45 min. IMT: consisted of home‐based training using Threshold IMT device (Respironics; Philips Healthcare, DA Best, The Netherlands), twice a day (5‐15 min), 5 d/week for 8 weeks. The training load ranged from 30%‐60% of PImax PR+IMT: this group received both interventions described above. Control/sham: this group did not receive any intervention.
Outcomes	HRQoL: SGRQ Respiratory function: FEV1  L %pred
Identification	Country: Poland Author's name: K. Wytrychowski Institution: Department of Internal Diseases, Geriatry andAllergology, Wroclaw Medical University Email: Polande‐mail:anhw@op.plK Address: 66 Sklodowskiej‐Curie St., 50‐369 Wroclaw, Poland
Notes

Masanga 2011.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): not reported/11 PR N (randomized/analyzed): not reported/9 Overall N (randomized/analyzed): not reported/21 Included criteria: stable patients with moderate to severe COPD Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: PR: consisted of Occupational therapy, Education, Dietary instruction. IMT: at a training load that ranged from 10 cmH2O to 40‐90 cmH2O. PR: Participants received only the PR program described above.
Outcomes	This abstract reported only adverse events: headache (6), jaw pain (6), neck pain (6), back pain (4), abdominal pain (2), cough (1), blood‐streaked sputum (1), shoulder pain (1) and chest pain (1)
Identification	Country: Philippines Author's name: L. Masanga
Notes

Nikoletou 2016.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 34/21 Loss to follow‐up or excluded: 13 Age, mean (SD) in years: 70.1 (8.4) Gender (M/F): 14/9 BMI, mean (SD), kg/m²: 24.9 (5.2) Control/sham N (randomized/analyzed): 34/18 Loss to follow‐up or excluded: 16 Age, mean (SD) in years: 71.1 (9.6) Gender (M/F): 11/8 BMI, mean (SD), kg/m²: 26.5 (8.1) Overall N (randomized/analyzed): 68/39 Loss to follow‐up excluded: 29 Gender (M/F): 25/17 COPD stage (GOLD): moderate to very severe Included criteria No exacerbation of COPD Not changed their medication for at least 4 weeks prior to the initial assessment Excluded criteria Patients with an a1‐antitrypsin deficiency Co‐existing heart disease, hypertension, or long‐term use of oral corticosteroids Patients with significant thoracic musculoskeletal abnormalities, such as kyphosis or scoliosis Patients unsuitable for magnetic stimulation, for example, those with cardiac pacemakers
Interventions	Intervention characteristics IMT: 2 sessions/d, 6 d/week for 7 weeks, home‐based, using the Powerbreathe inspiratory muscle trainer (HaB International, Southam, Warwickshire, UK). Participants started training at 30% of their baseline PImax and increased the intensity once a week as tolerated. The average weekly increase in the intensity of training was 5% and the mean (SD) intensity at the end of the program was 62% (SD: 11.7) of the baseline PImax Control/sham: participants received a similar IMT program and trained at 15% of PImax
Outcomes	Functional exercise capacity: incremental SWT Dyspnea: Borg Notes: Borg post‐incremental SWT HRQoL: CRQ Dyspnea Fatigue Emotion Mastery HRQoL: SF‐36 Questionnaire Physical functioning Physical problems Emotional problems Social functioning Mental health Energy/ Vitality Pain General health perception Change in health Respiratory muscle strength: PImax (FRC) Respiratory muscle endurance time: Tlim
Identification	Country: UK Setting: respiratory outpatient clinics at King’s College Hospital, GP practices and British Lung Foundation Breathe Easy groups Author's name: Dimitra Nikoletou Institution: School of Rehabilitation Sciences, Faculty of Health, Social care and Education, Kingston University and St George’s University of London Email: D.Nikoletou@sgul.kingston.ac.uk Address: Grosvenor Wing, Cranmer Terrace, London SW17 0RE, UK
Notes

Paneroni 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 15/12 Loss to follow‐up or excluded: 3 Age, mean (SD) in years: 67.8 (9.8) Gender (M/F): 10/2 BMI, mean (SD), kg/m²: 26.9 (5.3) PR N (randomized/analyzed): 11/10 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 65.8 (25.0) Gender (M/F): 8/2 BMI, mean (SD), kg/m²: 25.0 (5.1) Overall N (randomized/analyzed): 26/26 Loss to follow‐up or excluded: 4 Gender (M/F): 18/4 COPD stage (GOLD): mild to very severe Included criteria MVV of < 90% of predicted The ability to perform the 6MWD The absence of chronic respiratory failure (arterial oxygen (PaO2) and carbon dioxide (PaCO2) tensions at rest > 60 and < 45 mmHg, respectively) Excluded criteria Refusal to participate Cognitive impairment as assessed by a Mini‐Mental State Examination score of < 25 points Hemodynamic instability, and severe clinical events (e.g. COPD exacerbations or heart failure treated with infusions of vasopressors and/or inotropes) in the month before enrollment
Interventions	Intervention characteristics PR+IMT:  PR: incremental exercise training according to Maltais et al (Maltais 1997) until performing 30 min of continuous cycling at 50%–70% of the maximal load calculated based on the initial 6MWD plus 30‐min sessions of abdominal, upper, and lower limb muscle activities, lifting progressively increasing weights (300–500 g), shoulder and full arm circling, and other exercises according to Clark et al (Clark 1996) IMT: participants received twice a day, 5 d/week for 2 weeks, 20 15‐min sessions of normocapnic hyperpnea training using SpiroTiger device (Idiag; Fehraltorf, Switzerland). The initial target of minute ventilation (VE) was 66% of participants’ MVV. Once the participant was able to perform 15‐min training sessions without interruptions, the target VE was increased to 75% of MVV, by increasing only the respiratory rate. Thereafter, training involved further 10% increments of MVV, when the participant could perform without interruption three 15‐min sessions at the set level PR (+sham IMT): participants in this group received a similar training PR program, and they performed IMT using Threshold IMT device at no resistance
Outcomes	Functional exercise capacity: 6MWD Respiratory muscle strength: PImax Respiratory muscle endurance time: Tlim (normocapnic hyperpnea) Notes: it is the sustained VE at 66% of their MVV. If there was no exhaustion in 15 min, the test was repeated the following day at 75% of MVV Respiratory muscle endurance time: Tlim (Threshold device) Notes: the pressure was set at 30% of the patients' baseline PImax Respiratory muscle endurance: MVV
Identification	Country: Italy Setting: Respiratory Rehabilitation Divisions of the Lumezzane and Pavia Institutes of the Istituti Clinici Scientifici Maugeri, Italy Author's name: Mara Paneroni Institution: Istituti Clinici Scientifici Maugeri, IRCCS Address: Salvatore Maugeri 2, 27100 Pavia, Italy Clinical trial register: NCT01556139
Notes

Petrovic 2012.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 58.7 (5.2) Gender (M/F): 6/4 BMI, mean (SD), kg/m²: 24.3 (4.0) Control/sham N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 60.3 (5.3) Gender (M/F): 5/5 BMI, mean (SD), kg/m²: 24.2 (3.4) Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Gender (M/F): 11/9 COPD stage (GOLD): moderate to severe Included criteria Required participants to be < 70 years old Moderate to severe non‐reversible airflow obstruction Excluded criteria Exacerbation of COPD within the previous 6 weeks Neoplastic disease or the presence of a disease that could contribute to dyspnea, or exercise limitation (cardiovascular, neuromuscular, or other respiratory diseases) No patients had received treatment with systemic cortisone over the last 6 months
Interventions	Intervention characteristics IMT: performed once/d, 7 d/week, for 8 weeks using Respifit Sdevice (Mauerbach, Australia). The training consisted of strength training at 80% of PImax (≃12 min each session) and endurance training at 60% of PImax (≃ 13.5 min each session) Control/sham: this group did not receive any intervention
Outcomes	Dyspnea: Borg Constant cycle ergometer test Incremental cycle ergometer test Notes: the sensation of dyspnea was assessed every 2 min during exercise Respiratory muscle strength: PImax Respiratory muscle endurance time: Tlim Notes: measured through breathing against 60% PImax Laboratory exercise test: VO2peak (mL/min) Constant cycle ergometer test Incremental cycle ergometer test
Identification	Country: Austria Setting: outpatient clinic Author's name: Milos Petrovic Institution: Pulmonary Department and Karl Landsteiner Institute for Clinical and Experimental Pulmology Email: milos.petrovic@wienkav.at Address: Hietzing Hospital, Wolkensbergenstrasse 1, 1130 Vienna, Austria Clinical trial register: NCT00469313
Notes

Preusser 1994.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): /12 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 68/10 Gender (M/F): 6/6 Control/sham N (randomized/analyzed): /8 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 62/17 Gender (M/F): 1/7 Overall N (randomized/analyzed): 22/20 Loss to follow‐up or excluded: 2 Gender (M/F): 7/13 COPD stage (GOLD): severe to very severe Included criteria The presence of COPD as defined by the ATS  Willingness to participate in a 3‐month inspiratory muscle training protocol Excluded criteria Uncontrolled cardiac disease, diabetes, hypertension, musculoskeletal, and neuromuscular diseases
Interventions	Intervention characteristics IMT: consisted of supervised sessions, 3 days/week for 12 weeks, using Threshold IMT device at 52% of PImax. The training duration ranged from 5 min in week 1 to 18 min in week 12 Control/sham: this group received a similar IMT protocol and trained at 22% of PImax
Outcomes	Functional exercise capacity: 12MWD Respiratory muscle strength: PImax (FRC) Respiratory muscle endurance: Respiratory muscle endurance pressure (Pthmax) Notes: "After quietly breathing through an unloaded system for 2 min, the threshold load began at ‐4 cm H2O pressure and increased by ‐2 cm after every fifth breath (every 15 s) until the subject either signalled to stop the test or was unable to match respiratory rate, duty cycle and flow rate during three of the five breaths for that pressure load. The maximal threshold load successfully completed during the ramp test was recorded"
Identification	Sponsorship source: supported by grant F31 NR06378, National Center for Nursing Research; a grant from the American Nurses Foundation; and the Epsilon Chapter of Sigma Theta Tau Country: USA Author's name: Barbara A. Preusser Institution: College of Nursing, The Ohio State University, and the Department of Pulmonary Medicine, The Ohio State University Hospitals, Columbus Address: University of Utah College of Nursing, 25 South Medical Drive, Salt Lake City 84112
Notes

