Non‐invasive ventilation during exercise training for people with chronic obstructive pulmonary disease

Abstract

Background

Exercise training as a component of pulmonary rehabilitation improves health‐related quality of life (HRQL) and exercise capacity in people with chronic obstructive pulmonary disease (COPD). However, some individuals may have difficulty performing exercise at an adequate intensity. Non‐invasive ventilation (NIV) during exercise improves exercise capacity and dyspnoea during a single exercise session. Consequently, NIV during exercise training may allow individuals to exercise at a higher intensity, which could lead to greater improvement in exercise capacity, HRQL and physical activity.

Objectives

To determine whether NIV during exercise training (as part of pulmonary rehabilitation) affects exercise capacity, HRQL and physical activity in people with COPD compared with exercise training alone or exercise training with sham NIV.

Search methods

We searched the following databases between January 1987 and November 2013 inclusive: The Cochrane Airways Group specialised register of trials, AMED, CENTRAL, CINAHL, EMBASE, LILACS, MEDLINE, PEDro, PsycINFO and PubMed. 

Selection criteria

Randomised controlled trials that compared NIV during exercise training versus exercise training alone or exercise training with sham NIV in people with COPD were considered for inclusion in this review.

Data collection and analysis

Two review authors independently selected trials for inclusion in the review, extracted data and assessed risk of bias. Primary outcomes were exercise capacity, HRQL and physical activity; secondary outcomes were training intensity, physiological changes related to exercise training, dyspnoea, dropouts, adverse events and cost.

Main results

Six studies involving 126 participants who completed the study protocols were included. Most studies recruited participants with severe to very severe COPD (mean forced expiratory volume in one second (FEV₁) ranged from 26% to 48% predicted). There was an increase in percentage change peak and endurance exercise capacity with NIV during training (mean difference in peak exercise capacity 17%, 95% confidence interval (CI) 7% to 27%, 60 participants, low‐quality evidence; mean difference in endurance exercise capacity 59%, 95% CI 4% to 114%, 48 participants, low‐quality evidence). However, there was no clear evidence of a difference between interventions for all other measures of exercise capacity. The results for HRQL assessed using the St George's Respiratory Questionnaire do not rule out an effect of NIV (total score mean 2.5 points, 95% CI ‐2.3 to 7.2, 48 participants, moderate‐quality evidence). Physical activity was not assessed in any study. There was an increase in training intensity with NIV during training of 13% (95% CI 1% to 27%, 67 participants, moderate‐quality evidence), and isoload lactate was lower with NIV (mean difference ‐0.97 mmol/L, 95% CI ‐1.58mmol/L to ‐0.36 mmol/L, 37 participants, moderate‐quality evidence). The effect of NIV on dyspnoea or the number of dropouts between interventions was uncertain, although again results were imprecise. No adverse events and no information regarding cost were reported. Only one study blinded participants, whereas three studies used blinded assessors. Adequate allocation concealment was reported in four studies.

Authors' conclusions

The small number of included studies with small numbers of participants, as well as the high risk of bias within some of the included studies, limited our ability to draw strong evidence‐based conclusions. Although NIV during lower limb exercise training may allow people with COPD to exercise at a higher training intensity and to achieve a greater physiological training effect compared with exercise training alone or exercise training with sham NIV, the effect on exercise capacity is unclear. Some evidence suggests that NIV during exercise training improves the percentage change in peak and endurance exercise capacity; however, these findings are not consistent across other measures of exercise capacity. There is no clear evidence that HRQL is better or worse with NIV during training. It is currently unknown whether the demonstrated benefits of NIV during exercise training are clinically worthwhile or cost‐effective.

Plain language summary

Breathing support via a mask during exercise training for people with chronic obstructive pulmonary disease

Background: Quality of life and exercise tolerance are commonly reduced in people with chronic obstructive pulmonary disease (COPD). In addition, physical activity levels are lower compared with those of healthy people of a similar age. Exercise training as a part of a formal rehabilitation programme is an important component of management for people with COPD and has been shown to improve both quality of life and exercise tolerance. However, some individuals may have difficulty performing exercise at an adequate training intensity. Non‐invasive ventilation (NIV) is a method of providing breathing support using a machine called a ventilator. Breathing support is delivered via a mask that is worn over the nose, mouth or both, or via a mouthpiece. During a single exercise session, NIV has been shown to improve exercise tolerance and reduce breathlessness. Consequently, NIV used over multiple exercise sessions (during exercise training) may allow people with COPD to exercise at a higher intensity and potentially to achieve greater improvement in exercise tolerance, quality of life and physical activity.

Review question: We conducted a review to determine whether NIV during exercise training affects exercise tolerance, quality of life and physical activity compared with exercise training alone or exercise training with sham NIV (placebo) in people with COPD.

Study characteristics: The evidence is current to November 2013. We included six studies involving 126 participants who completed the study protocols. Most studies recruited participants with severe to very severe COPD. The average age of participants ranged from 63 to 71 years. Cycling or treadmill exercise training was performed in the studies. The duration of exercise training programmes ranged from six to twelve weeks.

Key results: The percentage change in peak exercise capacity increased by an average of 17% in three studies, and the percentage change in endurance exercise capacity by an average of 59% in two studies that provided NIV during training compared with training without NIV or training with sham NIV. However, these improvements in exercise capacity were not consistent findings as there was no clear evidence that NIV improved all other measures of exercise capacity. The results for quality of life were uncertain and our analysis did not exclude there being an effect with NIV during exercise training in two studies. Physical activity was not assessed in any of the studies. Non‐invasive ventilation allowed participants to exercise at a higher training intensity (average of 13% higher) in three studies, and evidence of a greater training effect on the muscles was found in two studies, as a marker in the blood (isoload blood lactate) was significantly lower by an average of 0.97 mmol/L. No information regarding adverse events or cost was reported. It is currently unknown whether demonstrated benefits of NIV during exercise training are clinically worthwhile or cost‐effective.

Quality of the evidence: This review was generally limited by the small number of included studies and the small numbers of participants within the included studies. The quality of the evidence was low for exercise capacity outcomes, largely because of issues with study design. Consequently, the effect of NIV during exercise training on exercise capacity is uncertain. The quality of the evidence for quality of life, training intensity and isoload blood lactate was moderate, and these findings can be interpreted with a greater degree of confidence.

Summary of findings

Summary of findings for the main comparison. Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation for people with chronic obstructive pulmonary disease.

Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation for people with chronic obstructive pulmonary disease
Patient or population: people with chronic obstructive pulmonary disease Settings: outpatient Intervention: non‐invasive ventilation during exercise training Comparison: exercise training alone or exercise training with sham non‐invasive ventilation
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No. of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Exercise training alone or exercise training with sham non‐invasive ventilation	Non‐invasive ventilation during exercise training
Exercise capacity: percentage change in peak work rate Incremental cycle or incremental treadmill test Follow‐up: 6 to 8 weeks	Exercise capacity: percentage change in peak work rate in the control groups ranged from a mean of 9% to 38%	Mean exercise capacity: percentage change in peak work rate in the intervention groups was 17% higher (7% to 27% higher)	17% (7% to 27%)	60 (3 studies)	⊕⊕⊝⊝ lowa
Exercise capacity: percentage change constant work rate endurance time Constant work rate cycle endurance test Follow‐up: 6 to 8 weeks	Exercise capacity: percentage change constant work rate endurance time in the control groups ranged from a mean of 74% to 88%	Mean exercise capacity: percentage change constant work rate endurance time in the intervention groups was 59% higher (4% to 114% higher)	59% (4% to 114%)	48 (2 studies)	⊕⊕⊝⊝ lowb,c	Mean change exceeds minimal important difference of 34%
Exercise capacity: endurance time (minutes) Constant work rate cycle endurance test Follow‐up: 6 to 8 weeks	Exercise capacity: endurance time (minutes) in the control groups ranged from a mean of 3.9 to 13.0 minutes	Mean exercise capacity: endurance time (minutes) in the intervention groups was 3.62 minutes higher (0.17 lower to 7.41 higher)	3.62 minutes (‐0.17 to 7.41 minutes)	48 (2 studies)	⊕⊕⊝⊝ lowb,d	CI crosses zero but does not rule out an effect
Health‐related quality of life Change in total score of St George's Respiratory Questionnaire. Scale from 0 to 100 Follow‐up: 6 to 8 weeks		Mean health‐related quality of life in the intervention groups was 2.45 points higher (2.3 lower to 7.2 higher)	2.45 points (‐2.3 to 7.2 points)	48 (2 studies)	⊕⊕⊕⊝ moderatee	CI crosses zero but does not rule out an effect
Physical activity: not measured	See comment	See comment	Not estimable	–	See comment	This outcome was not reported in any of the included studies
Training intensity: fInal training session (% baseline peak work capacity) Follow‐up: 6 to 8 weeks	Training intensity: change from baseline (%) in the control groups ranged from a mean of 75% to 93%	Mean training intensity: change from baseline (%) in the intervention groups was 13% higher (1% to 27% higher)	13% (1% to 27%)	67 (3 studies)	⊕⊕⊕⊝ moderatef	Heterogeneity between studies was explained by one study that recruited participants with milder disease compared with other studies in the analysis
Physiological outcomes: isoload blood lactate (mmol/L) Follow‐up: 6 to 12 weeks	Physiological outcomes: isoload blood lactate (mmol/L) in the control groups ranged from a mean of 2.50 to 2.61 mmol/L	Mean physiological outcomes: isoload blood lactate (mmol/L) in the intervention groups was 0.97 mmol/L lower (1.58 to 0.36 lower)	‐0.97 mmol/L (‐1.58 to ‐0.36 mmol/L)	37 (2 studies)	⊕⊕⊕⊝ moderateg
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (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.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

^a‐2 for risk of bias: None of the studies blinded participants or trainers, and only one study used a blinded assessor. It was unclear whether allocation concealment was adequate in two of the studies. Also, one study reported significant between‐group differences in baseline peak exercise capacity.
 ^b‐1 for risk of bias: One study did not blind participants or use a blinded assessor.
 ^c‐1 for imprecision: wide 95% confidence interval.
 ^d‐1 for imprecision: 95% confidence interval includes no effect, and upper confidence limit crosses the minimal important difference for benefit.
 ^e‐1 for risk of bias: Participants were not blinded in one study.
 ^f‐1 for risk of bias: Participants were not blinded in two of the studies, and trainers were not blinded in any of the studies.
 ^g‐1 for risk of bias: None of the studies blinded participants or trainers, which may have resulted in performance bias and could have indirectly affected this outcome.

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is a preventable but not curable disease that is generally progressive in nature (Viegi 2007). In 2010, COPD was one of the leading causes of mortality worldwide (Lozano 2012). Although variability between countries has been noted, it is estimated that the prevalence of COPD at GOLD (Global Initiative for Chronic Obstructive Lung Disease) stage II or higher (GOLD 2013) is 10.1% globally (Buist 2007). The economic and social costs of COPD are substantial (Pauwels 2004), and acute exacerbations of COPD that require admission to hospital are some of the largest contributors to direct healthcare costs (Viegi 2007). The number of years that people are living with disability due to COPD is also rising (Vos 2012).

Chronic obstructive pulmonary disease is characterised by expiratory flow limitation that is not fully reversible (O'Donnell 2006). In addition to pulmonary disease and dysfunction, COPD has a number of associated systemic manifestations including skeletal muscle dysfunction, weight loss and systemic inflammation (Agusti 2003). Dyspnoea is the hallmark symptom of COPD (Viegi 2007) and is more common in severe disease (Killian 1992). Dyspnoea can lead to a cycle of activity avoidance, deconditioning and reduced participation in society. Exercise capacity and health‐related quality of life (HRQL) are commonly reduced in people with COPD (Garrod 2006), and physical activity levels are lower than those of age‐matched healthy individuals (Pitta 2005).

Description of the intervention

Exercise training as a component of pulmonary rehabilitation is supported by high‐level evidence as one of the few effective interventions in the management of COPD (Rabe 2007; Ries 2007). Pulmonary rehabilitation has been shown to improve exercise capacity (Cambach 1999; Troosters 2000), HRQL and symptoms (Lacasse 2006), and to reduce the frequency of hospital admissions in those with a recent exacerbation (Puhan 2011). However, the effect of pulmonary rehabilitation on physical activity appears to be small (Ng 2012). High‐intensity exercise training may produce greater physiological improvement compared with lower‐intensity exercise training in people with COPD (Casaburi 1991; Gimenez 2000). However, some individuals may have difficulty performing exercise at an adequate intensity for the required duration (Maltais 1997) and may not achieve the same benefit from exercise training as those without a significant ventilatory limitation to exercise, particularly if peripheral muscle strength is relatively preserved (Garrod 2006; Plankeel 2005; Troosters 2001). Consequently, a number of adjuncts to exercise have been proposed, including non‐invasive ventilation (NIV), a type of breathing support delivered via a mask or mouthpiece.

How the intervention might work

In people with COPD, the use of NIV during a single session of lower limb exercise was shown in a systematic review (van't Hul 2002) to increase exercise endurance and reduce dyspnoea compared with exercise without NIV or exercise with sham NIV. Unloading of both inspiratory and expiratory components of the respiratory muscle pump has been observed with NIV during exercise (Kyroussis 2000), with the reduction in dyspnoea being proportional to respiratory muscle unloading (Maltais 1995). Improvement in pattern of breathing (Maltais 1995; van't Hul 2004) and in gas exchange (Dreher 2007; Hernandez 2001) was also noted. In addition, several extrapulmonary effects have been reported with NIV during exercise, including improved locomotor muscle perfusion (Borghi‐Silva 2008), decreased exercise‐induced lactic acidosis (Borghi‐Silva 2008; Polkey 2000) and associated reduction in symptoms of muscle fatigue (Bianchi 1998; Borghi‐Silva 2008).

Why it is important to do this review

Given the benefit of NIV during a single session of exercise, application of NIV over multiple sessions of exercise, that is, during exercise training, may allow people with COPD to exercise at a higher intensity for a greater duration. Therefore, exercise training with NIV could potentially lead to greater improvement in exercise capacity compared with exercise training alone. Such improvement in exercise capacity may also improve HRQL and increase physical activity levels in people with COPD.

Objectives

To determine whether NIV during exercise training (as part of pulmonary rehabilitation) affects exercise capacity, HRQL and physical activity in people with COPD compared with exercise training alone or exercise training with sham NIV.

Methods

Criteria for considering studies for this review

Types of studies

We included in this review randomised controlled trials (RCTs) comparing NIV during exercise training versus exercise training alone, or exercise training with sham NIV (control group). Randomised cross‐over trials were also considered for inclusion. Quasi‐RCTs, for example, those with alternate randomisation, were excluded.

Types of participants

Inclusion

We considered studies with participants with stable COPD for inclusion. Participants were considered to be stable if no history of an exacerbation was reported over the past month (Rabe 2007). The definition of COPD was based on:

1. a clinical diagnosis of COPD; and

2. a best recorded ratio of forced expiratory volume during one second (FEV₁) over forced vital capacity (FVC) < 70% and a best recorded FEV₁ < 80% predicted for individual study participants (equivalent to GOLD stage II to IV) (GOLD 2013).

Exclusion

We excluded studies that included participants with non‐COPD respiratory disease or participants with concomitant neuromuscular disease, a restrictive thoracic disorder, significant cardiac failure or cardiac disease if data from participants with COPD could not be analysed separately.

Types of interventions

Inclusion

The intervention for the active group consisted of the application of NIV (including bilevel, inspiratory pressure support and proportional assist ventilation) delivered via a mask or mouthpiece during all supervised exercise training sessions. The intervention for the control group was exercise training with or without sham NIV during all supervised exercise training sessions. Studies that involved the delivery of supplemental oxygen during exercise training in one group (e.g. exercise training with NIV and supplemental oxygen) were included provided that supplemental oxygen was also delivered to the alternative group (e.g. exercise training with supplemental oxygen). Similarly, studies that involved the use of nocturnal NIV were included only if both the actively treated group and the control group received nocturnal NIV. Training had to include lower limb and/or upper limb endurance exercise and had to comprise four or more weeks with a minimum of two supervised sessions per week.

Exclusion

Studies that used continuous positive airway pressure as the active treatment during exercise training were excluded.

Types of outcome measures

Primary outcomes

1. Exercise capacity (defined as peak exercise capacity, constant work rate (endurance) exercise capacity or functional exercise capacity measured post exercise training, without NIV).

2. Health‐related quality of life (measured using disease‐specific or generic HRQL instruments).

3. Physical activity: direct measurement (e.g. metabolic equivalents (METS), step count).

Secondary outcomes

1. Training intensity (e.g. peak training intensity, final session training intensity).

2. Physiological changes related to exercise training (e.g. blood lactate levels, minute ventilation).

3. Dyspnoea (e.g. Borg score, visual analogue scale score).

4. Dropouts.

5. Adverse events.

6. Cost.

Search methods for identification of studies

Electronic searches

We identified trials with assistance provided by the Cochrane Airways Group Trials Search Co‐ordinator using the Cochrane Airways Group Specialised Register of trials. This Register was derived from systematic searches of bibliographic databases including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Allied and Complementary Medicine Database (AMED) and PsycINFO, and from handsearching of respiratory journals and meeting abstracts, including annual meetings of the American Thoracic Society, the European Respiratory Society and the British Thoracic Society. All records in the Specialised Register coded as 'COPD' between 1 January 1987 and 24 November 2013 were searched using the following terms: (exercis* or physical* or train* or rehabilitat* or conditioning or ergometry or treadmill or endurance or "upper limb") AND (non‐invasive* or noninvasive* or "non invasive*" or NIV or "positive pressure" or NIPPV or NPPV or "pressure support" or IPS or "assist* ventilation" or PAV or "ventilatory support" or bilevel or BVS or "mechanical ventilation" or "artificial ventilation" or "artificial respiration" or mask* or BiPAP or IPAP or EPAP or nasal* or "positive airway*"). The search commenced from 1 January 1987, as the first reports in the literature of NIV delivered via a mask were dated 1987 (Ellis 1987; Kerby 1987).

To reduce the risk of missing eligible studies, separate searches were conducted on the following databases across the same time period: AMED, CENTRAL, CINAHL, EMBASE, Latin American and Caribbean Health Science Information Database (LILACS), MEDLINE, Physiotherapy Evidence Database (PEDro), PsycINFO and PubMed. See Appendix 1 for a list of search strategies for each database. Several clinical trials registers and search engines were also screened: Australian New Zealand Clinical Trials Register (www.anzctr.org.au); ClinicalTrials.gov (www.ClinicalTrials.gov); International Standard Randomised Controlled Trial Number Register (www.controlled‐trials.com/isrctn/); Netherlands Trial Register (www.trialregister.nl/trialreg/index.asp); University hospital Medical Information Network (UMIN) (www.umin.ac.jp/ctr/index/); Google Scholar (http://scholar.google.com.au/); and Web of Science (http://thomsonreuters.com/web‐of‐science/).

Searching other resources

We screened reference lists of included studies and of review articles obtained from the initial search for additional studies that potentially met the inclusion criteria. Authors of the included trials and international experts in the field of NIV were contacted and were asked to identify any other published or unpublished studies involving NIV during exercise training in COPD. Four of the six authors of included trials responded (Bianchi 2002; Hawkins 2002; Toledo 2007; van 't Hul 2006), and 11 of the 18 experts responded. No additional trials were identified. We also screened conference abstracts from the following meetings: American College of Chest Physicians, Asia Pacific Society of Respirology, German Society for Pneumology and Respiratory Medicine and the Thoracic Society of Australia and New Zealand. Abstracts were included in this review, and no language restrictions were applied.

Data collection and analysis

Selection of studies

Two review authors (CM and AJP) independently selected studies for inclusion in the review. Initially, titles and abstracts were reviewed, and studies that obviously did not fit the inclusion criteria were discarded. Full papers of the remaining studies were obtained for closer evaluation. Studies that met the inclusion criteria were selected. A list of excluded trials compiled from the group of full papers included the primary reason for exclusion (see Characteristics of excluded studies for details). Disagreements in study selection were resolved by consensus. We calculated a kappa coefficient to determine agreement between the two review authors on study inclusion from the initial selection of full papers (from titles and abstracts) and from the second selection of included studies (from full papers).

Data extraction and management

Two review authors (CM and AJP) independently extracted data from the included studies onto a predesigned form. We recorded the following information: study methods; participant characteristics; interventions; outcomes; and results. Although NIV was used during exercise training in the actively treated groups, post‐training primary and secondary outcome data were extracted only when study participants were evaluated while off NIV (e.g. unassisted test of exercise capacity). Discrepancies in the extracted data were resolved by consensus. If data were not presented numerically, a software programme (Engauge Digitizer, http://digitizer.sourceforge.net/) was used by one review author (KKW) to convert graphical images to numerical data. Two other review authors (CM and AJP) independently manually extracted numerical data from each graph using enlarged copies of the images. Discrepancies were resolved by consensus. Authors of included studies were contacted and were asked to provide missing information when applicable.

Assessment of risk of bias in included studies

Two review authors (CM and AJP) independently assessed the internal validity of the included studies. The strategy recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) was used and included assessment of randomisation sequence generation; allocation concealment; blinding; completeness of outcome assessment; selective outcome reporting; and other potential sources of bias. Unblinded studies were included in this review. Each item was graded as high, low or unclear risk of bias. Disagreements were resolved by consensus. Study authors were contacted to provide additional information when needed. 

Studies with multiple treatment groups

One study (Johnson 2002) consisted of two intervention groups and one control group. Data were extracted only from the intervention group that used NIV during exercise training and from the control group, which performed exercise training alone.

Measures of treatment effect

We recorded mean postintervention values and mean changes from baseline values and standard deviations (SDs) for continuous variables from both groups within each study. The mean difference (MD) and the 95% confidence interval (CI) were used when continuous data measured on the same scale were combined. The standardised mean difference (SMD) was used when studies reported data measured on different scales that could not be calculated back to a common scale. When possible, estimates of treatment effect and confidence limits were related to the minimal important difference (MID) for each outcome. When dichotomous data were combined, the treatment effect was defined as the odds ratio (OR) with 95% CI.

Unit of analysis issues

The unit of analysis was the participant.

Dealing with missing data

If the number of dropouts was large (> 15%), and results from intention‐to‐treat analyses (ITT) and per‐protocol analyses were reported, data were extracted from ITT analyses. If ITT analyses were not reported, data from the per‐protocol analyses were extracted for use in the meta‐analysis. If incomplete statistical results were reported in an included study for a given outcome (e.g. point estimate but no measure of variability), we contacted the study author and asked for the missing data. If the missing data were not provided, data were not extracted from the study for that particular outcome.

Assessment of heterogeneity

The effect of heterogeneity was quantified using the I² statistic. The I² statistic indicates the percentage of the total variation in observed intervention effects across studies that is due to heterogeneity rather than to chance alone (Deeks 2011). The following thresholds have been suggested to guide the interpretation of I²: 0% to 40% might not be important; 30% to 60% may indicate moderate heterogeneity; 50% to 90% may indicate substantial heterogeneity; and 75% to 100% represents considerable heterogeneity (Deeks 2011).

Assessment of reporting biases

As a result of the small number of included trials, we were not able to produce meaningful funnel plots to assess the likelihood of publication bias (Sterne 2011).

Data synthesis

When the included studies were clinically homogeneous, data were combined using Review Manager 5 software (RevMan 2012), and forest plots were generated. We used a fixed‐effect model for all analyses unless a moderate or greater degree of heterogeneity was detected (I² > 30%), in which case we used a random‐effects model.

Subgroup analysis and investigation of heterogeneity

The small number of studies included in this review precluded the investigation of heterogeneity between studies and the performance of subgroup analyses. However, if more studies are included in future updates of this review, the following subgroup analyses will be considered if I² indicates a moderate or higher level of heterogeneity (I² > 30%).