Ramirez Sarmiento 2002.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/7 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 65 BMI, mean (SD), kg/m²: 29 Control/sham N (randomized/analyzed):  not reported/7 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 66 (6) BMI, mean (SD), kg/m²: 26 (4) Overall N (randomized/analyzed): 16/14 Loss to follow‐up or excluded: 2 BMI, mean (SD), kg/m²: COPD stage (GOLD): severe to very severe Included criteria Patients with COPD Excluded criteria Hypoxemia (PaO2 < 60 mmHg breathing room air) Asthma; coronary disease; undernourishment (BMI < 20 kgm²) Chronic metabolic disease Orthopedic disease Previous abdominal or thoracic surgery Treatment with steroids, hormones or cancer chemotherapy
Interventions	Intervention characteristics IMT: participants trained 30 min/day, 5 d/week for 5 weeks. The sessions were supervised and using Threshold IMT device set at 50% of PImax Control/sham: participants received a similar IMT protocol and trained at no load
Outcomes	Functional exercise capacity:  6MWD Wmax Respiratory muscle strength: PImax Respiratory muscle endurance: Respiratory muscle endurance pressure (Pthmax) Notes: breathing against incremental loads (8 cmH2O) every 2 min until maximal sustainable threshold pressure was reached Respiratory muscle endurance time: Tlim Notes: breathing against a constant a submaximal constant load (equivalent to 80% of maximal threshold pressure) Laboratory exercise test: VO2peak (mL/kg/min) Notes: measured during incremental cycle test Respiratory function: FEV1 (%pred) Respiratory function: RV (%pred)
Identification	Sponsorship source: supported, in part, by grants BIOMED (BMH‐4‐CT98‐3406), FIS, SEPAR, and SIBEL Country: Spain Author's name: Mauricio Orozco‐Levi Institution: Servei Grup de Recera de Pneumologica, Hospital del MarIMIM Email: morozco@imim.es Address: Passeig Maritim 25, E‐080003, Barcelona (Catalonia), Spain
Notes

Saher 2021.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 23/17 Loss to follow‐up or excluded: 6 Age, mean (SD) in years: 62.8 (6.72) Gender (M/F): 9/5 BMI, mean (SD), kg/m² : 22.7 (4.25) Control N (randomized/analyzed): 23/17 Loss to follow‐up excluded: 6 Age, mean (SD) in years: 61.8 (11.93) Gender (M/F): 12/8 BMI, mean (SD), kg/m² : 23.7 (5.51) Overall N (randomized/analyzed): 46/34 Loss to follow‐up or excluded: 12 Gender (M/F): 21/13 COPD stage (GOLD): moderate to very severe Included criteria Hypercapnic moderate to severe COPD patients Patients were required to have hypercapnic respiratory failure (PaCO2 > 46 mmHg), already receiving non‐invasive ventilation for at least 8 h/d, and with a decreased inspiratory muscle strength (PImax < 60 cm H2O) Excluded criteria Patients with any other severe respiratory disease apart from COPD, or with any other severe non‐pulmonary disease limiting prognosis (e.g. metastatic cancer, active tuberculosis, congestive heart failure, liver cirrhosis)
Interventions	Intervention characteristics IMT: participants performed IMT 15 min/session, 2 session/day, for 10 days. They used an IMT threshold device (Powerbreath, Gaiam Ltd., Southam, UK). The training load started at 30% of PImax, and it was increased by 5%‐10% daily until 60% Control: this group did not receive any training Note: both groups received non‐invasive ventilation (NIV) as part of COPD management.
Outcomes	Functional exercise capacity: 6MWD Inspiratory muscle strength (PImax)
Identification	Country: India Setting: Metro Center for Respiratory Disease, Metro Hospital Author's name: T. Saher Institution: Center for Physiotherapy and Rehabilitation Sciences Email: jmoiz@jmi.ac.in  Address: Jamia Millia Islamia, New Delhi, 110025, India
Notes

Saka 2021.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 23/20 Loss to follow‐up or excluded: 3 Age, mean (SD) in years: 62.30 (7.43) Gender (M/F): 19/1 BMI, mean (SD), kg/m² : 26.04 (4.41) Sham IMT N (randomized/analyzed): 22/20 Loss to follow‐up excluded: 2 Age, mean (SD) in years: 62.10 (7.76)  Gender (M/F): 18/2 BMI, mean (SD), kg/m² : 27.11 (4.88) Overall N (randomized/analyzed): 45/40 Loss to follow‐up or excluded: 5 Gender (M/F): 37/3 COPD stage (GOLD): moderate to severe Included criteria FEV1:FVC ratio < 70% in PFT Age: ≥ 18 years old Being able to read written and understand spoken language Excluded criteria A history of COPD exacerbation in the last 6 weeks The presence of comorbidities affecting ambulation/activity (e.g. severe cardiac or neurological disorders, cancer, musculoskeletal problems) and cognitive disorders (Mini‐Mental State Examination  < 24)
Interventions	Intervention characteristics IMT: the training was performed 15 min twice a day, 5 d/week, for 8 weeks. Home‐based training with 1 session supervised by a physiotherapist. Participants used Threshold IMT device (Threshold IMT® Philips Respironics, UK) at a load of 30% of PImax. Sham IMT: participants received the same protocol with a training load set at 15% of PImax.
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD HRQoL: SGRQ Symptoms Activity Impacts Total HRQoL: CAT Inspiratory muscle strength (PImax) Respiratory function: FEV1 L %pred
Identification	Sponsorship: Scientific Research Projects Unit of Bezmialem Vakıf University, project number 9.2017/31 Country: Turkey Setting: Bezmialem Vakif University Division of Physiotherapy and Rehabilitation, Cardiopulmonary Physiotherapy and Rehabilitation Department. Author's name: Seda Saka Institution: Cardiopulmonary Physiotherapy Rehabilitation Department, Institute of Health Sciences, Bezmialem Vakif University, Istanbul, Turkey Email: fztsedasaka@gmail.com Address: Division of Physiotherapy and Rehabilitation, Faculty of Health Sciences, Bezmialem Vakif University, Silahtaraga St. No:189, Alibeykoy, 34060, Istanbul, Turkey Clinical trial register: NCT03517839
Notes

Sanchez Riera 2001.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 67 (4) Gender (M/F): 9/1 Control/Sham N (randomized/analyzed): 10/10 Loss to follow‐up/excluded: 0 Age, mean (SD) in years: 67.6 (5) Gender (M/F): 9/1 Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Gender (M/F): 18/2 COPD stage (GOLD): moderate to severe Included criteria The presence of COPD as defined by the ATS Patients were in stable condition Excluded criteria Clinical evidence of cardiovascular, musculoskeletal, or neuromuscular disease or of any other disease that might interfere with exercise
Interventions	Intervention characteristics IMT: participants trained at home 30 min/d, 6 d/week for 6 months using incentive flowmeter device (INSPIRx; Intertech Resources Inc; Ft. Myers, FL) set at 30% of PImax. The breathing pattern was controlled during the exercise. Control/sham: participants received a similar IMT protocol and trained at no load.
Outcomes	Dyspnea: Borg (Constant cycle ergometer test) Functional exercise capacity: SWT Notes: "Patients were required to walk 10 m back and forth. The walking speed was paced by an audio signal from a cassette that emitted beeps at regular intervals. The speed was increased each minute by 0.17 m/s until the next level was attained. The end of the test was determined by patients when they were too breathless to maintain the required speed, or by operators, if patients failed to complete a shuttle in the time allowed" Functional exercise capacity: Wmax  Notes: "performed on a cycle ergometer. After 1 min of unloaded pedalling, the work rate was increased 10 W/min at a time. The test was stopped when patients were unable to continue because of dyspnea or leg fatigue" Respiratory muscle strength: PImax (FRC) Laboratory exercise test: VO2peak (L/min)  Respiratory muscle endurance: Respiratory muscle endurance pressure (Pthmax)
Identification	Sponsorship source: Supported by “Junta de Andalucia” grant No. 94//535–119 Country: Spain Setting: home‐based training Author's name: Hildegard Sanchez Riera Institution: Pneumology Service, Virgen Del Rocio University Hospital Email: ablucil@mx2.redestb.es Address: “La Motilla,” C/Rayo 4, 41700 Dos Hermanas, Sevilla, Spain
Notes	CRQ was not included because study did not fully report control group data

Scherer 2000.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/15 Loss to follow‐up or excluded: not reported Age, mean (SD) in years: 66.9 (9.2) Gender (M/F): 9/6 BMI, mean (SD), kg/m²: 23.8 (3) Control/sham N (randomized/analyzed): 15 Loss to follow‐up or excluded: not reported Age, mean (SD): 71.0 (4.6) Gender (M/F): 10/5 BMI, mean (SD), kg/m²: 25.9 (3.48) Overall N (randomized/analyzed): 34/30 Loss to follow‐up or excluded: 4 Range age (min, max): 46,80 Gender (M/F): 19/11 COPD stage (GOLD): moderate to severe Included criteria Chronic airflow obstruction (FEV1 < 70%pred, FEV1/FVC < 70%pred, < 15% improvement in FEV1 after bronchodilatation with 200 mg of albuterol inhaled from a pressurized metered‐dose inhaler with a spacer) Aged 20‐80 years A stable clinical condition for at least 1 month The patients’ physical activity had to be limited by pulmonary dyspnea only Excluded criteria Patients with dyspnea at rest Cardiac disease, poor compliance, drug or alcohol abuse, pregnancy or lactation, a requirement for supplemental oxygen, CO2 retention, or use of any mechanical ventilatory support
Interventions	Intervention characteristics IMT: respiratory muscle endurance training was conducted twice daily, 5 d/week for 8 weeks. The device was developed and consisted of tubing (I.D. 5 19 mm) connecting a rebreathing bag with a mouthpiece at a 90° angle. The breathing frequency was 60% of MVV. Control/sham: participants trained following the same pattern of the description above, at no load using incentive spirometer (COACH 2 Volumetric Incentive Spirometer; DHD Healthcare, Canastota, NY). The respiratory rate was 6‐8 breaths/min, and the target inspiration was set at 70% of each participant’s ventilatory capacity.
Outcomes	Dyspnea: BDI‐TDI (Total) Functional exercise capacity:  6MWD Exercise time (treadmill) Respiratory muscle strength: PImax (RV) Laboratory exercise test: VO2peak (mL/kg/min) Respiratory muscle endurance time: Tlim Notes: respiratory muscle endurance was measured as sustained ventilation at 66% of each participant’s highest MVV. The time during which participants were able to sustain this target ventilation was recorded. If a participant surpassed 15 min of breathing at this level, the test was repeated on the following day at 75% of MVV. Respiratory function: FEV1 (%pred)
Identification	Sponsorship source: supported by grants from Astra Pharmaceutica, Dietikon, and Merck Sharpe and Dohme‐Chibret, and Rhône‐Poulenc Rorer Country: Switzerland Setting: outpatient clinic of the Pulmonary Division of the Triemli Hospital Author's name: Thomas A. Scherer Institution: LungenZentrum Hirslanden Email: thsche@swissonline.ch Address: Witellikerstrasse 36, 8008 Zurich, Switzerland
Notes