1. Study population (e.g. moderate vs severe to very severe disease (GOLD 2013)).

2. Blinding versus no blinding.

3. Type of exercise (e.g. treadmill vs cycling training, upper limb vs lower limb training).

4. Ventilatory settings (e.g. low‐ vs high‐level ventilatory assistance, mode of ventilation).

5. With versus without the use of supplemental oxygen during exercise training.

6. Duration of the training programme (e.g. standard vs long).

7. Primary limitation to peak exercise (e.g. ventilatory limited vs limited by leg fatigue).

Sensitivity analysis

We performed sensitivity analyses to determine the effects of the following on results: methodological design (blinding and allocation concealment), participant characteristics (disease severity), characteristics of the intervention (programme duration) and between‐group differences at baseline. Sensitivity analyses were limited to outcomes that included data from three or more studies in the initial analysis.

Results

Description of studies

See Characteristics of included studies and Characteristics of excluded studies.

Results of the search

The initial search of electronic databases identified 12,392 potentially relevant reports of studies. Of these, we excluded 12,299 by title and abstract. Full papers of the remaining 93 publications were retrieved for closer inspection. Substantial agreement was reported (Landis 1977) between the two review authors in selection of publications for retrieval of full papers and closer inspection (kappa = 0.78). After the full papers were examined, an additional article was identified from the study reference lists and was retrieved for detailed evaluation. Of the 94 full papers, six met the inclusion criteria of the present review. Perfect agreement was noted between the two review authors for final selection of included studies (kappa = 1.0). A flow chart of the study selection process is displayed in Figure 1. The latest search was run 24 November 2013.

1.

Study flow diagram.

Included studies

In total, six RCTs were included in the review (Bianchi 2002; Hawkins 2002; Johnson 2002; Reuveny 2005; Toledo 2007; van 't Hul 2006). Details of each included study are outlined in Characteristics of included studies, and a summary is provided in Appendix 2. All trials used a parallel‐group design and were published in English. When all studies were combined, data from a total of 126 participants who completed the study protocols (i.e. excluding dropouts) were analysed (control: N = 63; NIV during exercise training: N = 63). Individual study sample sizes ranged from 18 to 29 participants. Studies were conducted in Italy, the United Kingdom, the United States of America, Israel, Brazil and The Netherlands. The mean age of participants ranged from 63 to 71 years. Most participants were male (n = 93 of 108 participants from five studies; one study did not report the sex of participants). Most studies recruited participants with severe to very severe COPD (mean FEV₁ 26% to 41% predicted), and one study recruited participants with moderate to severe COPD (Bianchi 2002).

Exercise training programmes were conducted in the outpatient setting: two were hospital based (Bianchi 2002; Hawkins 2002); four were based in non‐hospital centres (Johnson 2002; Reuveny 2005; Toledo 2007; van 't Hul 2006). Exercise training programmes were similar between studies; most were conducted over six to eight weeks, with two to three sessions per week of 30 to 45 minutes of exercise training per session at a moderately high intensity. One study (Johnson 2002) encouraged participants to perform additional unsupervised exercise at home (without NIV). Based on log book records, this resulted in an average of two extra exercise sessions per week. All studies involved lower limb exercise training. None of the studies assessed upper limb training.

A variety of modes of NIV were used during exercise training, including bilevel, proportional assist ventilation (PAV) and inspiratory pressure support (IPS), with low to moderate levels of ventilatory support. Only one study compared exercise training with NIV versus exercise training with sham NIV (van 't Hul 2006). The remaining studies used exercise training without NIV as the control intervention. Three studies used supplemental oxygen during exercise training (Hawkins 2002; Johnson 2002; Reuveny 2005). Delivery of oxygen was reported as equivalent between groups. None of the studies included participants receiving domiciliary NIV.

All studies used exercise capacity to evaluate treatment effects, and two studies evaluated HRQL. None of the studies used physical activity as an outcome measure. We attempted to contact authors from all six trials to obtain additional information about study design, outcomes or funding support for the study. Three study authors provided the requested information, one gave a partial response and two did not respond.

Excluded studies

A list of studies excluded (N = 88) during the second round of selection (i.e. from the list of full papers that were evaluated in detail) and reasons for exclusion are presented in Characteristics of excluded studies. The primary reasons for exclusion included the following: not an RCT (N = 38); exercise training not evaluated (N = 37); no COPD (N = 6); NIV not used during exercise (N = 4); wrong comparison (N = 2); no stable COPD (N = 1). Of the excluded trials in which the wrong comparison was made, one study (Pires Di Lorenzo 2003) compared nocturnal NIV plus exercise training versus NIV during exercise training without nocturnal NIV. This study was excluded because nocturnal NIV has been shown to augment the benefits of pulmonary rehabilitation (Duiverman 2008; Garrod 2000; Kohnlein 2009), and this could have confounded the results. The second study (Borghi‐Silva 2010) compared supplemental oxygen during exercise training versus NIV during exercise training. This study was excluded because supplemental oxygen during exercise training has been shown to increase both training intensity and exercise capacity in people with COPD compared with exercise training alone (Emtner 2003), which also could have confounded the results. Four excluded studies were written in Portuguese, three in German, one in Russian, one in French and one in Norwegian. The abstract and method sections of these studies were translated before exclusion. The remaining studies were published in English.

Risk of bias in included studies

Details of the review authors' judgements on risk of bias for each included study can be seen in Figure 2, Figure 3 and Characteristics of included studies.

2.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

3.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Allocation

The method of randomisation sequence generation was described and was judged to be adequate in half of the studies (Bianchi 2002; Hawkins 2002; van 't Hul 2006). The remaining three studies did not report randomisation sequence generation (Johnson 2002; Reuveny 2005; Toledo 2007), and inability to contact study authors prevented a conclusive assessment of bias in two studies (Johnson 2002; Reuveny 2005). However, despite the use of sealed, opaque envelopes to conceal group allocation, it may have been possible to predict group allocation for a small number of participants (4/29) in the study by (Hawkins 2002), as randomisation blocks were of a fixed size, the study was performed at a single centre and investigators were not blinded to group allocation.

Blinding

Personnel who trained participants were not blinded to group allocation in any of the studies. Similarly, participants were not blinded in most of the studies; this may have introduced bias for outcomes such as exercise capacity and HRQL, whereas physiological outcomes were less likely to be affected. Consequently, high risk of performance bias was observed for five of the six studies. Lack of blinding of participants largely reflects the difficulty of providing an adequate sham intervention for NIV during exercise training. However, one study (van 't Hul 2006) did blind participants using sham NIV (IPS 5 cmH₂O), which previously has been shown to have an equivalent effect on exercise performance as unassisted exercise in people with severe COPD (van't Hul 2004). Half of the studies (Reuveny 2005; Toledo 2007; van 't Hul 2006) reported using blinded assessors to evaluate clinical outcomes. Two studies (Bianchi 2002; Hawkins 2002) did not use blinded assessors and were judged as having high risk of detection bias. One study (Johnson 2002) did not report whether outcome assessors were blinded, and the study author could not be contacted to provide clarification.

Incomplete outcome data

Five studies reported the number of dropouts and the reasons for dropping out, and one study (Toledo 2007) did not report the number of dropouts. Intolerance of NIV was reported as a reason for dropping out in two studies: In one study (Reuveny 2005), all dropouts from the NIV during training group (n = 3/12 or 25%) were due to NIV intolerance; in the other study (Bianchi 2002), 28% of participants (n = 5/18) dropped out as the result of NIV intolerance. An ITT analysis was performed in two studies (Bianchi 2002; van 't Hul 2006). The study authors stated that the results did not differ from per‐protocol analyses, although data from ITT analyses were not reported.

Selective reporting

Although most studies were free from selective outcome reporting, two studies (Reuveny 2005; Toledo 2007) did not report results for between‐group comparisons for exercise capacity or for a number of physiological variables despite reporting post‐training within‐group differences.

Other potential sources of bias

In one study (Johnson 2002), the results may have been confounded by contamination, as the group randomly assigned to exercise training with NIV also performed unsupervised exercise training without NIV for an average of two sessions per week. In addition, the same study (Johnson 2002) reported significant between‐group differences in baseline exercise capacity, which may have affected the response to NIV during exercise training. The efficacy of the control intervention (unassisted exercise training) was questionable in one study (Reuveny 2005), as within‐group improvement in exercise capacity did not occur. The group that trained with NIV did improve. However, as trainers and participants were not blinded to the intervention, bias cannot be excluded. However, the progression of training intensity was standardised, which should have helped to ensure that participants were exposed to the same training programme.

Effects of interventions

See: Table 1

See Table 1.

Primary outcomes

Exercise capacity

Peak exercise capacity

All six trials included in the review reported the effects of NIV during exercise training on peak exercise capacity. Three studies (Bianchi 2002; Hawkins 2002; Reuveny 2005) assessed peak exercise capacity using an incremental cycle ergometer test in a combined total of 28 participants who trained with NIV and 29 participants who trained without NIV. No clear evidence of a difference was found between training with or without NIV (MD 6.34 watts; 95% CI ‐1.66 to 14.34; Analysis 1.1). Two studies evaluated peak exercise capacity using incremental treadmill tests. One study (Johnson 2002) used a protocol that increased walking speed and incline, with performance measured in METS, and the other study (Toledo 2007) used a protocol that increased walking speed only, while performance was measured in kilometres per hour. Although both studies used incremental treadmill tests to assess peak exercise capacity, results were not combined, as different constructs were measured (one protocol measured peak work, the other measured peak walking speed) (Table 2). Peak oxygen consumption during an incremental treadmill test was also reported in two studies (Reuveny 2005; Toledo 2007). No clear evidence of a difference was found between exercise training with or without NIV (MD 0.12 L/min; 95% CI ‐0.08 to 0.31; Analysis 1.2). The remaining study (van 't Hul 2006) measured peak exercise capacity using the incremental shuttle walk test (ISWT) (Singh 1992). The individual study effect size of 17.0 metres (95% CI ‐ 2.4 to 36.4) was lower than the reported MID of 47.5 metres (95% CI 38.6 to 56.8) for this test (Singh 2008) (Table 2). A significant difference in peak exercise capacity in favour of training with NIV was observed when the percentage change in peak work rate was assessed in three studies (Hawkins 2002; Johnson 2002; Reuveny 2005) in a combined total of 30 participants who received NIV during exercise training and 30 participants who received exercise training alone (MD 17%; 95% CI 7 to 27; Figure 4; Analysis 1.3).

1.1. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 1 Exercise capacity: peak cycle work rate (watts).

1. Results for individual studies.

Outcome or subgroup	Study	Participants	Effect estimate (mean difference (95% confidence interval))
Exercise capacity
Peak work rate (metabolic equivalents (METS))	Johnson 2002	22	0.2 (‐0.8 to 1.2)
Peak treadmill walking speed (km/h)	Toledo 2007	18	0 (‐0.9 to 0.9)
Peak oxygen consumption (% change)	Reuveny 2005	19	21 (5 to 37)
Incremental shuttle walk test (m)	van 't Hul 2006	29	17.0 (‐2.4 to 36.4)
Incremental shuttle walk test (% change)	van 't Hul 2006	29	13 (2 to 24)
Six‐minute walk test (m)	Bianchi 2002	19	4.3 (‐64.1 to 72.7)
Incremental treadmill test distance (m)	Toledo 2007	18	30.9 (‐250.6 to 312.4)
Training intensity
Peak cycle training intensity (W)	Hawkins 2002	19	5.0 (‐4.0 to 14.0)
Peak treadmill training speed (km/h)	Reuveny 2005	19	0 (‐0.5 to 0.5)
Physiological outcomes
Peak exercise minute ventilation (L/min)	Reuveny 2005	19	10.7 (‐4.0 to 25.5)
Peak exercise aerobic threshold (% change)	Reuveny 2005	19	6 (‐8 to 21)
Peak exercise oxygen pulse (mL/beat)	Reuveny 2005	19	1.0 (‐0.8 to 2.8)

1.2. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 2 Exercise capacity: peak VO₂ (L/min).

4.

Forest plot of comparison: 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, outcome: 1.3 Exercise capacity: percentage change.

1.3. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 3 Exercise capacity: percentage change.

Three sensitivity analyses were also performed (Table 3). The analysis for peak work rate (watts) was re‐run first after exclusion of data from one study (Bianchi 2002) that recruited participants with milder disease severity, and second after exclusion of data from another study (Reuveny 2005) that did not report adequate allocation concealment. No change in effect size was observed in either case. Finally, the analysis for percentage change in peak work rate was rerun without data from one study (Johnson 2002) with significant between‐group differences in exercise capacity at baseline. Similarly, no differences in effect size were observed.

2. Sensitivity analysis.

Outcome	Initial analysis	Sensitivity analysis
	Effect estimate	Study removed	Reason for removal	Included studies	Effect estimate
Exercise capacity
Peak cycle work rate (watts)	6.34 (‐1.66 to 14.34)	a	Disease severity	b, d	6.33 (‐2.02 to 14.67)
Peak cycle work rate (watts)	6.34 (‐1.66 to 14.34)	d	Allocation concealment	a, b	5.45 (‐4.48 to 15.37)
Peak work rate (% change)	17 (7 to 27)	c	Baseline differences	b, d	18 (7 to 29)
Training intensity
Training intensity (% change)	13 (1 to 27)	a	Disease severity	b, f	20 (12 to 28)
Training intensity (% change)	13 (1 to 27)	f	Blinding	a, b	10 (‐9 to 28)
Physiological outcomes
Isotime exercise VE (L/min)	‐0.08 (‐2.82 to 2.67)	a	Disease severity	b, f	‐0.68 (‐4.89 to 3.52)
Peak exercise VE (L/min)	2.68 (‐2.02 to 7.37)	a	Disease severity	b, d	2.60 (‐2.35 to 7.55)
Isotime exercise VE (L/min)	‐0.08 (‐2.82 to 2.67)	f	Blinding	a, b	0.30 (‐2.70 to 3.30)
Peak exercise La (mmol/L)	‐0.35 (‐1.10 to 0.41)	d	Allocation concealment	b, e	‐0.61 (‐1.59 to 0.37)
Peak exercise VE (L/min)	2.68 (‐2.02 to 7.37)	d	Allocation concealment	a, b	2.54 (‐3.09 to 8.16)
Peak exercise La (mmol/L)	‐0.35 (‐1.10 to 0.41)	e	Programme duration	b, d	0.04 (‐0.55 to 0.62)
Dyspnoea
Isotime dyspnoea (Borg)	‐0.33 (‐0.83 to 0.16)	a	Disease severity	b, e	‐0.34 (‐0.86 to 0.17)
Isotime dyspnoea (Borg)	‐0.33 (‐0.83 to 0.16)	e	Programme duration	a, b	‐0.39 (‐0.92 to 0.14)
Dropouts
Dropouts	OR 1.26 (0.61 to 2.59)	a	Disease severity	b, c, d, f	OR 1.07 (0.46 to 2.48)

Effect estimate is presented as mean difference (95% confidence interval) unless otherwise indicated;^aBianchi 2002; ^bHawkins 2002; ^cJohnson 2002; ^dReuveny 2005; ^eToledo 2007; ^fvan 't Hul 2006.

La: lactate; OR: odds ratio; VE: minute ventilation.

Endurance exercise capacity

Endurance exercise capacity was assessed with a constant work rate cycle ergometer test in two studies (Hawkins 2002; van 't Hul 2006) in a combined total of 25 participants who trained with NIV and 23 participants who performed exercise training alone or with sham NIV. The reported MID for the constant work rate cycle endurance test (performed at 75% peak work capacity) is 101 seconds (95% CI 86 to 116) (Puente‐Maestu 2009). A trend for increased exercise endurance was found to favour exercise training with NIV (MD 3.62 minutes; 95% CI ‐0.17 to 7.41; Analysis 1.4). However, the lower limit of the confidence interval crossed zero. When the summary effect for each study was expressed as the percentage change from baseline, rather than as post‐intervention values, a significant effect in favour of exercise training with NIV was observed when the results were combined (MD 59%; 95% CI 4 to 114; Figure 4; Analysis 1.3). Although the mean effect size for percentage change in endurance time was greater than the reported MID for percentage change in constant work rate cycle endurance of 34% (95% CI 29 to 39) (Puente‐Maestu 2009), the lower limit of the confidence interval was less than the MID.

1.4. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 4 Exercise capacity: constant work rate cycle endurance time (minutes).

Functional exercise capacity

Functional exercise capacity was measured in one study (Bianchi 2002) by the six‐minute walk test (6MWT). The MID for the 6MWT in people with COPD is 25 metres (95% CI 20 to 61) (Holland 2010). Individual study results demonstrated no statistically or clinically significant difference between training with NIV and exercise training alone (MD 4.3 metres; 95% CI ‐64.1 to 72.7) (Table 2).

Health‐related quality of life

Health‐related quality of life was measured in two studies (Bianchi 2002; van 't Hul 2006) with the St George’s Respiratory Questionnaire (SGRQ) in a total of 24 participants who trained with NIV and 24 participants who trained without NIV or with sham NIV. A reduction of four points in the SGRQ total score represents a clinically worthwhile improvement in HRQL (Jones 2002). No clear evidence of an effect on HRQL was found for the SGRQ total score (MD 2.5 points; 95% CI ‐2.3 to 7.2). Similar results were found for the three subscales of the SGRQ: symptoms (MD 0.9 points; 95% CI ‐10.2 to 11.9); activity (MD 0.1 points; 95% CI ‐14.9 to 15.0); and impacts (MD 0.1 points; 95% CI ‐6.8 to 7.1) (Figure 5; Analysis 1.5). Heterogeneity between studies was considerable (I² = 77%) for the activity subsection of the SGRQ.

5.

Forest plot of comparison: 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, outcome: 1.5 Health‐related quality of life: St George's Respiratory Questionnaire.

1.5. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 5 Health‐related quality of life: St George's Respiratory Questionnaire.

Physical activity

None of the included studies reported physical activity as an outcome.

Secondary outcomes

Training intensity

Three studies (Bianchi 2002; Hawkins 2002; van 't Hul 2006) reported the training intensity achieved during the final training session (expressed as a percentage of baseline peak work capacity) in a combined total of 34 participants who trained with NIV and 33 participants who performed exercise training alone or with sham NIV. A significant effect on training intensity was found to favour training with NIV during exercise (MD 13%; 95% CI 1 to 27; Figure 6; Analysis 1.6). However, heterogeneity between studies was substantial (I² = 72%).

6.

Forest plot of comparison: 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, outcome: 1.6 Training intensity: Final training session (% baseline peak work capacity).

1.6. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 6 Training intensity: final training session (% baseline peak work capacity).

Two sensitivity analyses were conducted (Table 3). First, the analysis was rerun without data from one study (Bianchi 2002) that recruited participants with milder disease. The effect size increased to a mean of 20% (95% CI 12 to 28), and heterogeneity was reduced to 0%. Second, the analysis was rerun without data from one study (van 't Hul 2006) that blinded participants to determine whether the effect size was different (e.g. overestimated) if only studies with unblinded participants were included. The effect size was slightly reduced and the 95% CI widened, with the lower limit of the 95% CI crossing zero (MD 10%; 95% CI ‐9 to 28). Heterogeneity also increased to I² = 83%.

Physiological outcomes

A significant decrease in isoload blood lactate was observed to favour training with NIV when data from two studies (Hawkins 2002; Toledo 2007) with 19 participants who trained with NIV and 18 participants who trained without NIV (MD ‐0.97 mmol/L; 95% CI ‐1.58 to ‐0.36; Figure 7; Analysis 1.7) were combined. There was no clear evidence of an effect between exercise training with NIV and exercise training alone or exercise training with sham NIV for peak exercise blood lactate, isotime exercise minute ventilation (VE), post‐training peak exercise VE, or change in oxygen consumption at the anaerobic threshold (Analysis 1.8; Analysis 1.9; Analysis 1.10; Analysis 1.11). A moderate level of heterogeneity between studies was found for the analysis of peak exercise blood lactate (I² = 59%).

7.

Forest plot of comparison: 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, outcome: 1.7 Physiological outcomes: Isoload lactate (mmol/L).

1.7. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 7 Physiological outcomes: isoload lactate (mmol/L).

1.8. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 8 Physiological outcomes: peak exercise lactate (mmol/L).

1.9. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 9 Physiological outcomes: isotime exercise minute ventilation (L/min).

1.10. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 10 Physiological outcomes: peak exercise minute ventilation (L/min).

1.11. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 11 Physiological outcomes: change in VO₂ at anaerobic threshold (L/min).

Several sensitivity analyses were conducted (Table 3). First, analyses for isotime exercise VE and for peak exercise VE were rerun without data from one study (Bianchi 2002) that recruited participants with milder disease. For each outcome, the effect size did not change. The analysis for isotime exercise VE was rerun without data from one study (van 't Hul 2006), which blinded participants to determine whether the effect size was different if only studies with unblinded participants were included. The effect size did not change substantially. Analyses for peak exercise blood lactate and peak exercise VE were also rerun with data excluded from one study (Reuveny 2005) that did not report adequate allocation concealment. A slight increase in effect size was noted for peak exercise blood lactate from ‐0.35 mmol/L (95% CI ‐1.10 to 0.41) in the initial analysis to ‐0.62 mmol/L (95% CI ‐1.22 to ‐0.01), and heterogeneity between studies did not change. The effect size for peak exercise VE was not altered. Finally, the effect size for peak exercise blood lactate was mildly reduced to 0.04 mmol/L (95% CI ‐0.55 to 0.62) when the analysis was repeated without data from one study (Toledo 2007) with a programme duration approximately twice as long as that of other studies included in the review, and heterogeneity between studies decreased to 0%.

Dyspnoea

Post‐training isotime exercise dyspnoea was measured in three studies (Bianchi 2002; Hawkins 2002; Toledo 2007) in a total of 28 participants who trained with NIV and 28 participants who performed exercise training alone. No significant effect on dyspnoea, as measured on the Borg scale, was noted between participants performing exercise training with and without NIV (MD ‐0.18; 95% CI ‐1.09 to 0.72; Analysis 1.12). A sensitivity analysis that excluded data from one study (Bianchi 2002), which recruited participants with milder disease, did not change the size of the effect. Similarly, a sensitivity analysis that excluded data from one study (Toledo 2007) with a longer programme duration did not alter the effect size (Table 3).

1.12. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 12 Dyspnoea: isotime exercise dyspnoea (Borg scale).

Dropouts

Dropouts were reported in five studies (Bianchi 2002; Hawkins 2002; Johnson 2002; Reuveny 2005; van 't Hul 2006) from a total of 151 participants (78 participants who were randomly assigned to exercise training with NIV and 73 participants who were randomly assigned to exercise training without NIV or with sham NIV). There was no evidence of a clear effect on dropouts with NIV during exercise training compared with exercise training alone, or exercise training with sham NIV (OR 1.26; 95% CI 0.61 to 2.59; Analysis 1.13). A sensitivity analysis that excluded data from one study (Bianchi 2002), which recruited participants with milder disease, did not change the magnitude of the effect (Table 3).

1.13. Analysis.

Comparison 1 Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation, Outcome 13 Dropouts.

Adverse events

Adverse events were not reported in any of the studies.

Cost

Cost was not reported in any of the studies.

Discussion

Summary of main results

Pulmonary rehabilitation, with exercise training as a key component, is well established as a standard of care for people with COPD, with demonstrated improvement in exercise capacity, HRQL and dyspnoea (Lacasse 2006). The aim of this systematic review was to determine whether NIV during exercise training could provide benefit for exercise capacity, HRQL and physical activity above that of exercise training alone in people with COPD. The current review showed that NIV during exercise training allowed participants to achieve a greater percentage improvement in lower limb peak and endurance exercise capacity, to exercise at a higher training intensity and to gain a greater physiological training effect compared with exercise training alone or exercise training with sham NIV. There was no clear evidence that HRQL was better or worse with NIV during exercise training, and the effect of NIV during exercise training on physical activity is unknown, as none of the included studies reported this outcome.