Schultz 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 307/300 Loss to follow‐up or excluded: 26 Age, mean (SD) in years: 57.7 (8.2) Gender (M/F): 188/112 BMI, mean (SD), kg/m²: 26.6 (6.4) PR N (randomized): 304/302 Loss to follow‐up or excluded: 24 Age, mean (SD) in years: 57.9 (6.6) Gender (M/F): 201/101 BMI, mean (SD), kg/m²: 26.9 (6.6) Overall N (randomized): 611/602 Loss to follow‐up or excluded: 50 Gender (M/F): 389/213 COPD stage (GOLD): moderate to severe Included criteria Medical history of COPD FEV1/FVC < 70% and FEV1 %pred < 80% post‐bronchodilation Excluded criteria Lack of language or cognitive abilities to fill out questionnaires Hypercapnic respiratory failure (arterial carbon dioxide tension > 50 mmHg at rest) Indication for intermittent noninvasive ventilation Contraindications for inspiratory muscle training (e.g. a history of recent lung surgery, recent pulmonary embolism; a history of recurrent spontaneous pneumothorax) Severe comorbidities that confer significantly greater morbidity than COPD (e.g. active cancer without successfully completed curative tumor therapy)
Interventions	Intervention characteristics PR+IMT:  PR: consists of 2 components  Obligatory components: (mostly 30‐ to 60‐min sessions) including physical training (endurance training: 4 or 5 sessions per week; strength training: three sessions/week; whole‐body vibration muscle training: 7 sessions/week), patient education (≥ 7 sessions) and respiratory physiotherapy in groups (2‐4 sessions/week).  Optional components: smoking cessation (8 sessions), mucolytic physiotherapy, saline inhalation, psychological interventions, social counselling, nutritional counselling and occupational therapy IMT: conducted 21 min/d, 7 d/week for 3 weeks, using Threshold IMT device (POWERbreathe Medic; POWERbreathe International, Southam, UK). The initial training load was at least 30% of PImax and was progressively increased to at least 60%, and about half of the sessions were supervised PR (+ sham IMT): this group received a similar training protocol as described above and the IMT was conducted at no load.
Outcomes	Dyspnea: BDI‐TDI (Total) Functional exercise capacity: 6MWD HRQoL SGRQ (Total) CAT CCQ: (Total score) Respiratory muscle strength: PImax Respiratory function: FEV1 (L)
Identification	Sponsorship source: this study was supported by Deutsche Rentenversicherung Bayern Süd. Funding information for this article has been deposited with the Crossref Funder Registry Country: Germany Setting: Bad Reichenhall Clinic Author's name: Konrad Schultz Institution: Center for Rehabilitation, Pulmonology and Orthopedics, Klinik Bad Reichenhall Email: konrad.schultz@klinik‐bad‐reichenhall.de Address: Salzburger Strasse 8–11, 83435 Bad Reichenhall, Germany Clinical trial register: DRKS00004609
Notes	Adjusted mean differences with 95% CI were reported

Sykes 2005.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 PR N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Overall N (randomized/analyzed): 40/40 Loss to follow‐up or excluded: 0 Range age (min, max): 60,84 Gender (M/F): 34/6 Included criteria: not reported Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: participants received exercise training and IMT with Threshold IMT at a load ranged from 30%‐60% of PImax PR: this group received only exercise training
Outcomes
Identification	Country: China Setting: Tai Po Hospital Author's name: Sykes
Notes	The systematic review (Gosselink 2011) reported data from this trial.

Tounsi 2021.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 17/16 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 62 (5) Gender (M/F): 17/0 BMI, mean (SD), kg/m²: 23.13 (4.37) PR N (randomized): 18/16 Loss to follow‐up or excluded: 2 Age, mean (SD) in years: 63 (4) Gender (M/F): 18/0 BMI, mean (SD), kg/m²: 23.41 (5.19) Overall N (randomized): 35/32 Loss to follow‐up or excluded: 3 Gender (M/F): 35/0 COPD stage (GOLD): moderate to very severe Included criteria patients with forced expiratory volume in 1 s (FEV1) <80% predicted and FEV1 /forced vital capacity (FVC) <70%  Age between 45–75 years Clinically stable Exclusion criteria Cardiovascular problem  Psychiatric or cognitive disorders Progressive neuromuscular diseases Severe orthopedic problems with a significant impact on daily activities  Prior inclusion in a rehabilitation program (<1 year)
Interventions	Intervention characteristics PR+IMT:  PR:  an 8‐week program with three sessions per week. the training program consists of aerobic supervised exercise training. The training consisted of 30 min of supervised treadmill exercise per session. Each session ended with upper and lower limb stretching. Each participant received an individualized program based on 60% to 80% of the average speed achieved during the six‐minute walk test. IMT: The training was performed once a day, 7 days per week for 8 weeks, using a handle device (PowerBreathe1 Medic, IMT Technologies Ltd, Birmingham, UK). The training consists in making two sets of 30 breaths (4–5 min/set) with 5–10 min of rest between each set. The Respiratory resistive load was set at 50% of the initial PImax and then increased by 10% of the initial PImax every two weeks of training. Part of the IMT was performed and well instructed in the pulmonary rehabilitation center (3 days/ week) for 8 weeks; the other part was home‐based training. PR: this group received a similar training protocol as described above without IMT.
Outcomes	Functional exercise capaticy: 6MWD Respiratory muscle strength: PImax
Identification	Sponsorship source: the authors received no funding for this work Country: Tunisia Setting: Farhat Hached Hospital of Sousse Author's name: Bilel Tounsi Institution: 1/ Laboratory of Exercise Physiology and Rehabilitation (APERE, UR‐EA 3300), Sport Sciences Department, Picardie Jules Verne University, Amiens, France, 2/ Research Laboratory of Exercise Physiology and Pathophysiology: From Integral to Molecular Biology, Medicine and Health (LR19ES09), Faculty of Medicine of Sousse, University of Sousse, Sousse, Tunisia Email: bilel.tounsi@u‐picardie.fr Address: Department of sports sciences, Picardie Jules Verne University, Amiens, France Clinical trial register:  NCT04084405
Notes

Tout 2013.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 61 (9.32) PR N (randomized/analyzed): 10/10 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 58.1 (8.72) Overall N (randomized/analyzed): 20/20 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 60.38 (8.02) Gender (M/F): 19/21 BMI, mean (SD), kg/m²: 28.3 (3.11) COPD stage (GOLD): mild to moderate Included criteria Co‐operating COPD patients Diagnosed clinically and through spirometric measurement (grades 1 and 2 in the Gold classification); presenting with 50% < FEV1 < 80% of predicted or theoretical value in the spirometric test Presenting with an improvement < 15% of the FEV1 following use of bronchodilators From 45‐75 years of age Of either sex Excluded criteria Heart failure or associated cardiac pathology Previous pulmonary or cardiac surgery Patient depending on oxygen therapy or undergoing cortisone treatment Associated neuromuscular pathologies
Interventions	Intervention characteristics PR+IMT: PR: 16 rehabilitation sessions that consisted of bronchial decluttering meant to clear the airways; diaphragmatic rehabilitation (solicitation of the physiological diaphragmatic contraction) aimed at improved stamina; reinforcing of the lower limb muscles so as to limit functional deconditioning; psychological support and therapeutic education IMT: 20‐30 min/session, twice/week for 8 weeks, using Threshold IMT at a training load ranged from 30%‐60% of PImax PR: this group received only the PR protocol described above.
Outcomes	Functional exercise capacity: 6MWD HRQoL: SGRQ Symptoms Activity Impact Total Respiratory function: FEV1 (L)
Identification	Country: Lebanon Author's name: Rola Tout Institution: Institut de Physiothérapie, université Saint‐Joseph Email: rolatout@yahoo.com Address: URAF, rue de Damas, Beyrouth, Lebanon
Notes

Wang 2017.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group Subgroup analysis: PImax less or more than 60 cmH2O
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 28/28 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 70.8 (4.5) BMI, mean (SD), kg/m²: 21.31 (2.76) PR N (randomized/analyzed): 27/27 Loss to follow‐up/ excluded: 0 Age, mean (SD) in years: 70.6 (6.3) BMI, mean (SD), kg/m²: 22.40 (2.85) Overall N (randomized/analyzed): 55/55 Loss to follow‐up or excluded: 0 COPD stage (GOLD): moderate to severe Included criteria Age > 40 years Diagnosis of stable COPD based on the GOLD guidelines; No participation in any PR program in the previous 2 months The participants had to be able to understand the investigator’s instructions and complete the tests required in the study Excluded criteria Acute or chronic airway diseases other than COPD, cardiovascular disorders (such as acute coronary syndrome), metabolic conditions (such as diabetes or hyperthyroidism), or other health problems that would interfere with exercise performance or the testing procedures
Interventions	Intervention characteristics PR+IMT: PR: consisted of 30 min of cycle ergometer training 3 times/week for 8 weeks. The exercise intensity threshold in lower limbs was calculated as 70% VO2max during cardiopulmonary exercise training IMT: conducted 3 times/week for 8 weeks using a Threshold IMT device (model HS730, Philips, Amsterdam, the Netherlands), around 14 min/session at 30% of PImax PR: this group received only the PR protocol described above
Outcomes	Dyspnea: Borg  6MWD Incremental cycle ergometer test Notes: Borg was measured during peak cycle endurance test Dyspnea: mMRC Functional exercise capacity:  6MWD Wmax: measured through an incremental load of 5 w/min or 10 w/min HRQoL: CAT HRQoL: SGRQ (Total) Respiratory muscle strength: PImax Respiratory muscle endurance: MVV Laboratory exercise test: VO2peak  mL/min mL/kg/min Respiratory function: FEV1 %pred L
Identification	Sponsorship source: this work was supported by Guangzhou Municipal Science and Technology Project (201507020033), Medical Scientific Research Foundation of Guangdong Province (A2016399), Open Project of State Key Laboratory of Respiratory Disease (SKLRD2016OP019), and Clinical Research training program of Southern Medical University (LC2016PY032). Country: China Setting: Zhujiang Hospital affiliated to Southern Medical University Author's name: Xin Chen Institution: Department of Respiratory Medicine, Zhujiang Hosptial, Southern Medical University Email: chen_xin1020@163.com Address: 253 Gongye Road, Guangzhou 510282, China Clinical trial register: NCT02285400
Notes	Adjusted P values were reported. Authors reported a subgroup analysis for intervention group participants with or without respiratory muscle weakness.