Results for the effect of NIV during exercise training on exercise capacity should be interpreted with caution, as differences were found only when percentage change from baseline values rather than post‐intervention values were used in analyses. One possible explanation for the difference in results is that if large interindividual or intergroup baseline differences were present, the use of change from baseline values rather than postintervention values would provide greater statistical power to detect treatment effects. In addition, the overall quality of the evidence for percentage change in peak and endurance exercise capacity was judged as low (see Table 1). The clinical significance of the treatment effect for percentage change in peak exercise capacity is unknown, and the effect size may have been exaggerated because of the high risk of bias of studies included in the analysis. Endurance exercise capacity may be more relevant to people with COPD than peak exercise capacity, given that most daily activities are performed at a submaximal level (Pitta 2005). However, interpretation of the clinical significance of the effect size for percentage change in endurance exercise capacity is also unclear. Although the mean effect of 59% was above the reported MID of 34% (Puente‐Maestu 2009), the 95% CI was very wide, with the lower limit of the CI (4%) considerably below the MID.

The finding of an improvement in some aspects of exercise capacity with NIV during exercise training compared with exercise training alone or exercise training with sham NIV may relate to the fact that NIV during exercise training permits higher‐intensity exercise training and results in a greater physiological training effect, as reflected by lower isoload blood lactate levels. It is interesting to note that although isoload lactate was reduced with NIV during exercise training, no evidence was found of a significant reduction in isotime VE or isotime dyspnoea. Although overall assessments of the quality of the evidence for training intensity and isoload lactate were moderate (see Table 1), it is unknown whether the size of the treatment effects is clinically meaningful.

As no cure for COPD is known, treatment aims to relieve symptoms, slow disease progression, optimise function and overall health and prevent and treat exacerbations (GOLD 2013). As such, HRQL is an important outcome for people living with COPD. Although the overall quality of the evidence for HRQL was judged as moderate (see Table 1), only two of the studies included in this review assessed HRQL. In addition, significant heterogeneity was found across studies for the activity subsection of the SGRQ and could not be investigated further because of the small number of studies included in the analysis. As a result, the effect of NIV during exercise training on this domain remains uncertain. The effect of NIV during exercise training on functional exercise capacity is also unclear, as this outcome was measured in only one study (Bianchi 2002) by the 6MWT. Changes in six‐minute walk distance are an important prognostic indicator for people with COPD and have been shown to relate to mortality (Polkey 2013) and risk of hospitalisation (Spruit 2012). Similarly, the effect of NIV during upper limb exercise training is unknown, as none of the included studies used upper limb exercise as a training modality. Upper limb training is recommended as part of a comprehensive pulmonary rehabilitation programme (Spruit 2013), and some evidence suggests that NIV during unsupported arm exercise improves endurance exercise capacity during a single exercise session (Menadue 2009a). Consequently, HRQL, functional exercise capacity and upper limb training should be considered as outcomes for future studies.

Among the combined total of 63 participants who trained with NIV, no adverse events were reported. However, as the total number of participants who trained with NIV was relatively small, the effect of NIV during exercise training on adverse events is unclear in people with moderate to very severe COPD.

Overall completeness and applicability of evidence

The studies included in the current review recruited participants with severe to very severe COPD (GOLD 2013), with the exception of one study (Bianchi 2002), which recruited participants with moderate to severe COPD. The impact of disease severity on the efficacy of training with NIV could not be formally assessed in the present review. However, based on outcomes from the individual included studies, it appears that disease severity could be an important factor in patient selection for this technique, with greater benefit reported in studies in which individuals with severe to very severe COPD were recruited, compared with those with moderate disease (Bianchi 2002). In addition to selecting people with severe COPD, two studies (Reuveny 2005; van 't Hul 2006) selected participants who demonstrated a very limited ventilatory reserve at peak unassisted exercise, suggesting a ventilatory limitation to exercise. In the latter study (van 't Hul 2006), participants were included only if they were tolerant of NIV, indicating that participants were highly selected. It is unclear whether a trial of NIV was undertaken to test acceptability before enrolment in the other included studies. Three of the included studies (Reuveny 2005; Toledo 2007; van 't Hul 2006) also did not report the number of patients screened during the recruitment process. Of those studies that did report the number of patients screened during the recruitment process (Bianchi 2002; Hawkins 2002; Johnson 2002), no information was provided regarding the number of patients who declined to take part because of the intervention (NIV). Subsequently, the potential for participants to have been highly selected cannot be excluded. In addition, two studies reported dropouts due to poor tolerance of NIV (Bianchi 2002; Reuveny 2005), which could have related to selection of participants with less severe COPD (Bianchi 2002) or the provision of lower levels of ventilatory support (Bianchi 2002; Reuveny 2005) compared with other studies (Hawkins 2002; van 't Hul 2006). Consequently, the findings of the present review may not be applicable to all people with moderate to very severe COPD.

Although the studies included in the present review were reasonably homogeneous and representative of current clinical practice with respect to the exercise training programmes, substantial diversity was reported regarding the delivery of NIV. Three different NIV modes were used (bilevel, PAV and IPS), and ventilatory support ranged from a low to a moderate level. None of the included studies assessed high‐level pressure support, which has shown promising results during ground walking in people with very severe COPD (Dreher 2007). During pressure preset ventilation, the amount of tidal volume assistance delivered will vary, depending on factors such as respiratory system compliance, airways resistance and inspiratory time (Mehta 2001). As a result, a given level of pressure support can have a different effect on tidal volume between participants and even within an individual, for example, if dynamic hyperinflation occurs during exercise and respiratory system compliance is reduced. However, as subgroup analyses could not be performed, the influence of these factors on the treatment effects associated with NIV during exercise training is unclear. As yet, the optimal mode and settings for NIV during exercise training are unknown. FInally, although NIV during exercise training could potentially benefit select individuals with COPD, implementation of this technique does have resource implications and would require experienced staff and access to appropriate equipment and may involve extra costs, which could limit the feasibility of this technique in some settings.

Quality of the evidence

Six studies with a combined total of 126 participants who completed the study protocols (63 with NIV during exercise training; 63 with exercise training alone or exercise training with sham NIV) were included in the current review. Limitations in the literature were noted in terms of the small number of studies included in the analyses, the small numbers of participants within the included studies and issues related to methodological quality such as lack of blinding or inadequate reporting of allocation concealment. These limitations are reflected in assessments of the quality of the evidence, which ranged from low for exercise capacity outcomes to moderate for HRQL, training intensity and post training isoload blood lactate levels (see Table 1).

The key methodological limitation of the studies was lack of blinding. Only one study blinded participants, three used blinded assessors and none of the included studies blinded trainers, which may have introduced performance or detection bias. Consequently, important outcomes such as exercise capacity and HRQL could be influenced by bias, as all analyses included data from unblinded studies. Unblinded studies have been shown to overestimate treatment effect size by 9% compared with blinded studies (Pildal 2007). Blinding an intervention such as NIV is difficult but may be achieved with the use of sham NIV. However, if an inappropriate sham NIV is used, the treatment effect size could be altered by sham NIV either impeding exercise performance or assisting exercise performance compared with what would have occurred during unassisted exercise. Therefore a sham NIV would have to be shown to be appropriate for a given patient population before commencement of a training study. For example, the sham NIV used by one study (van 't Hul 2006) was previously demonstrated to have an equivalent effect on exercise performance as unassisted exercise (van't Hul 2004). Sensitivity analyses were performed to determine whether lack of blinding exaggerated effect sizes in the present review. Because of the small number of included studies, this could be performed only for isotime exercise VE and training intensity. No substantial changes in effect size were observed, suggesting that these results are robust.

Allocation concealment was adequately performed and reported in most of the trials. However, two studies (Johnson 2002; Reuveny 2005) did not provide an adequate description of allocation concealment in the paper, and the study authors could not be contacted to provide additional information. In addition, although allocation concealment was adequately described in another study (Hawkins 2002), it may have been compromised for a small number of participants as the result of block randomisation. If the size of the blocks used during block randomisation is fixed and known, it may be possible to predict future group allocation for some participants in an unblinded trial (Berger 2005). Inclusion of studies without adequate allocation concealment has been reported to overestimate effect size by 18% to 37% (Moher 1998, Pildal 2007). In the present review, only a limited number of sensitivity analyses could be conducted to assess the impact of including trials without adequate description of allocation concealment. The change in effect size was small to negligible for peak work rate (watts), training intensity, peak exercise blood lactate and peak exercise VE, indicating that these results were also robust.

Another factor that may have impacted the results of the current review is lack of statistical power. Only six studies were eligible for inclusion in the review, and within each study, sample sizes were generally small, with dropout rates in five of the studies ranging from 21% to 42%. In addition, a variety of measurement tools were used to assess outcomes of interest. Consequently, only a small number of meta‐analyses could be performed, often with results from only two to four studies combined, occasionally with data from as few as 37 participants. Statistical power also was probably compromised in individual studies. Two studies (Reuveny 2005; Toledo 2007) failed to present post‐training results for between‐group comparisons of expected outcomes such as exercise capacity, despite conducting parallel RCTs to assess the effects of training with NIV versus exercise training alone. This reporting bias may have occurred because significantly different results between groups were not detected. Another study (Hawkins 2002) was powered to assess post‐training isoload blood lactate. However, the combination of dropouts and difficulty gaining vascular access in some participants reduced the power of the study to detect differences between groups. When isoload blood lactate data from this study were combined with data from Toledo 2007 in the present review, statistical power was improved, and a difference between interventions was found to favour training with NIV. A larger number of RCTs with greater numbers of participants are needed to achieve sufficient statistical power to confidently assess the effects of NIV during exercise training on key outcomes.

Significant heterogeneity across studies was detected in only three analyses: HRQL (activity subsection of the SGRQ); peak exercise blood lactate; and training intensity. The most likely reason for heterogeneity for the activity subsection of the SGRQ was a difference in disease severity between the two studies. The condition of participants from one study (van 't Hul 2006) was more severe (based on FEV₁ and ventilatory reserve at peak exercise) than that of participants recruited by the second study (Bianchi 2002), and a trend was found to favour NIV during training improving this outcome. In contrast, the second study (Bianchi 2002) reported a trend for improvement in this outcome in favour of the control group. It is unlikely that the treatment effect was overestimated by the first study (van 't Hul 2006), as allocation concealment was adequate and both participants and assessors were blinded. However subgroup and sensitivity analyses could not be performed to investigate the cause of heterogeneity, as only two studies reported this outcome. Three studies were included in the analysis of peak exercise blood lactate. The training programme in one study (Toledo 2007) was substantially longer (12 weeks) than the programmes in the other two studies (six to eight weeks) (Hawkins 2002; Reuveny 2005). Greater training effects can be achieved with a training programme duration of 12 weeks or longer in comparison with programmes with a duration of six to eight weeks (Ries 2007). This factor appeared to account for the difference in effect sizes between studies, as demonstrated in a sensitivity analysis for which data from the study with the longer programme duration (Toledo 2007) were removed, with the summary effect size reduced and I² decreased to zero. Finally, differences in disease severity appeared to explain heterogeneity across studies in the analysis of training intensity. Two studies (Hawkins 2002; van 't Hul 2006) recruited participants with severe to very severe COPD and reported an increase in training intensity with NIV during exercise training compared with control, whereas another study (Bianchi 2002) recruited a 'milder' group of participants and found no difference in training intensity between those who trained with NIV and the control group, suggesting that individuals with less severe disease may not derive benefit from NIV during exercise. A sensitivity analysis that excluded data from Bianchi 2002 increased the summary effect size and reduced I² to zero.