Weiner 1992.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 12/12 Loss to follow‐up or excluded: 0 Age, mean (SD in years): 67.2 (9) PR N (randomized/analyzed): 12/12 Loss to follow‐up or excluded: 0 Age, mean (SD in years): 64.4 (10.3) Overall N (randomized/analyzed): 24/24 Loss to follow‐up or excluded: 0 COPD stage (GOLD): severe to very severe Included criteria Patients with spirometric evidence of chronic airflow limitation that was not corrected by bronchodilator therapy Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: PR: consisted of  20 min of cycling on cycle ergometer. Participants started cycling with low load that was then gradually increased, about 5% each session, to reach 50% of the initial Wmax.  10 min of rowing  15 min of muscle exercises to strengthen upper and lower extremities and abdominal muscles IMT: 15 min/session, 3 times/week for 6 months using the Threshold IMT device. Participants breathed at 15% of their PImax for 1 week. The resistance was then increased by 5% to reach 60% of PImax at the end of the first month and 80% at the end of the second month. PR+ (sham IMT): this group received the same training protocol as described above, and IMT was conducted at no load.
Outcomes	Functional exercise capacity:  12MWD Exercise time (constant cycle ergometer test) Notes: endurance work time at 2/3 of Wmax Respiratory muscle strength: PImax (RV) Respiratory muscle endurance: respiratory muscle endurance pressure (Pthmax) Notes: participantss inspired through a 2‐way HansRudolph valve whose inspiratory port was connected to a chamber and plunger to which weights could be added externally. Inspiratory elastic work was then increased by the progressive addition of 25‐100 g weights at 2‐min intervals. The pressure achieved with the heaviest load (tolerated for at least 60 s) was defined as the peak pressure (Pthmax) Respiratory function: FEV1 (%pred)
Identification	Country: Israel Author's name: Paltiel Weiner Institution: Department of Medicine, Hillel‐Yaffe Medical Center, and the Institute for Respiratory Disease Address: Hadera, Israel
Notes

Weiner 2000.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 12/11 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 62.5 (8.31) PR N (randomized/analyzed): 5/4 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 61.0 (5.4) Overall N (randomized/analyzed): 17/15 Loss to follow‐up or excluded: 2 COPD stage (GOLD): severe Included criteria Spirometric evidence of chronic airflow limitation Diagnosis of moderate‐to‐severe COPD according to the criteria of the ATS Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: the training program is the same as described in (Weiner 1992). Participants trained 3 times/week and each session consisted of 1 h of supervised training. When only exercise training was performed, participants trained the whole hour, and when IMT was added, exercise training was cut to 30 min. PR+ SHAM: participants in this group received the same rehabilitation protocol and IMT was conducted at no load
Outcomes	Respiratory muscle strength: PImax (RV)
Identification	Country: Israel Author's name: Paltiel Weiner Institution: Department of Medicine A, Hillel‐Yaffe Medical Center Address: Hadera, Israel
Notes

Weiner 2003.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 8/8 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 63.1 (8.7) Gender (M/F): 6/2 Control/sham N (randomized/analyzed): 8/8 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 61.8 (9) Gender (M/F): 7/1 Overall N (randomized/analyzed): 16/16 Loss to follow‐up or excluded: 0 Gender (M/F): 13/3 COPD stage (GOLD): severe Included criteria Spirometric evidence of significant chronic airflow limitation (i.e. FEV1 of < 50% of predicted and FEV1/FVC ratio of 70% of predicted) in whom COPD had been diagnosed, according to the criteria of the ATS Excluded criteria Patients with cardiac disease Poor compliance A requirement for supplemental oxygen therapy CO2 retention
Interventions	Intervention characteristics IMT: participants trained daily, 1 h/d, 6 times/week for 3 months using Threshold IMT device. They started breathing at a resistance equal to 15% of their PImax or PEmax for 1 week. The resistance then was increased incrementally 5% to 10% each session, to reach 60% of their PImax or PEmax at the end of the first month of training. Control/sham: participants trained at 7 cmH2O throughout the trial
Outcomes	Dyspnea: BDI‐TDI:  Functional impairment Magnitude of task  Magnitude of effort Focal score Functional exercise capacity: 6MWD Respiratory muscle strength: PImax (RV) Respiratory muscle endurance: respiratory muscle endurance pressure (Pthmax)
Identification	Country: Israel Author's name: Paltiel Weiner Institution: Department of Medicine A, Hillel Yaffe Medical Center Email: weiner@hillel‐yaffe.health.gov.il Address: Hadera, Israel 38100
Notes	Because only the IMT group data were reported in the results, and there were differences between the graphs and what was reported numerically, we included both groups' data from the graphs to guarantee consistency.

Weiner 2006.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 63.0 (7.66) Gender (M/F): 8/6 BMI, mean (SD), kg/m²: 28.0 (5.72) Control/sham N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 62.7 (7.58) Gender (M/F): 8/6 BMI, mean (SD), kg/m²: 28.0 (5.68) Overall N (randomized/analyzed): 28/28 Loss to follow‐up or excluded: 0 Gender (M/F): 16/12 COPD stage (GOLD): severe Included criteria Spirometric evidence of severe chronic air flow limitation (FEV 1 < 50% of predicted) and FEV 1/FVC < 70% of predicted Excluded criteria: not reported
Interventions	Intervention characteristics IMT: participants trained daily, 1 h/d, 6 d/week for 8 weeks using Powerbreathe (Southam, UK). They started breathing at a resistance equal to 15% of their PImax for 1 week. The resistance was then increased incrementally (5%–10% each session), to reach 60% of their PImax at the end of the 1st month. IMT was then continued at 60% of their PImax adjusted weekly to the new PImax achieved Control/sham: participants trained with a resistance of 7 cmH20 following the same protocol described above
Outcomes	Respiratory muscle strength: PImax (RV)
Identification	Country: Israel Author's name: Paltiel Weiner Institution: Department of Medicine A, Hillel Yaffe Medical Center Email: weiner@hillel‐yaffe.health.gov.il Address: Hadera 38100 Israel
Notes

Wu 2017.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT (Threshold device) N (randomized/analyzed): 19/19 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 59.74 (6.14) BMI, mean (SD), kg/m²: 18.40 (2.19) IMT (resistive device) N (randomized/analyzed): 21/21 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 62.24 (7.36) BMI, mean (SD), kg/m²: 19.25 (2.17) Control/sham N (randomized/analyzed): 20/29 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 60.30 (6.55) BMI, mean (SD), kg/m²: 18.54 (2.58) Overall N (randomized/analyzed): 60/60 Loss to follow‐up or excluded: 0 COPD stage (GOLD): moderate to severe Included criteria Moderate, severe and very severe COPD (post‐bronchodilator FEV1/FVC < 70% and FEV1 < 50% of predicted (GOLD B, C and D, respectively) Inspiratory muscle weakness (PImax < 60 cm H2O) Bronchial dilation test (BDT) negative No history of PR Excluded criteria Time from most recent exacerbation > 2 months With no medication changes in 1 month prior to enrollment Obesity (BMI > 30 m2/kg) Severe orthopedic problems having a major impact on ADL Previous inclusion in a rehabilitation program (< 1 year) Concomitant heart failure and pulmonary vascular diseases Diagnosed psychiatric or cognitive disorder Progressive neurological or neuromuscular disorder
Interventions	Intervention characteristics IMT: participants trained twice a day, 15 min/session for 8 weeks using either Threshold IMT (Respironics Inc; Pittsburgh, PA, USA) or Pflex (Respironics Inc, Pittsburgh, PA, USA) devices set at 60% of PImax Control/sham: no intervention received by this group
Outcomes	Dyspnea: BDI‐TDI Functional impairement  Magnitude of task Magnitude of effort Focal score Functional exercise capacity: Wmax (Incremental cycle ergometer test: measured by increasing the work rate by 10 w/min after one minute of unloaded pedalling) Exercise time (Incremental cycle ergometer test) HRQoL: CRQ Dyspnea Fatigue Emotion Mastery Total Respiratory muscle strength: PImax (RV) Laboratory exercise test: VO2peak (L/min) (Incremental cycle ergometer test)  Respiratory function: FEV1 L %pred
Identification	Sponsorship source: the study was supported by the Science and Technology Project of Guangdong Province (2017A020211018) and the Guangzhou Healthcare collaborative innovation major project (201604040012) and State's Key Project of Research and Development Plan(2017YFSF11078). Country: China Setting: Guangzhou Institute of Respiratory Disease Author's name: Rongchang Chen Institution: Guangzhou Institute of Respiratory Disease, State Key Laboratory of Respiratory Disease Email: chenrc_vip@163.com Address: Guangzhou, China Clinical trial register: NCT03101774
Notes

Xu 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group Subgroup analysis: PImax: < or > 60 cmH2O
Participants	Baseline characteristics IMT N (randomized/analyzed): 23/23 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 67.49 (6.17) BMI, mean (SD), kg/m²: 22.09 (3.37) Control/sham N (randomized/analyzed): 23/23 Loss to follow‐up or excluded: 0 Age, mean (SD) in years: 69.43 (6.44) BMI, mean (SD), kg/m²: 20.86 (4.41) Overall N (randomized/analyzed): 46/46 Loss to follow‐up or excluded: 0 COPD stage (GOLD): moderate to severe Included criteria Patients with clinically stable COPD Naive to PR and willing to participate Excluded criteria Cognitive disorders Organ failure Malignant tumors Metabolic diseases
Interventions	Intervention characteristics IMT: participants trained daily, at home, 48 min/d for 8 weeks using Threshold IMT (Respironics, USA). The training load ranged from 30%‐45% of PImax Control/sham: participants received the same protocol and trained at no load.
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD HRQoL:  SGRQ (Total) CAT Respiratory muscle strength: PImax Respiratory function: FEV1  L %pred
Identification	Sponsorship source: this work was funded unconditionally by Clinical Research training program of Southern Medical University (LC2016PY032), National Key R&D Program of China (2017YFC1310601), The Guangzhou Healthcare Collaborative Innovation Major Project (201604020012), Guangzhou Innovation and Entrepreneurship Education Project of Universities (201709T26), Special Fundsfor the Cultivation of Guangdong College Students’ Scientific and Technological Innovation (PDJHB0101). The sponsors have no any role in design, conduct, data interpretation of the study, and preparation, review or approval of this manuscript Country: China Setting: Zhujiang Hospital of Southern Medical University Author's name: Xin Chen Institution: Department of Respiratory Medicine, Zhujiang Hosptial, Southern Medical University Email: chen_xin1020@163.com Address: 253 Gongye Road, Guangzhou 510282, China Clinical trial register: NCT02326181
Notes

ZhouL 2016.

Study characteristics
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 22/22 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 66.4 (5.5) Gender (M/F): 17/5 BMI, mean (SD), kg/m²: 22.5 (2.1) Control/sham N (randomized/analyzed): 22/22 Loss to follow‐up or excluded: 1 Age, mean (SD) in years: 66.8 (6.2) Gender (M/F): 16/6 BMI, mean (SD), kg/m²: 21.8 (2.2) Overall N (randomized/analyzed): 44/44 Loss to follow‐up or excluded: 2 Gender (M/F): 33/11 BMI, mean (SD), kg/m² COPD stage (GOLD): severe to very severe Included criteria Age: 40‐80 years Severe or extremely severe COPD in lung functions Combined with chronic respiratory failure (under normal inhalation, blood gas analysis PCO2 ≥ 50 mmHg after 1 h of rest No acute occurrence of exacerbating episode in the past 4 weeks Excluded criteria People who smoke > 10 cigarettes/d Patients with unstable cardiac hemodynamics, such as acute left heart failure, unstable angina Combined with other respiratory diseases, such as typical bronchiectasis, typical pulmonary fibrosis, sleep apnea, lung tumors, sequelae of tuberculosis (damage to the lung) Those suffering from neuromuscular diseases or sequelae of severe cerebrovascular accidents
Interventions	Intervention characteristics IMT: participants trained twice a day, without supervision, 30 min/d, 6 d/week for 8 weeks using Threshold IMT device set at 60% of PImax. This group also received non‐invasive positive pressure ventilation Control: this group received long‐term oxygen therapy
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD Respiratory muscle strength: PImax
Identification	Sponsorship source: National Natural Science Foundation of China (81361128004); Public welfare research and Capacity building special fund project (2014A020215033); Guangzhou Medical University Scientific Research Fund (2014c22) Country: China Setting: Guangzhou Institute of Respiratory diseases Comments: Author's name: Chen Rongchang Institution: Frist Affiliated Hospital of Guangzhou Medical University Email: member@wc.rf.org Address: Guangzhou 510120, China Clinical trial register: 0192675
Notes