Potential biases in the review process

Strengths of the review process include adherence to a predefined protocol (Menadue 2009b), with the exception of several small alterations (see Differences between protocol and review), and the performance of a comprehensive literature search (including non‐English trials and grey literature). A potential weakness of the review process was the inability to assess for the likelihood of publication bias because of the small number of included trials. To reduce the risk of publication bias, a number of clinical trial registers were searched, conference abstracts were reviewed and international experts in the field of NIV were asked to identify further published or unpublished trials. However, no additional potential studies were found. Not all of the studies included in the present review reported results in favour of NIV during exercise training. Nevertheless, publication bias cannot be excluded. Finally, not all of the authors of included studies could be contacted to provide additional information regarding study design or data. This may have affected the judgement of some categories of risk of bias and limited the data included in meta‐analyses for some outcomes.

Agreements and disagreements with other studies or reviews

Two non‐Cochrane systematic reviews and meta‐analyses have previously investigated the effects of NIV during exercise. Ricci 2013 evaluated the physiological effects of NIV during exercise training in people with stable COPD compared with control (exercise training alone or exercise training with sham NIV or supplemental oxygen). In addition to the six RCTs included in the present review, Ricci 2013 included one quasi‐randomised study (Costes 2003) and one study that compared supplemental oxygen during exercise training with NIV during exercise training (Borghi‐Silva 2010). In contrast to the present review, Ricci 2013 found no difference between NIV during exercise training and control with respect to lactate. Also no difference between NIV during exercise training and control was found for heart rate, oxygen consumption (VO₂) and workload. However, very limited data were provided, as the study authors did not state which studies were included in each meta‐analysis and did not report the summary effect for outcomes other than VO₂. The second systematic review (van't Hul 2002) reported significant benefit for exercise endurance time and dyspnoea during a single application of NIV during exercise in people with COPD compared with control (exercise without NIV). The present review also found some evidence that endurance exercise capacity may be improved with NIV during exercise, although, in contrast to van't Hul 2002, for which included studies tested endurance exercise capacity while participants breathed on NIV, studies included in the present review tested exercise capacity post training and without NIV. Also, the present review did not find a reduction in dyspnoea associated with NIV during training. However, as dyspnoea was measured after training and without NIV during exercise in the studies included in the present review, the difference in results may simply reflect the fact that the reduction in dyspnoea is a temporary phenomenon related to respiratory muscle unloading (Maltais 1995) as a direct result of the application of NIV.

To date, three literature reviews have specifically addressed the role of NIV during exercise training as part of a pulmonary rehabilitation programme (Araujo 2005; Corner 2010; De Backer 2010). Meta‐analyses were not performed in any of the reviews. Araujo 2005 did not discuss any of the studies included in the present review, and Corner 2010 included five RCTs from the current review (Bianchi 2002; Hawkins 2002; Johnson 2002; Reuveny 2005; van 't Hul 2006) and one quasi‐randomised study (Costes 2003) that was excluded from the present review, and excluded one RCT (Toledo 2007) that was included in the present review. De Backer 2010 included all six RCTs from the present review, as well as one quasi‐randomised study (Costes 2003) and one randomised cross‐over study (Barakat 2007), all of which were excluded from the present review. Conclusions were similar between the three reviews, namely, that NIV may permit patients with moderate to very severe COPD to train at a higher intensity and gain greater improvement in exercise capacity compared with exercise training alone. The current review largely supports these findings. However, it is unclear whether the observed improvement in exercise capacity is clinically worthwhile.

Several pulmonary rehabilitation practice guidelines have included recommendations regarding the role of NIV as an adjunct to exercise training during pulmonary rehabilitation. Most recently, the American Thoracic Society/European Respiratory Society Statement 'Key Concepts and Advances in Pulmonary Rehabilitation' (Spruit 2013) referred to the findings of the literature review by Corner 2010, which concluded that NIV appears to enhance the effects of exercise training, with greatest benefit observed in individuals with severe disease. It was also stated that as NIV is a difficult and labour‐intensive intervention, its use may be feasible only in centres with significant expertise with NIV, and for individuals with demonstrated benefit from using NIV during exercise (Spruit 2013). Also in 2013, the British Thoracic Society Guideline on Pulmonary Rehabilitation in Adults (Bolton 2013) stated that NIV should not be used routinely during pulmonary rehabilitation in patients with chronic hypercapnic respiratory failure who do not already receive domiciliary NIV, and that patients with chronic hypercapnic respiratory failure who use domiciliary NIV should be offered the opportunity to exercise with NIV during pulmonary rehabilitation, provided that this is tolerable and accepted by the patient. In 2007, The Joint American College of Chest Physicians/American Association of Cardiovascular and Pulmonary Rehabilitation Evidence‐Based Clinical Practice Guidelines stated that NIV during exercise training may be of benefit for select patients with severe COPD and may permit modest improvements in exercise performance above that of exercise training alone (Ries 2007). The current review provides evidence to support some of these recommendations, for example that NIV during exercise may lead to improvement in endurance exercise capacity above that of exercise training alone. However, regarding patient selection for this technique, although the results of individual studies suggest that individuals with severe to very severe COPD may respond better to NIV during exercise training than those with milder disease, an insufficient number of included studies in the present review precluded the performance of subgroup analyses to determine the effect of disease severity on outcomes. The present review does not provide evidence for the use of NIV during exercise training in people with chronic hypercapnic respiratory failure secondary to COPD, as none of the studies included in the present review recruited participants with chronic hypercapnia. However, as benefit for exercise capacity and dyspnoea has been observed in this population during an acute application of NIV during exercise (Bianchi 1998), investigation of the role of NIV during exercise training is warranted. Some guidelines (Spruit 2013) recommend selecting individuals for NIV during training who have previously demonstrated an acute benefit from exercise with NIV. However, currently no evidence is available to suggest that selecting individuals on these grounds will result in greater training effects with NIV, as the predictive validity of the acute response is low (van 't Hul 2006).

Although a number of pulmonary rehabilitation review articles (Spruit 2013; Troosters 2005; Troosters 2010) allude to NIV during exercise training as a difficult, costly and time‐consuming intervention, none of the studies included in the present review reported the additional costs associated with using NIV during exercise training when compared with exercise training alone. One of the included studies (Bianchi 2002) did report that staff spent an average of 11 ± 3 minutes setting up the ventilator. However, it is unclear whether this time was spent during the initial session or during each session of the rehabilitation programme, or how this compared with the amount of time spent with the group who performed exercise training without NIV. The issue of cost, in terms of staff time and additional resources, requires further investigation before conclusions can be drawn as to whether NIV could or could not be a cost‐effective adjunct to exercise training for people with COPD.

Authors' conclusions

Implications for practice.

This review provides evidence that NIV during exercise training may allow people with COPD to exercise at a higher training intensity and to achieve a greater physiological training effect compared with exercise training alone or exercise training with sham NIV. Although some evidence suggests that NIV during exercise training improves the percentage change in peak and endurance exercise capacity, these findings are not consistent across other measures of peak and endurance exercise capacity. The results for quality of life were uncertain and our analysis did not exclude there being an effect with NIV during exercise. It is currently unknown whether the demonstrated benefit of NIV during exercise training is clinically worthwhile or cost‐effective.

Implications for research.

To conclusively determine the effects of NIV during exercise training, additional RCTs with larger numbers of participants are needed. It is essential that studies have a strong methodological design to minimise the risk of bias, as well as high‐quality reporting to enable accurate assessment of the risk of bias. In particular, blinding of participants, trainers and assessors is required, although arguably difficult, with an intervention such as NIV. Important outcomes that should be evaluated include endurance exercise capacity, HRQL and physical activity. Assessment of exercise capacity using tests for which the MID is known may help to clarify whether clinically relevant improvements in exercise capacity can be obtained. Longer‐term follow‐up of study participants (e.g. 12 months) should also be performed. In addition, future studies need to quantify the extra time and costs associated with NIV during exercise training, so that this potential barrier can be weighed against any potential benefits.

The optimal mode and settings for NIV during exercise training are not well defined and may directly alter the efficacy of NIV during exercise training. However, a larger number of studies are required before subgroup analyses can be performed to determine the effects of ventilator mode and settings on important outcomes. Studies included in the present review used only low to moderate levels of ventilatory support during exercise training. Assessment of high‐level ventilatory support during exercise training is warranted. Evaluation of NIV during exercise training in subgroups, such as those with or without a limited ventilatory reserve at peak (unassisted) exercise and individuals with or without significant dynamic hyperinflation during exercise, may help to better define a target population for this intervention. People with chronic hypercapnic respiratory failure who are considered appropriate candidates for nocturnal NIV could potentially benefit from NIV during exercise training and should be assessed in future studies.

Appendices

Appendix 1. Search strategies

A randomised controlled trial (RCT) filter was applied to all database searches except for CENTRAL and PEDro, as these databases contain only controlled trials and systematic reviews. The filter was developed by grouping database‐specific terms for ‘RCT’ using ‘or.’ Then, database‐specific search terms for ‘NIV' were grouped using ‘or,’ database‐specific search terms for ‘Exercise’ were grouped using ‘or’ and database‐specific search terms for ‘COPD’ were grouped using ‘or.’ Afterwards, the following four search strategies were used for each database: (I) ‘RCT’ and ‘COPD’ and ‘NIV’ and ‘Exercise’;  (ii) ‘RCT’ and ‘COPD’ and ‘NIV’;  (iii) ‘RCT’ and ‘COPD’ and ‘Exercise’; and (iv) ‘NIV’ and ‘Exercise.’ We searched the following databases via Ovid: EMBASE; MEDLINE; Allied and Complementary Medicine Database (AMED); Cumulative Index to Nursing and Allied Health Literature (CINAHL); PsycINFO. See below for detailed search strategies for each database.

Allied and Complementary Medicine Database (AMED)

	RCT	NIV	Exercise	COPD
Search terms	#1 Randomized controlled trial.pt. #2 Controlled clinical trial.pt. #3 Randomized.ab. #4 Placebo.ab. #5 Randomly.ab. #6 Trial.ab. #7 Groups.ab. #8 #1 or #2 or #3 or #4 or #5 or #6 or #7 #9 (animals not (humans and animals)).sh. #10 #8 not #9	#11 Respiration, Artificial/ or Positive‐Pressure Respiration/ #12 Intermittent Positive‐Pressure Ventilation/ #13 Ventilators mechanical/ #14 #11 or #12 or #13	#15 Exercise/ or Exercise Therapy/ or Exercise Test/ or Exercise Tolerance/ #16 "Physical Education and Training"/ #17 Rehabilitation/ #18 Physical Fitness/ or Physical Endurance/ #19 Ergometry/ #20 Walking/ #21 Bicycling/ #22 Upper Extremity/ #23 #15 or #16 or #17 or #18 or #19 or #20 or #21 or #22	#24 Pulmonary Disease, Chronic Obstructive/ #25 Lung diseases obstructive/ #26 #24 or #25

The following combinations of groups of search terms using ‘and’ was performed: RCT (#10) and COPD (#26) and NIV (#14) and Exercise (#23); RCT (#10) and COPD (#26) and NIV (#14); RCT (#10) and COPD (#26) and Exercise (#23); NIV (#14) and Exercise (#23).

Cochrane Central Register of Controlled Trials (CENTRAL)

This search was performed in addition to the search conducted by the Cochrane Airways Group Trials Search Co‐ordinator. CENTRAL is a database of randomised trials and systematic reviews; therefore a randomised controlled trial filter was not required.

	NIV	Exercise	COPD
Search terms	#1 Non‐invasive ventilation #2 NIV #3 Proportional assist ventilation #4 PAV #5 Inspiratory pressure support #6 IPS #7 #1 or #2 or #3 or #4 or #5 or #6	#8 Exercise	#9 COPD

The following combinations of groups of search terms using ‘and’ was performed: COPD (#9) and NIV (#7) and Exercise (#8); COPD (#9) and NIV (#7); COPD (#9) and Exercise (#8); NIV (#7) and Exercise (#8).