6MWD: six‐minute walk distance; 12MWD: 12‐minute walk distance; ADL: activities of daily living; ATS: American Thoracic Society; BDI: Baseline Dyspnea Index; BMI: body mass index; CAT: Chronic Obstructive Pulmonary Disease Assessment Test; CCQ: Clinical COPD Questionnaire; CI: confidence interval; COPD: chronic obstructive pulmonary disease; CPET: cardiopulmonary exercise testing; CRQ: Chronic Respiratory Disease Questionnaire; ERS: European Respiratory Society; FEV1: forced expiratory volume at 1 second; FRC: functional residual capacity; FVC: forced vital capacity; GOLD: Global Initiative for Chronic Obstructive Lung Disease; HRQoL: health‐related quality of life; IMT: inspiratory muscle training; ISWT: incremental shuttle walk test; IVC: inspiratory vital capacity; MDP: Multidimensional Dyspnea Profile; MIP: maximal inspiratory pressure; mMRC: Modified Medical Research Council; MVV: maximal voluntary ventilation; PEmax: maximal expiratory pressure; PFT: pulmonary function test; PImax: maximal inspiratory pressure; PR: pulmonary rehabilitation; P_thmax: respiratory muscle endurance pressure; RCT: randomized controlled trial; RV: residual volume; SD: standard deviation; SE: standard error; SEM: standard error of the mean; SGRQ: St George's Respiratory Questionnaire; SpO₂: Peripheral oxygen saturation; SWT: shuttle walk test; TDI: Transition Dyspnea Index; VO₂max: maximal oxygen consumption; Wmax: maximum exercise workload

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Ahmad 2013	Ineligible intervention
Aldrich 1985	Ineligible study design
Anand 2013	Ineligible intervention
Baines 2005	Ineligible comparator
Basso Vanelli 2016	Ineligible comparator
Battaglia 2009	Ineligible intervention
Belman 1994	Ineligible comparator
Bgin 1991	Ineligible study design
Bissett 2016	Ineligible patient population
Bjerre Jepsen 1981	Ineligible intervention
Cader 2010	Ineligible patient population
Chen 1985	Ineligible intervention
Daynes 2018	Ineligible study design
de Andrade 2005	Ineligible study design
de Lucas Ramos 1998	Ineligible study design
Di Mambro 2007	Ineligible study design
DRKS00005637	Ineligible comparator
DRKS00006021	Cancelled Clinical trial
Elbouhy 2014	Ineligible patient population
Elmorsi 2016	Ineligible study design
Enright 2005	Ineligible comparator
Garcia 2008	Ineligible study design
Goldstein 1989	Ineligible intervention
Gregg 1989	Ineligible study design
Guyatt 1992	Ineligible intervention
Hart 2000	Ineligible patient population
Heydari 2015	Ineligible comparator
Hopp 1996	Ineligible study design
Ibakordor 2013	Ineligible comparator
Ionescu 2005	Non‐RCT
Izumizaki 2008	Ineligible study design
Johnson 1996	Ineligible study design
Kivastik 2015	Ineligible study design
Koch 2020	Ineligible study design
Kolesnikova 2016	Ineligible intervention
Levine 1986	Ineligible comparator
Liao 2015	Ineligible intervention
Lin 2012	Ineligible intervention
Lisboa 1994	Ineligible patient population
Lisboa 1995a	Ineligible study design
Lisboa 1995b	Ineligible patient population
Lisboa 1998	Ineligible study design
Madariaga 2007	Ineligible comparison
Madsen 1985	Ineligible comparator
Martin 2006	Ineligible intervention
McKeon 1986	Ineligible patient population
Meshcheriakova 2006	Ineligible intervention
Minoguchi 2002	Ineligible study design
NCT01218295	Ineligible study design
NCT01556139	Ineligible patient population
NCT01747694	Ineligible intervention
NCT01945398	Ineligible study design
NCT01956565	Ineligible study design
NCT02186340	Ineligible study design
NCT02278523	Ineligible comparator
NCT02579200	Ineligible patient population
NCT02914093	Ineligible comparator
NCT02935166	Ineligible study design
NCT03186092	Ineligible patient population
NCT03500042	Ineligible outcomes
NCT03739879	Ineligible study design
NCT03844711	Ineligible comparator
NCT03880630	Ineligible study design
NCT04084405	Ineligible comparator
NCT04117399	Ineligible outcomes
NCT04460261	Ineligible study design
Neves 2014a	Ineligible intervention
Neves 2014b	Ineligible study design
Nield 2007	Ineligible intervention
Noseda 1987	Ineligible comparator
O'Connor 2019	Ineligible study design
Okura 2019	Ineligible study design
Okura 2020	Ineligible study design
PACTR201703002095224	Ineligible comparator
Padula 2001	Ineligible patient population
Perez 2010	Ineligible intervention
Pescaru 2016	Ineligible study design
Quintero 1999	Ineligible study design
Richardson 1989	Ineligible study design
Rocha 2015	Ineligible intervention
Sassoon 1992	Ineligible intervention
Serón 2005	Ineligible patient population
Shahin 2008	Ineligible study design
Shioya 2007	Ineligible intervention
Similowski 1994	Ineligible study design
Sivashanmugam 2019	Ineligible comparator
Soicher 1998	Ineligible study design
Sonne 1982	Ineligible study design
Sudo 1997	Ineligible intervention
Sugiyama 2010	Ineligible study design
Sun 2003	Ineligible intervention
TCTR20191009004	Ineligible intervention
UMIN000030937	Ineligible intervention
Van't Hul 2006	Ineligible intervention
Villafranca 1998	Ineligible study design
Wada 2016	Ineligible intervention
Wu 2006	Ineligible intervention
Xi 2015	Ineligible intervention
Yamaguti 2012	Ineligible intervention
Yan 1996	Ineligible intervention
Yang 2005	Ineligible intervention
Zhang 2008	Ineligible study design

Characteristics of studies awaiting classification [ordered by study ID]

Barter 1987.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Bustamante 1997.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Cassidy 2009.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Gender (M/F): 7/7 Control/sham N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Gender (M/F): 5/9 Overall N (randomized/analyzed):  28/28 Loss to follow‐up or excluded: 0 Gender (M/F): 12/16 COPD stage: moderate to severe Included criteria: patients with COPD, following an acute exacerbation Excluded criteria: not reported
Interventions	Intervention characteristics IMT: participants underwent 8 weeks of unsupervised IMT using a Threshold IMT device Control/sham: participants received sham IMT
Outcomes	Respiratory muscle strength: PImax Functional exercise capacity: 6MWD HRQoL: CRQ
Notes	Country: Ireland Setting: home‐based training Author's name: C.Cassidy Institution: Respiratory Assessment Unit, CREST Directorate Address: St. James’s Hospital, Dublin 8, Ireland

Cejudo 1998.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Chen 2017.

Methods	Study design: RCT Study grouping: parallel‐group Clinical trial register: NCT02200549
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 24/not reported PR N (randomized/analyzed): 25/not reported Overall N (randomized/analyzed): 49 not reported Included criteria: not reported Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: PR: 8 weeks of cycle ergometer trainning. IMT: 8 weeks of IMT PR: training with a cycle ergometer for 8 weeks
Outcomes	Dyspnea Functional exercise capacity HRQoL: CRQ Respiratory muscle strength
Notes	Country: China Author's name: X.Chen Institution: Zhujiang Hospital‐ Southern Medical University, Department of Respiratory Medicine Address: Guangzhou, China

Croitoru 2013.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Age mean (SD) in years: 63.4 (8) PR N (randomized/analyzed): 14/14 Loss to follow‐up or excluded: 0 Age mean (SD) in years: 60.3 (11) Overall N (randomized/analyzed): 28/28 Loss to follow‐up or excluded: 0 COPD stage: moderate to severe Included criteria: not reported Excluded criteria: not reported
Interventions	Intervention characteristics PR+IMT: PR:  outpatient, 8 weeks, lower and upper limbs training, education, and psychological support IMT: daily, at home, with Threshold device, 30 min/d PR: participants received the same PR protocol described above.
Outcomes	Functional exercise capacity: 6MWD HRQoL: SGRQ Respiratory muscle strength: PImax
Notes

Del Castillo Otero 1998.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Di Marzo 2000.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Di Marzo 2002.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Downes Vogel 2002.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Eastwood 2005.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Gething 2001.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Göhl 2006.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

IRCT201104266299N1.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall N (randomized/analyzed): 30 Included criteria Patients with mild to very severe COPD Having an established treatment plan BMI < 35 Absence of other diseases such as neurological disorders, musculoskeletal system, peripheral vascular disease, and independence to the long‐term oxygen therapy Exclusion criteria Changes in the treatment plan during the study.
Interventions	Intervention characteristics IMT: participants trained at home daily for 21 min, 6 d/week for 8 weeks Control/sham: no intervention
Outcomes	Dyspnea HRQoL: SGRQ
Notes

IRCT20180205038633N1.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 15/unknown PR N (randomized/analyzed): 15/unknown IMT N (randomized/analyzed): 16/unknown Control/sham: N (randomized/analyzed): 15/unknown Overall N (randomized/analyzed): 61/unknown COPD stage: moderate to severe Inclusion/exclusion criteria Willing to participate in the study Grade 2 or 3 COPD based on GOLD criteria Age 40‐70 years; Lack of other pulmonary diseases Lack of severe limb limbs Lack of pulmonary surgery in the last 12 months Recent fracture failure at the level of ribs (6 months) No history of psychotropic disease and related drugs and alcohol and psychotropic substances
Interventions	Intervention characteristics IMT: the intensity of training was 40%‐60% (S‐Index 40%‐60%), 2 d/week, and 5 sessions of dermal muscle training with repeat 50 (5 Repeat 10) for about 15 min/session.  PR: aerobic exercise method: with a treadmill and foot pedometer, 2 times a week, with 40%‐60% heart rate reserve for 40 min/session PR+IMT: in each session, breathing exercises were performed first and then aerobic exercises were performed on the lower extremities Control: this group did not take any special intervention other than the usual treatments (control group)
Outcomes	Dyspnea: mMRC Respiratory muscle strength: PImax  Functional exercise capacity: 6MWD Respiratory function: FEV1
Notes

ISRCTN19258620.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria 80 non‐hypercapnic patients with moderate (FEV < 40%) recruited from consultant hospital and community chest clinics All patients will be receiving optimum medical management and will have been stable for at least 4 weeks prior to their initial assessment Exclusion criteria Hypercapnia (PaCO2 > 45 mmHg) Any patient who is unsuitable for magnetic stimulation (pacemakers, artificial heart valves, metal prosthesis)
Interventions	IMT: using Powerbreathe device
Outcomes	Dyspnea: Borg Functional exercise capacity: SWT Respiratory muscle strength: PImax Respiratory muscle endurance pressure: Pthmax Respiratory muscle endurance time: Tlim
Notes

Jones 1985.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall N (randomized/analyzed): 21 Inclusion criteria: FEV1 < 1.2 L
Interventions	Intervention characteristics IMT: training with resistive device for 10 weeks Control/sham: placebo training group
Outcomes	Functional exercise capacity: Wmax Functional exercise capacity: 12MWD
Notes

Koppers 2004.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Liu 1989.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Manuel Vargas 1995.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall N (randomized/analyzed): 23
Interventions	Intervention characteristics IMT: participants trained with a Threshold device at 30% of PImax Control/sham: no training was provided
Outcomes	Functional exercise capacity
Notes

Mendoza 2007.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes	Unfound

Meshcherykova 2018.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics IMT N (randomized/analyzed): not reported/17 Age mean (SD) in years: 62.5 (7.5) Control/sham N (randomized/analyzed): not reported/11 Age mean (SD) in years: 63.5 (9.8) Overall N (randomized/analyzed): not reported/ 28 COPD stage: severe Included criteria: patients with severe COPD
Interventions	Intervention characteristics IMT: participants received IMT for 3 months Control/sham: training with no load
Outcomes	Dyspnea: BDI‐TDI Functional exercise capacity: 6MWD Respiratory function: FEV1 Respiratory function: RV
Notes

NCT01056081.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall: N (randomized/analyzed): 19/unknown Inclusion criteria Patients with a clinical and spirometric diagnosis of moderate to severe COPD according to GOLD  In a stable condition (without exacerbations or infections for at least a month) Had to be former smokers (> 6 months without smoking) Patients referred by a physician to the pulmonary rehabilitation program Exclusion criteria Patients with a known history of asthma, or severe and/or unstable heart disease or any other pathological condition that could impair their physical activities
Interventions	Intervention characteristics PR+IMT PR: no information IMT: participants trained with a Threshold inspiratory muscle trainer (Respironics HealthScan, Inc, Cedar Grove, New York, USA). The participants performed the IMT training in a seated position, with the upper limbs supported. The total duration of the respiratory training was 30 min, with sequences of 3 min of training followed by pauses of 2 min. The initial load was equivalent to 30% of the individual's MIP. This load was progressively increased over the first 4 weeks, according to the participant's tolerance, to reach 60% of the MIP. This level was then maintained until the end of the training. PR: this group received only PR
Outcomes	Dyspnea Functional exercise capacity: exercise time Respiratory muscle strength: PImax
Notes

NCT01903772.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion Criteria: Patients with COPD Eligible to participate in an inpatient pulmonary rehabilitation program of 3 weeks PImax <60 cmH20 or <50% of the predicted normal value Exclusion Criteria: Major comorbidities preventing successful participation in an 8‐week exercise training intervention use of non‐invasive ventilation
Interventions	Intervention characteristics IMT: Inspiratory Muscle Training Three times daily inspiratory muscle training (2x30 breaths) at an intensity of >50% PImax Sham IMT: Twice daily inspiratory muscle training (3x30 breaths) at an intensity of 5 centimeters of water (H2O)
Outcomes	Dyspnea Functional exercise capacity Respiratory muscle strength: PImax Inspiratory muscle endurance capacity
Notes	It is unclear whether the study was completed, and no contact details were found.

NCT02392715.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria COPD demonstrated by spirometry using GOLD criteria Patient referred by a pneumologist to the ambulatory PR program at the Riviera‐Chablais Hospital, Monthey Patient with maximal inspiratory pressure < 60 cmH20 Patient > 40 years Exclusion criteria Orthopaedic or neurological troubles that could slant the 6MWD
Interventions	Intervention characteristics PR+IMT: PR:  general exercise training IMT: participants trained with a Threshold device 3 times/week with a total of 36 sessions. The training load was increased from 15%‐60% of PImax. PR: (+sham IMT): this group received general exercise training with a sham IMT set at 5 cmH2O.
Outcomes	Functional exercise capacity: 6MWD HRQoL: SGRQ Inspiratory muscle strength (PImax)
Notes

NCT02673242.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria COPD grade 3‐4 Inhabitant of Hedmark (Løten, Våler, Åsnes, Hamar, Elverum) Exclusion criteria Undergoing exercise‐based physical therapy treatment Not able to do IMT physically or mentally
Interventions	Intervention characteristics IMT: using a Threshold device for 6 weeks Control/sham: participants received sham training or another intervention
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD HRQoL: CAT
Notes

NCT03080662.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria To be eligible for the PR program participants must have stable COPD (at least 4 weeks), inspiratory muscle weakness (PImax < 70%) and pulmonary hyperinflation (TLC > 120%). Patients that have signed informed consent Exclusion criteria Hospitalization within the previous 14 days Current participation in a rehabilitation program Locomotor or neurological condition or disability limiting the ability to perform exercise Lung transplantation or lung volume reduction surgery foreseen within 1 month after discharge
Interventions	Intervention characteristics IMT: training with ORYGEN DUAL Sham Valve for 5 weeks Control/sham: training with the same protocol with no load
Outcomes
Notes

NCT03438019.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria Male or female aged > 40 years with a confirmed diagnosis of COPD by a pulmonologist (presence of risk factors and airflow obstruction) Evidence of inspiratory muscle weakness as defined by the latest ATS/ERS statement on respiratory muscle training Exclusion criteria Refusal to participate in the study Patients actively undergoing PR Inability to perform the required manoeuvres (i.e. patients with a cerebrovascular accident or tracheostomy) Patients not in their stable state (i.e. having an acute exacerbation or within 4 weeks of having one) or the presence of important comorbidities that may confound the interpretation of TIRE (test of incremental respiratory endurance) measures (i.e. decompensated heart failure, diaphragmatic paralysis, prior lung surgery, active cancer treatment, etc.).
Interventions	Intervention characteristics The TIRE IMT group:  will receive a tablet with the TIRE software installed and a PrO2 device through which they will train. Training consists of 6 levels (A‐F) with 6 inspirations at each level for a total of 36 breaths. Recovery times between breaths range from 40‐5 seconds as the participant advances each level. TIRE data will be stored in the tablet for subsequent interrogation and data retrieval. The standard IMT group: will receive a Threshold Inspiratory Muscle Trainer. This device incorporates a flow‐independent one‐way valve to ensure consistent resistance and features an adjustable specific pressure setting to be set based on MIP values of each participant. Participants will be instructed to perform up to 36 breaths daily. To compare with TIRE training, we will ask participants to perform this within a 30‐min session. The sham IMT group: will also receive a Threshold device and undergo the exact protocol of group 2 but with minimal resistance applied (7 cm H2O, the lowest in the device).
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD HRQoL:  SGRQ CAT Respiratory muscle strength: PImax
Notes

NCT03790410.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Inclusion criteria Stable COPD Do not take part in other treatment Be able to learn the usage of inspiratory muscle trainer Exclusion criteria Acute exacerbation Balance problems with a neurological cause The patient unable to cooperate Participation in pulmonary rehabilitation 3 months before the study IMT use in the 3 months before the study
Interventions	Intervention characteristics PR+IMT: PR consists of breathing exercises, bicycle ergometer conditionings, relaxations, and strength and endurance training. IMT consists of strengthening exercises on the diaphragm muscle. PR: This group undertook only PR.
Outcomes	Respiratory muscle strength: PImax
Notes

Newall 1998.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Newall 2000.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes	Unfound

NTR2990.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Pertuze 1994.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Ramirez Sarmiento 2000.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Reidi 2005.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall N (randomized/analyzed): 18/unknown Age mean (SD) in years: 65.1 (6.6) Inclusion criteria: FEV1 < 60%
Interventions	Intervention characteristics IMT: the training was performed through the Threshold set at 30 of PImax. The program consisted of 4 weeks, with 3 weekly sessions where both groups were reassessed weekly Control/sham: the training consisted of performing the threshold with or without the natural resistance of the equipment < 7 cmH2O
Outcomes	Respiratory muscle strength: PImax
Notes

Valderramas 2009.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Vargas 1995.

Methods	Study design: RCT Study grouping: parallel‐group
Participants
Interventions	Intervention characteristics IMT: participants used an inexpensive pressure threshold load valve constructed according to the Appropriate Technology principles of the WHO, adjusted at 30% of MIP for 3 months Control/sham: this group did not receive any intervention.
Outcomes	Respiratory muscle strength: PImax
Notes

Vargas 1998.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Wang 2004.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics Overall N (randomized/analyzed): 64
Interventions	Intervention characteristics IMT: twice a day for 6 months Control/sham: no intervention was received
Outcomes	Respiratory function: FEV1 (%pred)
Notes

Wanke 1994.

Methods	Study design: RCT Study grouping: parallel‐group
Participants	Baseline characteristics PR+IMT N (randomized/analyzed): 21/unknown PR N (randomized/analyzed): 21/unknown
Interventions	Intervention characteristics PR+IMT: PR: cycle ergometer training for 8 weeks IMT: for 8 weeks PR: cycle ergometer training alone
Outcomes	Functional exercise capacity
Notes

Weiner 2006a.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

Wolstenholme 1998.

Methods	Unfound
Participants	Unfound
Interventions	Unfound
Outcomes	Unfound
Notes

6MWD: six‐minute walk distance; 12MWD: 12‐minute walk distance; ATS: American Thoracic Society; BDI: Baseline Dyspnea Index; BMI: body mass index; CAT: COPD Assessment Test ; COPD: chronic obstructive pulmonary disease; CRQ: Chronic Respiratory Disease Questionnaire; ERS: European Respiratory Society; FEV1: forced expiratory volume at 1 second; GOLD: Global Initiative for Chronic Obstructive Lung Disease; HRQoL: health‐related quality of life; IMT: inspiratory muscle training; MIP: maximal inspiratory pressure; mMRC: Modified Medical Research Council; PImax: maximal inspiratory pressure; PR: pulmonary rehabilitation; RCT: randomized controlled trial; RV: residual volume; SGRQ: St George's Respiratory Questionnaire; SWT: shuttle walk test; TDI: Transition Dyspnea Index; TIRE: test of incremental respiratory endurance; WHO: World Health Organization; Wmax: maximum exercise workload

Characteristics of ongoing studies [ordered by study ID]

CTRI/2020/11/029226.

Study name	Effects of inspiratory muscle training vs autogenic drainage in hospitalised COPD patients
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Diagnosed case of COPD from 4‐10 years Patients of both genders aged between 50‐70 years Smokers and non‐smokers Clinically stable patients Exclusion criteria Resting SBP > 200 mm Hg or DBP > 110 mm Hg Pre‐hypertension and hypotension Diagnosed cases of orthopnea Diagnosed cases of chronic kidney disease Injured musculoskeletal or fractured since 3 months.
Interventions	IMT: conducted for 30 min with a total of 10 sessions Control: this group did not receive any intervention
Outcomes	HRQoL: SGRQ Pulmonary function tests
Starting date	20 November 2020
Contact information	Sponsorship source: KLE Academy of Higher Education and Research Country:  India Author's name: Varun Naik  Institution: Physiotherapy Department Nehru Nagar Belgaum Belgaum  Email: drvarunnaik@gmail.com Address: KARNATAKA 590010 Clinical trial register: CTRI/2020/11/029226
Notes

CTRI/2021/05/033469.