Cumulative Index to Nursing and Allied Health Literature (CINAHL)

	RCT	NIV	Exercise	COPD
Search terms	#1 Clinical trials or placebo	#2 Positive pressure ventilation #3 Ventilators, mechanical #4 #2 or #3	#5 Exercise #6 Physical endurance #7 Endurance #8 Rehabilitation #9 Rehabilitation exercise #10 Conditioning #11 Cardiopulmonary #12 Ergometry #13 Treadmills #14 Exercise test #15 Upper extremity #16 Upper extremity exercises #17 #5 or #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13 or #14 or #15 or #16	#18 Lung diseases, obstructive

The following combinations of groups of search terms using ‘and’ was performed: RCT (#1) and COPD (#18) and NIV (#4) and Exercise (#17); RCT (#1) and COPD (#18) and NIV (#4); RCT (#1) and COPD (#18) and Exercise (#17); NIV (#4) and Exercise (#17).

EMBASE

	RCT	NIV	Exercise	COPD
Search terms	#1 Random$ #2 Factorial$ #3 Crossover$ #4 Cross over$ #5 Cross‐over$ #6 Placebo$ #7 Doubl$ adj blind$ #8 Singl$ adj blind$ #9 Assign$ #10 Allocate$ #11 Volunteer$ #12 Crossover‐procedure #13 Double‐blind procedure #14 Single‐blind procedure #15 Randomized controlled trial #16 #1 or #2 or #3 or #4 or #5 or #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13 or #14 or #15	#17 Artificial ventilation #18 Assisted ventilation #19 Intermittent positive pressure ventilation #20 Pressure support ventilation #21 #17 or #18 or #19 or #20	#22 Exercise #23 Arm exercise #24 Exercise test #25 Exercise tolerance #26 Treadmill exercise #27 Leg exercise #28 Pulmonary rehabilitation #29 Rehabilitation #30 Training #31 Ergometry #32 #22 or #23 or #24 or #25 or #26 or #27 or #28 or #29 or #30 or #31	#33 Chronic obstructive pulmonary disease

The following combinations of groups of search terms using ‘and’ was performed: RCT (#16) and COPD (#33) and NIV (#21) and Exercise (#32); RCT (#16) and COPD (#33) and NIV (#21); RCT (#16) and COPD (#33) and Exercise (#32); NIV (#21) and Exercise (#32).

 

Latin American and Caribbean Health Science Information Database (LILACS)

	RCT	NIV	Exercise	COPD
Search terms	#1 Random$ #2 Placebo$ #3 Trial$ #4 #1 or #2 or #3	#5 Respiration, artificial	#6 Exercise$ #7 Physical$ #8 Train$ #9 Rehabilitat$ #10 Conditioning #11 Ergometry #12 Treadmill #13 Endurance #14 #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13	#15 COPD #16 COAD #17 DPOC #18 Emphysema$ #19 Bronchit #20 #15 or #16 or #17 or #18 or #19

The following combinations of groups of search terms using ‘and’ was performed: RCT (#4) and COPD (#20) and NIV (#5) and Exercise (#14); RCT (#4) and COPD (#20) and NIV (#5); RCT (#4) and COPD (#20) and Exercise (#14); NIV (#5) and Exercise (#14).

MEDLINE

	RCT	NIV	Exercise	COPD
Search terms	#1 Randomized controlled trial.pt. #2 Controlled clinical trial.pt. #3 Randomized.ab. #4 Placebo.ab. #5 Randomly.ab. #6 Trial.ab. #7 Groups.ab. #8 #1 or #2 or #3 or #4 or #5 or #6 or #7 #9 (animals not (humans and animals)).sh. #10 #8 not #9	#11 Respiration, Artificial/ or Positive‐Pressure Respiration #12 Intermittent Positive‐Pressure Ventilation #13 #11 or #12	#14 Exercise/ or Exercise Therapy/or Exercise Test/ or Exercise Tolerance #15 "Physical Education and Training" #16 Rehabilitation #17 Physical Fitness/or Physical Endurance #18 Ergometry #19 Walking #20 Bicycling #21 Upper Extremity #22 #14 or #15 or #16 or #17 or #18 or #19 or #20 or #21	#23 Pulmonary Disease, Chronic Obstructive

The following combinations of groups of search terms using ‘and’ was performed: RCT (#10) and COPD (#23) and NIV (#13) and Exercise (#22); RCT (#10) and COPD (#23) and NIV (#13); RCT (#10) and COPD (#23) and Exercise (#22); NIV (#13) and Exercise (#22).

PEDro

PEDro is a database of randomised trials and systematic reviews; therefore a randomised controlled trial filter was not required. PEDro was searched with the following terms: COPD and NIV or pressure support or ventilation and exercise or training; COPD and NIV or pressure support or ventilation; NIV or pressure support or ventilation and exercise or training; NIV and exercise.

PsycINFO

	RCT	NIV	Exercise	COPD
Search terms	#1 Randomized controlled trial.pt. #2 Controlled clinical trial.pt. #3 Randomized.ab. #4 Placebo.ab. #5 Randomly.ab. #6 Trial.ab. #7 Groups.ab. #8 #1 or #2 or #3 or #4 or #5 or #6 or #7 #9 (animals not (humans and animals)).sh. #10 #8 not #9	#11 Respiration, Artificial/ or Positive‐Pressure Respiration #12 Intermittent Positive‐Pressure Ventilation #13 Ventilators mechanical #14 Exp Artificial Respiration #15 #11 or #12 or #13 or #14	#16 Exercise/ or Exercise Therapy/ or Exercise Test/ or Exercise Tolerance #17 "Physical Education and Training" #18 Rehabilitation #19 Physical Fitness/ or Physical Endurance #20 Ergometry #21 Walking #22 Bicycling #23 Upper Extremity #24 #16 or #17 or #18 or #19 or #20 or #21 or #22 or #23	#25 Pulmonary Disease, Chronic Obstructive #26 Lung diseases obstructive #27 Exp Lung Disorders #28 #25 or #26 or #27

The following combinations of groups of search terms using ‘and’ was performed: RCT (#10) and COPD (#28) and NIV (#15) and Exercise (#24); RCT (#10) and COPD (#28) and NIV (#15); RCT (#10) and COPD (#28) and Exercise (#24); NIV (#15) and Exercise (#24). 
 
  

PubMed

	RCT	NIV	Exercise	COPD
Search terms	#1 Randomized controlled trial #2 Controlled clinical trial #3 Randomized #4 Placebo #5 Randomly #6 Trial #7 Groups #8 #1 or #2 or #3 or #4 or #5 or #6 or #7	#9 Respiration, artificial #10 Positive pressure respiration #11 Intermittent positive‐pressure ventilation #12 #9 or #10 or #11	#13 Exercise #14 Exercise test #15 Exercise therapy #16 Exercise tolerance #17 Physical education and training #18 Rehabilitation #19 Physical endurance #20 Physical fitness #21 Ergometry #22 Walking #23 Bicycling #24 Upper extremity #25 #13 or #14 or #15 or #16 or #17 or #18 or #19 or #20 or #21 or #22 or #23 or #24	#26 Pulmonary disease, chronic obstructive

The following combinations of groups of search terms using ‘and’ was performed: RCT (#8) and COPD (#26) and NIV (#12) and Exercise (#25); RCT (#8) and COPD (#26) and NIV (#12); RCT (#8) and COPD (#26) and Exercise (#25); NIV (#12) and Exercise (#25).

Appendix 2. Summary of characteristics of included studies

	Bianchi 2002	Hawkins 2002	Johnson 2002	Reuveny 2005	Toledo 2007	van 't Hul 2006
Sample size	19	19	22	19	18	29
Disease severity	Moderate to severe	Very severe	Severe to very severe	Severe to very severe	Severe to very severe	Severe
Setting	Outpatient, hospital‐based	Outpatient, hospital‐based	Outpatient, centre‐based	Outpatient, centre‐based	Outpatient, centre‐based	Outpatient, centre‐based
Programme length	6 weeks	6 weeks	6 weeks	8 weeks	12 weeks	8 weeks
Supervised sessions/wk	3	3	2	2	3	3
Session duration	30 minutes	30 minutes	20 minutes	45 minutes	30 minutes	45 minutes
Type of exercise	Cycling	Cycling	Treadmill	Treadmill	Treadmill	Cycling
Training intensity	50% to 70% peak work capacity	70% peak work capacity	50% to 60% maximum METs	65% to 70% initial maximum walking speed	70% baseline walk speed	65% peak work capacity
Comparison	Unassisted versus NIV	Unassisted versus NIV	Unassisted versus NIV	Unassisted versus NIV	Unassisted versus NIV	Sham NIV versus NIV
NIV mode	PAV	PAV	Bilevel	Bilevel	Bilevel	IPS
NIV settings	FA: 3.5 (1.6) cmH2O/L/s VA: 6.6 (2.2) cmH2O/L CPAP: 2 cmH2O	FA: 3.6 (0.7) cmH2O/L/s VA: 12.7 (1.5) cmH2O/L CPAP: 0	IPAP: 8 to 12 cmH2O EPAP: 2 cmH2O	IPAP: 7 to 10 cmH2O EPAP: 2 cmH2O	IPAP: 10 to 15 cmH2O EPAP: 4 to 6 cmH2O	IPS: 5 versus IPS: 10 cmH2O EPAP: 0 cmH2O

CPAP: continuous positive airway pressure; EPAP: expiratory positive airway pressure; FA: flow assist; IPAP: inspiratory positive airway pressure; IPS: inspiratory pressure support; METs: metabolic equivalents; NIV: non‐invasive ventilation; PAV: proportional assist ventilation; VA: volume assist.