Study name	To test the efficacy of inspiratory muscle training device Airofit in reducing breathlessness in COPD patients
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria All patients with spirometry‐proven, stable COPD (GOLD stage ≥ 2) screened for inclusion Patients with MIP < 100% of predicted Patients aged 35‐65 years Exclusion criteria Patients without access to a suitable smartphone or tablet for the duration of intervention Patients with an inability to read and understand written and verbal instructions in English Patients with a history of hospitalization during the previous 4 weeks Patients with severe orthopedic problems during the previous 4 weeks Patients with diagnosed psychiatric or cognitive disorders Patients with a progressive neurological or neuromuscular disorder Patients on the waiting list for lung transplantation Patients with previous inclusion in a rehabilitation program < 1 year Patients with previous experience with IMT Patients with underlying bronchiectasis, innate lymphoid cell, post‐COVID and post‐tuberculosis sequelae All COPD patients with active COVID status or within 15 days of post‐COVID recovery
Interventions	Intervention characteristics IMT (Airofit device group): the Airofit is a new IMT device, which combines a calibrated inspiratory flow resistance with a pressure transducer and mobile device app. As the user inhales through the device, a pressure load is created; the magnitude of the load is proportional to the rate of air flow, as well as to the flow resistance properties of the load setting. The former has been identified as a limitation of flow‐resistive IMT. To overcome the flow‐dependency of the training load, the Airofit uses a pressure transducer to communicate with bespoke software, in real time. The measurement of pressure provides the user with instantaneous visual feedback of the pressure created by their inspiratory muscles, i.e. their training effort. The software also provides the user with a personalized visual training intensity target (50% of MIP). A total of 19 patients with a spirometry‐proven stable COPD will be enrolled in the experimental arm.  IMT (Powerbreathe device group): 19 participants will train at a load of 50% of PImax. Sham IMT: the participants will use control device ‘Breathing pacer’
Outcomes	Inspiratory muscle strength (PImax) Adverse events
Starting date	1 June 2021
Contact information	Sponsorship source:  Airofit AS Copenhagen, Denmark Country: India Author's name: Atul Deshmukh Institution: Padmashree Dr. D Y Patil University Navi Mumbai  Email: atul.deshmukh@dypatil.edu   Address: Centre for Interdisciplinary Research, Ground floor, Central Research Facility, Near Simulation Laboratory Padmashree Dr. D Y Patil Univerisity Campus Sector 7 Nerul Navi Mumbai, MAHARASHTRA, 400706, India
Notes

CTRI201712010952.

Study name	Effect of breathing exercises to improve the strength of respiratory muscles in COPD population
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria: diagnosed COPD patients Exclusion criteria: women with COPD, those unable to comprehend
Interventions	Intervention : IMT+PR Control: PR
Outcomes	6MWD Respiratory muscle strength (PImax) CRQ
Starting date	10 April 2017
Contact information	veenakiran_nambiar@yahoo.co.in
Notes

De Souza 2019.

Study name	Does inspiratory muscle training (IMT) reduce depression in patients with COPD?
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Patients with COPD presenting with inspiratory muscle weakness
Interventions	Intervention characteristics IMT: participants trained at 50% of PImax, 2 sessions/d of 30 breaths for 8 weeks, with a weekly face‐to‐face session Control/sham: participants trained at 10% of PImax with a similar protocol
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD Respiratory muscle strength: PImax
Starting date
Contact information	souzayr@gmail.com
Notes

Formiga 2020.

Study name	Novel versus traditional inspiratory muscle training regimens as home‐based, stand‐alone therapies in COPD: protocol for a randomized controlled trial
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Patients > 40 years with a clinical and functional diagnosis of COPD according to GOLD guidelines – stages 1‐4 Evidence of inspiratory muscle weakness, defined as a MIP ≤ 80 cmH2O and a SMIP ≤ 427 pressure‐time units The ability to operate a computer, tablet or smartphone independently and follow the training instructions Clinical stability with no history of infections or exacerbation of respiratory symptoms for at least 2 months prior to study enrollment Non‐participation in exercise programs in the past 12 months Exclusion criteria History of lung surgery, lung cancer, as well as individuals with any diagnosed cognitive (i.e. Mini Mental State Examination score < 24), orthopedic, neurological or neuromuscular disorders that might prevent them from appropriately performing the required physical tests and/or completing the study questionnaires Patients will not be excluded based upon their current bronchodilator regimen. If they experience acute exacerbations or respiratory infections during the training period, they will be examined by a pneumologist who will decide whether the participant should continue with the training or not.
Interventions	Intervention characteristics All groups will train once a day for 8 weeks TIRE: participants will use an on‐screen training template set at 50% of their MIP and SMIP. The training will consist of 6 levels (A‐F) with 6 inspirations per level for up to 36 efforts per session. Pre‐set recovery times between breaths: 60 seconds at level A to 50, 40, 30, 20 and 10 seconds at levels B to F, respectively Threshold IMT: participants will train using a one‐way spring‐loaded valve set at 50% of their MIP. The training will consist of 36 inspirations performed using the device within a 30‐min period. Sham‐IMT: participants will use a one‐way spring‐loaded valve set to its minimal resistance (−9 cmH2O). The training will consist of 36 inspirations performed using the device within a 30‐min period.
Outcomes	Dyspnea: mMRC Respiratory muscle strength: PImax  HRQoL: CAT Functional exercise capacity: 6MWD Respiratory muscle strength (PImax) Pulmonary function test
Starting date	1 May 2021
Contact information	dosbaba.filip@fnbrno.cz
Notes	Clinical tiral register: NCT04415788

JPRN‐UMIN000039893.

Study name	The effectiveness of inspiratory muscle training in patients with chronic obstructive pulmonary disease
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Age minimum: 20 years Age maximum: not applicable Gender: male Exclusion criteria Participants have no severe and/or unstable cardiac disease, orthopedic disease or mental disorder
Interventions	Intervention characteristics PR+IMT: PR:  IMT: 30%‐50% of PImax intensity, 30 breaths/session, 2 sessions/d, every day for 3 months PR +sham IMT: this group will undergo the same rehabilitation program with an IMT load set at 10% of PImax
Outcomes	Functional exercise capacity: 6MWD HRQoL: CAT Inspiratory muscle strength (PImax)
Starting date	01/11/2016
Contact information	Sponsorship source: Akita University Graduate School of Health Sciences Country: Japan Author's name: Takanobu Shioya Institution: Akita University Graduate School of Health Sciences Department of Physical Therapy Email: shioya@hos.akita‐u.ac.jp Address: 1‐1‐1, Hondo, Akita Japan Clinical trial register: JPRN‐UMIN000039893
Notes

JPRN‐UMIN000043099.

Study name	Effect of inspiratory muscle training on diaphragm and exercise tolerance in patients with COPD
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Age minimum: 65 years Age maximum: 85 years Gender: male and female Exclusion criteria The exclusion criteria include unstable medical conditions that cause or contribute to breathlessness (i.e. metabolic, cardiovascular, or other respiratory diseases) or any other disorders that interfere with exercise testing, such as neuromuscular diseases or musculoskeletal problems
Interventions	Intervention characteristics IMT: IMT will be performed for 12 weeks Control: this group will not receive any intervention
Outcomes	Exercise capacity
Starting date	22 January 2021
Contact information	Sponsorship source: Kindai University Hospital Country: Japan Author's name: Masashi Shiraishi Institution: Kindai University Hospital Department of Rehabilitation Email: masashi‐shiraishi@med.kindai.ac.jp Address: 377‐2 Onohigashi, Osakasayama‐city 589‐8511 Japan Clinical trial register: JPRN‐UMIN000043099
Notes

NCT04120142.

Study name	Effect of inspiratory muscle training during PR on dyspnoea and exercise tolerance in COPD patients
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria COPD diagnosed by pulmonary function testing Clinically stable Absence of other obstructive diseases Signed written consent Exclusion criteria Were previous pneumonectomy or lobectomy in the past 6 months Spontaneous risk of pneumothorax or rib fracture Incapacity to follow a standard rehabilitation programme (locomotor deficits, acute cardiac failure and acute exacerbation of COPD at the beginning of the program) The absence of written informed consent
Interventions	Intervention characteristics PR+IMT: IMT + aerobic exercise  PR: aerobic exercise alone
Outcomes	Dyspnea Respiratory muscle strength (PImax)
Starting date	1 February 2019
Contact information
Notes

NCT04201522.

Study name	The effect of respiratory training on exercise tolerance in COPD (ERTET)
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Age ≥ 40 years Chronic airflow obstruction: FEV1/FVC < 0.7, FEV1 of 30%‐80%pred, after bronchodilation Exclusion criteria Inability to perform a cycling exercise Diagnosed of ≥ 1 comorbidities that may limit exercise tolerance: cardiovascular, metabolic, endocrine, gastrointestinal, renal, neurological or rheumatological disease Recent COPD exacerbation (< 3 months) Recent cancer A daily dose of Prednisone > 10 mg Hypoxemia at rest or during exercise: PaO2 < 60 mmHg or SpO2 ≤ 88% BMI > 30 kg/m² Pregnancy Skinfold at intercostal or vastus lateralis muscle > 1.5 cm
Interventions	Intervention characteristics IMT: participants will train for 6 weeks, 15 min twice daily, 5 d/week at 60% of the peak of minute ventilation, at home by means of a respiratory device (SpiroTiger, Idiag, Fehraltorf, CH) Control/sham: participants will undergo the training protocol at rest's minute ventilation
Outcomes	Respiratory muscle strength (PImax)
Starting date	14 March 2017
Contact information	ferid.oueslati@criucpq.ulaval.ca
Notes

NCT04387318.

Study name	Inspiratory muscle training and neuromuscular electrical stimulation in chronic obstructive pulmonary disease
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria Clinical diagnosis of COPD, in stages 2, 3 and 4 according to GOLD criteria Clinically stable, i.e. absence of infections or exacerbations in the last 3 months The medical team allows the patient to exercise Availability of attending the rehabilitation program Exclusion criteria Unstable primary pathologies (cardiovascular, renal, metabolic, psychiatric) Hemodynamic instability Nutritional supplementation in the 4 weeks preceding the study Severe hearing or a visual impairment recorded on patient chart or self‐referred Obesity (BMI > 30 kg/m2) A neurological or musculoskeletal condition that severely limits mobility and postural control, thus making it impossible to carry out the assessments Electronic devices, such as heart pacemakers and implantable cardioverter‐defibrillator Skin injuries and infection where electrodes would be placed Prior participation in pulmonary rehabilitation programs 3 months previous to the study Vertigo
Interventions	PR+IMT: PR: aerobic and resistance exercise for 8 weeks IMT: participants will train with POWERbreathe Medic Plus (POWERbreathe International Ltd., England, UK) inspiratory training device for 5 sets of 10 repetitions each, with a 1‐minute interval between each set. The initial load set will be 30% of PImax, during the first 2 weeks to allow for an adjustment period. After that, load increases occurred as follows: 35% of PImax in week 3, 40% of PImax in week 4, 45% of PImax in week 5, 50% of PImax at week 6, 55% of PImax in week 7, and 60% of PImax in week 8 PR: this group will undergo only the PR protocol described above
Outcomes	HRQoL: SGRQ Respiratory muscle strength (PImax) Respiratory muscle endurance
Starting date	1 October 2019
Contact information	albuisa@gmail.com
Notes

NCT04802096.

Study name	Effects of inspiratory muscle training in addition to pulmonary rehabilitation in patients with moderate to severe COPD exacerbation
Methods	Study design: RCT Study grouping:parallel‐group
Participants	Inclusion criteria Age > 20 years old Diagnosed as moderate exacerbation of COPD MIP < 80 cmH2O Exclusion criteria Any clinical diagnosis that will influence the measurement, including any history of neuromyopathy Angina, acute myocardial infarction in the previous  month Pregnancy Participated in IMT program in the previous 3 months Any psychiatric or cognitive disorders, e.g. Mini‐Mental State Examination < 24, that will disturb the communication and co‐operation of the study
Interventions	PR+IMT: PR: aerobic exercise training, strength training, and education related to airway clearance and drug utilization. All the participants will receive 8‐week PR IMT: the intensity of IMT will be set at 30% of MIP. Participantsin this group will perform 15 breaths/set, 6 sets/d on a daily basis. The intervention will last for 8 weeks PR: this group will undergo the same PR program as the intervention group
Outcomes	Functional exercise capacity: 6MWD HRQoL: SGRQ HRQoL: CAT
Starting date	15 March 2021
Contact information	Sponsorship source: National Taiwan University Hospital Country: Taiwan  Author's name: Wei‐Yu Huang Institution: School and Graduate Institute of Physical Therapy of National Taiwan University Email:  r08428013@ntu.edu.tw Address: Taipei, Zhongzheng Dist, Taiwan, 100 Clinical trial register: NCT04802096
Notes

RBR‐10nyzcqc.

Study name	Effects of Inspiratory Muscle Training on Breathlesness, Exercise Capacity and Postural Control in Patients with COPD
Methods	Study design: RCT Study grouping:parallel‐group
Participants	Inclusion criteria Age> 40 years  Diagnosis of COPD according to GOLD (2016) Absence of exacerbations in the last 30 days  Presence of inspiratory muscle weakness (PImax <70cmH2O or <70% of predicted) Exclusion criteria Diagnosis of psychiatric or cognitive diseases that compromise the understanding of the study's guidelines Progressive neurological disease  Neuromuscular disease; Vestibular disorders Orthopedic changes that compromise the results of field and range tests Presence of other comorbidities that, at the researcher's discretion, may interfere with the results of the study
Interventions	IMT: 22 patients will train at around 50% of their PImax. The training consists of 30 breaths in a row, with deep and strong inspiration (lasting 4‐5 minutes), twice a day, every day for 8 weeks. Each week, the patient will return to our center to perform a new manovacuometry and update the load value, maintaining the ~ 50% of the PImax value, and after this update, one of the sessions of the day will be held in person. Sham IMT: 16 patients will train at around 10% of their PImax. The training consists of 30 breaths in a row, with deep and strong inspiration (lasting 4‐5 minutes), twice a day, every day for 8 weeks. Every week, the patient will return to our center to perform a new manovacuometry, the data will be recorded, however the load will be kept at around 10% of the initial PImax value, after this registration, one of the sessions of the day will be held in person. This value will not make any changes during the study period. After the eight‐week protocol, patients will return to perform the initial evaluations and tests again, in the same order they did the first time, and will receive a new medical order for serum vitamin D and calcium measurement for comparison with the first. The CG participants after the study period will undergo the same protocol applied to the IG as treatment.
Outcomes	Dyspnea: mMRC Functional exercise capacity: 6MWD
Starting date
Contact information	yves.souza@uva.br
Notes

RBR 42rmqy.

Study name	Inspiratory muscle training in chronic obstructive pulmonary disease oxygen‐dependent patients: a randomized controlled trial
Methods	Study design: RCT Study grouping: parallel‐group
Participants	IMT N (randomized): 30 Control/sham N (randomized): 30   Inclusion criteria Individuals with COPD Both genders Stable clinical condition Without cardiac disease (heart failure, angina pectoris) No history of respiratory infection at least 30 days before inclusion of study Exclusion criteria Inability to walk  Inability to do 6MWD Neuromuscular disease Non‐controlled comorbidies such as arterial hypertension and diabetes mellitus
Interventions	Intervention characteristics IMT: the experimental group will have 30 individuals that will receive home‐based IMT, 7 d/week, over 4 weeks. The daily training will be split into 2 daily sessions (i.e. morning and afternoon). Each daily session will be comprised of four 4‐min sets of respiratory training with 1‐min rest intervals between each set. The resistance will be provided by the Power Breathe Classic, which allows individuals to exercise the inspiratory muscles during the training session. Training will be individually tailored for each participant. The initial training load for each participant will be set at 50% of his/her PImax. The participants will be trained and instructed to do the exercise program on their own with no supervision. Once a week, during the physical therapist’s home visit, the clinician will determine the new values for maximal inspiratory strength, and will adjust the load to 50% of the new values. The device will be covered with opaque material, so that participants will be blinded to the training load. Control/sham: the control group with 30 individuals will receive a sham intervention. Sham respiratory muscle training will be delivered using a Power Breathe with no resistance (0 cmH2O) or progression, over the 4‐week period, 40 min/d, 7 d/week
Outcomes	HRQoL:  ADL Respiratory muscle strength (PImax) Laboratory exercise test: VO2peak
Starting date	30 October 2019
Contact information	viniciusmaldaner@gmail.com
Notes

TCTR20210604001.

Study name	Is single bout of inspiratory muscle training alter blood pressure and cardio autonomics modulation in COPD patients? : a pilot study
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Inclusion criteria GOLD stage 2‐3 COPD patients, which was defined by FEV1 30%‐80%pred Free of acute exacerbation of COPD > 8 weeks before the study BMI 18.5 kg/m2‐29.9 kg/m2 Exclusion criteria History of recent lung surgery, spontaneous pneumothorax or rib fracture < 12 months prior to the study History of stroke, diabetes, cardiovascular or neuromuscular disorder Resting blood pressure > 180/110 mmHg or < 90/60 mmHg Fracture at upper, lower extremity or spine < 6 months prior to the study Lower extremity and spine pain which was assessed by visual analog scale > 3  Regular exercise which was defined as having exercise or perform physical activity Unable to communicate and follow instructions Resting supplemental oxygen Age minimum: 40 years Age maximum: 70 years Gender: male
Interventions	Intervention characteristics IMT: participants will perform single session of IMT exercise with 60% of MIP. They will complete 6 breaths/set for 5 sets with 1‐min rest between each set Sham IMT: participants will undergo the same program described above at no load
Outcomes
Starting date	11 January 2021
Contact information	Sponsorship source: Khon Kaen University Khon Kaen, Thailand Country: Thailand Author's name: EAKARACH WONGSAYA Institution: School of Physical Therapy, Faculty of Associated Medical Sciences, Khon Kaen University Email: eakarach.wo@up.ac.th Address: 123 School of Physical Therapy, Faculty of Associated Medical Sciences, Khon Kaen University University 40002 Khon Kaen Thailand Clinical trial register: TCTR20210604001
Notes

Tctr 2022.

Study name	Effect of inspiratory muscle training on cardiovascular autonomic functions in chronic obstructive pulmonary disease patients
Methods	Study design: RCT Study grouping: parallel‐group
Participants	Patients with COPD
Interventions	IMT Home‐based IMT will be performed daily using PowerBreathe device. Total eight week of IMT will be provided. The IMT group will start training at 40% of their initial PImax. The new PImax value will be measured every week by the researcher. The training load in each week will be increased continuously over time by adjusting to at least 50% of PImax or gradually increase to the highest tolerable intensity during each of the supervised sessions. The highest tolerable intensity means that the rates of perceived inspiratory effort using a modified Borg dyspnea scale will be the range 4 to 6 of 10 (moderate to very severe). Daily training will consist of 2 sessions of 30 breaths.,The participants will receive only the standard care treatment from the physicians such as pharmacological treatment. They will meet the physician at week 1 and week 8. Control This group will not receive any intervention
Outcomes	Dyspnea:  mMRC BDI‐TDI Functional exercise capacity: 6MWD Respiratory muscle strength: PImax Spirometry measurements
Starting date	2022
Contact information
Notes

6MWD: six‐minute walk distance; ADL: activities of daily living; BMI: body mass index; CAT: Chronic Obstructive Pulmonary Disease Assessment Test; COPD: chronic obstructive pulmonary disease; CRQ: Chronic respiratory disease questionnaire; DBP: diastolic blood pressure; FEV1: forced expiratory volume at 1 second; FVC: forced vital capacity; GOLD: Global Initiative of Chronic Obstructive Lung Disease; HRQoL: health‐related quality of life; IMT: inspiratory muscle training; MIP: maximal inspiratory pressure; mMRC: Modified Medical Research Council; PImax: maximal inspiratory pressure; PR: pulmonary rehabilitation; RCT: randomized controlled trial; SBP: systolic blood pressure; SGRQ: St George's Respiratory Questionnaire; SMIP: sustained maximal inspiratory pressure; TIRE: test of incremental respiratory endurance

Differences between protocol and review

Because few studies used the Borg scale and reported it at different levels of effort, we decided to replace it with PImax for subgroup analysis. The 6MWD and PImax are respectively clinical and physiological outcomes, and we believed that analysing both outcomes together would present an opportunity to check for the possible correlation between them.

Because many RCTs were included, we conducted a sensitivity analysis by keeping just the studies at an overall low risk of bias.

We conducted a sensitivity analysis by removing (Preusser 1994) because we noticed large differences in baseline data between the two groups.

We did not use Robvis to generate weighted bar plots because it was possible to create them with Revman Web, and finally we kept just the PRISMA flow diagram and funnel plots.

Contributions of authors

Conception of the review: OA/WF
Co‐ordination of the review: OA
Search and collection of studies for inclusion in the review: OA/WF
Data extraction: OA/AK
Assessment of the risk of bias in the included studies: OA/TL
Analysis of data: OA/AK
Assessment of certainty in the body of evidence: OA/TL
Interpretation of data: WF/SK
Writing the review: OA
Clinical and statistical comments: RG/AR
 

Contributions of editorial team

Sally Spencer (Co‐ordinating Editor) edited the review; advised on methodology, interpretation and content; and approved the review prior to publication.
Rebecca Fortescue (Co‐ordinating Editor) checked the data in the review.
Anne Holland (Contact Editor): edited the review; and advised on methodology, interpretation and content.
Emma Dennett (Deputy Co‐ordinating Editor): advised on methodology, interpretation and content; and edited the review.
Emma Jackson (Managing Editor): co‐ordinated the editorial process; conducted peer review; obtained translations; and edited the plain language summary and reference sections of the review. 
Elizabeth Stovold (Information Specialist): designed the search strategy; ran the searches; and edited the search methods section.

Vittoria Lutje (Freelance Information Specialist): ran the updated search.

Kayleigh Kew (Freelance Editor): edited the review and advised on methodology.

Sources of support

Internal sources

• University of Sfax, Faculty of Medicine, Tunisia

  Omar Ammous is supported by the University of Sfax, Faculty of Medicine, to conduct this systematic review as his medical thesis.

External sources

• National Institute for Health and Care Research (NIHR), UK

  Provision of Cochrane Infrastructure funding to Cochrane Airways

Declarations of interest

OA: none known
RG: none known
AK: none known
WF: none known
TL: none known
AR: none known
SK: none known

 

Edited (no change to conclusions)