Data and analyses

Comparison 1. Non‐invasive ventilation during exercise training versus exercise training alone or exercise training with sham non‐invasive ventilation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exercise capacity: peak cycle work rate (watts)	3	57	Mean Difference (IV, Fixed, 95% CI)	6.34 [‐1.66, 14.34]
2 Exercise capacity: peak VO2 (L/min)	2	37	Mean Difference (IV, Fixed, 95% CI)	0.12 [‐0.08, 0.31]
3 Exercise capacity: percentage change	4		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
3.1 Percentage change peak work rate	3	60	Mean Difference (IV, Fixed, 95% CI)	17.01 [6.83, 27.19]
3.2 Percentage change constant work rate endurance time	2	48	Mean Difference (IV, Fixed, 95% CI)	58.66 [3.72, 113.60]
4 Exercise capacity: constant work rate cycle endurance time (minutes)	2	48	Mean Difference (IV, Fixed, 95% CI)	3.62 [‐0.17, 7.41]
5 Health‐related quality of life: St George's Respiratory Questionnaire	2		Mean Difference (Random, 95% CI)	Subtotals only
5.1 SGRQ: total score (points)	2	48	Mean Difference (Random, 95% CI)	2.45 [‐2.30, 7.20]
5.2 SGRQ: symptoms (points)	2	48	Mean Difference (Random, 95% CI)	0.87 [‐10.19, 11.93]
5.3 SGRQ: activity (points)	2	48	Mean Difference (Random, 95% CI)	0.05 [‐14.92, 15.02]
5.4 SGRQ: impacts (points)	2	48	Mean Difference (Random, 95% CI)	0.11 [‐6.82, 7.05]
6 Training intensity: final training session (% baseline peak work capacity)	3	67	Mean Difference (IV, Random, 95% CI)	13.31 [0.05, 26.57]
7 Physiological outcomes: isoload lactate (mmol/L)	2	37	Mean Difference (IV, Fixed, 95% CI)	‐0.97 [‐1.58, ‐0.36]
8 Physiological outcomes: peak exercise lactate (mmol/L)	3	56	Mean Difference (IV, Random, 95% CI)	‐0.35 [‐1.10, 0.41]
9 Physiological outcomes: isotime exercise minute ventilation (L/min)	3	53	Mean Difference (Fixed, 95% CI)	‐0.08 [‐2.82, 2.67]
10 Physiological outcomes: peak exercise minute ventilation (L/min)	3	47	Mean Difference (Fixed, 95% CI)	2.68 [‐2.02, 7.37]
11 Physiological outcomes: change in VO2 at anaerobic threshold (L/min)	2	38	Mean Difference (Fixed, 95% CI)	0.08 [‐0.09, 0.24]
12 Dyspnoea: isotime exercise dyspnoea (Borg scale)	3	56	Mean Difference (Fixed, 95% CI)	‐0.18 [‐1.09, 0.72]
13 Dropouts	5	151	Odds Ratio (M‐H, Fixed, 95% CI)	1.26 [0.61, 2.59]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Bianchi 2002.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: male participants with COPD diagnosed by American Thoracic Society criteria; referred to outpatient pulmonary rehabilitation; stable clinical condition Exclusion criteria: chronic respiratory failure; other organ failure; cancer; inability to co‐operate; arterial exercise–induced hypertension (systolic blood pressure > 200 mmHg or diastolic blood pressure > 130 mmHg) Statistical analysis: between groups: repeated measures analysis of variance; differences between dropouts and programme completers: Fisher’s exact test; P value < 0.05 significance
Participants	Participants recruited from: referred to outpatient multi‐disciplinary pulmonary rehabilitation programme Number screened: 83 Sample size: 9 NIV group; 10 control group Age mean (range), years: 64 (61 to 67) NIV group; 65 (61 to 69) control group Gender: 0 female participants NIV group; 0 female participants control group FEV1 mean (SD) % predicted: 48 (19) NIV group; 40 (12) control group FVC mean (SD) % predicted: 77 (15) NIV group; 74 (19) control group RV mean (SD) % predicted: 161 (69) NIV group; 181 (49) control group PaO2 mean (SD) kPa: 10.0 (1.1) NIV group; 10.0 (1.1) control group PaCO2 mean (SD) kPa: 5.2 (0.6) NIV group; 5.2 (0.5) control group PImax mean (SD) cmH2O: 82.9 (25.8) NIV group; 72.4 (23.4) control group PEmax mean (SD) cmH2O: 147.9 (26.8) NIV group, 139.2 (25.1) control group Domiciliary NIV: nil Long‐term oxygen therapy: nil
Interventions	NIV: proportional assist ventilation: flow assist 3.5 (1.6) cmH2O/L/s, volume assist 6.6 (2.2) cmH2O/L with EPAP 2 cmH2O during exercise training Control: unassisted exercise training Supplemental oxygen during exercise training: no Type of exercise training: cycle ergometry Intensity of training: 50% to 70% of peak work capacity Number of sessions per week: 3 Duration of each session: 30 minutes, supervised Total duration of training: 6 weeks Number of training sessions attended (/18): NIV group 17.4 ± 1.3 sessions; control group 17.0 ± 1.8 sessions Programme setting: outpatient, hospital‐based
Outcomes	Maximal cycle ergometry: unassisted, load increased 10 watts/min; six‐minute walk test; dyspnoea; leg discomfort; St George's Respiratory Questionnaire Evaluation: assessment performed at baseline and immediately following the training programme
Dropouts	A total of 33 participants were recruited. Of the 18 randomly assigned to NIV, 9 completed the protocol (50% dropout rate). Reasons for dropout: intolerance of NIV (N = 5); acute exacerbation of COPD (N = 2); exercise‐induced hypertension (N = 1); and unexpected coronary disease (N = 1). Of the 15 randomly assigned to the control group, 10 completed the protocol (33% dropout rate). Reasons for dropout: acute exacerbation of COPD (N = 2); exercise‐induced hypertension (N = 2); unexpected coronary disease (N = 1)
Funding	No information provided
Notes	Country: Italy Study author contacted: full response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote (from correspondence): "The sequence of allocation was computer generated"
Allocation concealment (selection bias)	Low risk	Quote (from correspondence): "...concealed envelopes"
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote (from report): "Patients and investigators were unblinded during the study"
Blinding of outcome assessment (detection bias) All outcomes	High risk	Quote (from report): "Patients and investigators were unblinded during the study."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: Total number of participants recruited, number of dropouts from each group and reasons for dropping out are reported. Dropout rate was higher in the group that trained with NIV (44%) compared with the group that trained without NIV (33%). Five of the eight dropouts in the NIV group were due to NIV intolerance. Both intention‐to‐treat analyses and per‐protocol analyses were performed for key outcomes
Selective reporting (reporting bias)	Low risk	Comment: Published report contains all expected outcomes, including those that were prespecified
Other bias	Low risk	None evident

Hawkins 2002.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: severe COPD; current smoker or had a history of smoking Exclusion criteria: orthopaedic or cardiovascular contraindications to exercise Statistical analysis: within groups: paired t‐tests; between groups: unpaired t‐tests; strength of relationship between physiological variables: Pearson’s correlation coefficient; P value < 0.05 significance
Participants	Participants recruited from: respiratory medicine clinic Number screened: 34 Sample size: 10 NIV group; 9 control group Age mean (SD), years: 66 (7) NIV group; 68 (9) control group Gender: 0 female participants NIV group; 2 female participants control group FEV1 mean (SD) % predicted: 28 (7) NIV group; 26 (7) control group RV/TLC mean (SD) %: 62 (9) NIV group; 59 (15) control group PaO2 mean (SD) kPa: 8.6 (0.9) NIV group; 8.1 (1.2) control group PaCO2 mean (SD) kPa: 5.6 (0.7) NIV group; 5.8 (0.9) control group Domiciliary NIV: no information provided Long‐term oxygen therapy: no information provided
Interventions	NIV: proportional assist ventilation: flow assist 3.6 (0.7) cmH2O/L/s, volume assist 12.7 (1.5) cmH2O/L during exercise training Control: unassisted exercise training Supplemental oxygen during exercise training: yes (two participants with unassisted training; one participant with non‐invasive ventilation during exercise training) Type of exercise training: cycle ergometry Intensity of training: initially 70% of peak work capacity, then increased progressively by 5 watts when able to maintain the existing work rate for cycle of 30 minutes Number of sessions per week: 3 Duration of each session: 30 minutes, supervised Total duration of training: 6 weeks Number of training sessions attended (/18): not reported Programme setting: outpatient, hospital‐based
Outcomes	Isoload arterialised venous blood lactate concentration; maximal cycle ergometry: unassisted, workload increased 5 to 10 watts/min; constant power cycle ergometry: unassisted, at 70% of baseline peak work capacity Evaluation: Baseline testing was performed on two occasions before the training programme was begun, and on two occasions within one week of completion of the training programme
Dropouts	A total of 29 participants were recruited. Of the 14 randomly assigned to NIV, 10 completed the protocol (29% dropout rate). Of the 15 randomly assigned to the control group, 9 completed the protocol (40% dropout rate). Reasons for dropout: acute exacerbation of COPD (N = 4); non‐compliance with the exercise programme (N = 4); non‐pulmonary hospitalisation (N = 2)
Funding	Peter Hawkins was funded by grant (F97/1) from the British Lung Foundation. Dimitra Nikoletou was funded by Respironics Inc, which also provided the ventilators used in the study
Notes	Country: United Kingdom Study author contacted: full response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote (from correspondence): "The envelopes were shuffled and then numbered in sequence"
Allocation concealment (selection bias)	Low risk	Quote (from report): "Patients were randomised using sealed envelopes" Quote (from correspondence): "The envelopes were opaque" Quote (from correspondence): "Subjects were randomised in groups of 6" Comment: It may have been possible to predict group allocation for 4/29 participants because the randomisation blocks were of a fixed size at a single centre and investigators were not blinded to group allocation. However, as the proportion of participants potentially affected was very low, the overall judgement of risk of bias was low
Blinding of participants and personnel (performance bias) All outcomes	High risk	Comment: Participants and trainers would not have been blinded, as sham non‐invasive ventilation was not used
Blinding of outcome assessment (detection bias) All outcomes	High risk	Quote (from correspondence): "The investigators performing the post exercise test were aware which group the subject was in"
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: Total number of participants recruited, number of dropouts from each group and reasons for dropping out are reported. Numbers of dropouts were reasonably balanced between groups. Per‐protocol analyses were performed and data reported for key outcomes. An intention‐to‐treat analysis was not reported
Selective reporting (reporting bias)	Low risk	Comment: Published report contains all expected outcomes, including those that were prespecified
Other bias	Low risk	None evident

Johnson 2002.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: diagnosis of COPD; FEV1 < 50% predicted; ability to walk on a treadmill; referred to outpatient pulmonary rehabilitation Exclusion criteria: exertional angina; congestive cardiac failure; valvular heart disease; uncontrolled cardiac dysrhythmias; other conditions limiting ability to exercise or use a nasal mask Statistical analysis: within groups: paired t‐tests; between groups: unpaired t‐tests; P value < 0.05 significance
Participants	Participants recruited from: referred to outpatient pulmonary rehabilitation programme Number screened: 39 Sample size: 11 NIV group; 11 control group; 10 heliox group Age mean (SD), years: 69 (9) NIV group; 67 (8) control group; 72 (9) heliox group Gender: 3 female participants NIV group; 4 female participants control group; 6 female participants heliox group FEV1 mean (SD) % predicted: 32 (9) NIV group; 31 (11) control group; 34 (13) heliox group FVC mean (SD) % predicted: 57 (15) NIV group; 53 (10) control group; 57 (16) heliox group RV mean (SD) % predicted: 203 (63) NIV group; 199 (73) control group; 191 (66) heliox group PaO2 mean (SD) mmHg: 72 (10) NIV group; 69.2 (9) control group; 70.3 (6.0) heliox group PaCO2 mean (SD) mmHg: 42.4 (4.3) NIV group; 43.3 (4.5) control group; 43.5 (3.7) heliox group Domiciliary NIV: no information provided Long‐term oxygen therapy: no information provided
Interventions	NIV: bilevel support: IPAP 8 to 12 cmH2O, EPAP 2 cmH2O during exercise training Control: humidified air at 10 L/min via a non‐rebreather mask during exercise training Heliox: 10 L/min of humidified heliox (79% helium, 21% oxygen) via a non‐rebreather mask during exercise training Supplemental oxygen during exercise training: yes. Oxygen was titrated to maintain oxygen saturation of at least 90% throughout the entire testing and training periods. Maximum oxygen delivery flow rate during unassisted training: 2.7 ± 1.6 L/min; during training with NIV: 2.6 ± 1.8 L/min. Information was not provided regarding the number of participants in each group who used supplemental oxygen Type of exercise training: treadmill Intensity of training: 50% to 60% of maximum METS based on initial incremental treadmill test, then increased (speed and grade) once 20 minutes of continuous exercise was achieved at this level Number of sessions per week: 2 Duration of each session: 20 minutes, supervised Total duration of training: 6 weeks Number of training sessions attended (/12): Quote: "Nearly all of the patients were present for 10 sessions" Programme setting: outpatient, centre‐based
Outcomes	Incremental maximal treadmill test: unassisted, workload increased 0.5 mph every 4 minutes, when predetermined maximum speed achieved, incline was increased at 3% every 4 minutes Evaluation: Testing was performed the week before and within the week following completion of training
Dropouts	A total of 39 participants were recruited. Of the 15 randomly assigned to NIV, 11 completed the protocol (27% dropout rate). Of the 13 randomly assigned to the control group, 11 completed the protocol (15% dropout rate). Of the 11 randomly assigned to Heliox, 10 completed the protocol (9% dropout rate). Reasons for dropout: exertional angina (N = 1); congestive heart failure (N = 1); flare of chronic liver disease (N = 1); acute exacerbation of COPD (N = 1); tibial fracture (N = 1); scheduling conflict (N = 1); non‐compliance (N = 1)
Funding	Funded through Brooke Army Medical Center local research funds
Notes	Country: United States of America Study author contacted: no response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Comment: No information provided
Allocation concealment (selection bias)	Unclear risk	Comment: No information provided
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote (from report): "No attempt was made to blind the Heliox group and unassisted exercise training group patients to their training modality" Comment: It would not have been possible to blind participants or trainers in the NIV training group either
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	No information provided regarding blinding of outcome assessors. This probably was not done
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: Total number of participants recruited, number of dropouts from each group and reasons for dropping out are reported. Numbers of dropouts were reasonably well balanced between groups. Per‐protocol analyses (not intention‐to‐treat analyses) were performed for key outcomes
Selective reporting (reporting bias)	Low risk	Comment: Published report contains all expected outcomes, including those that were prespecified
Other bias	High risk	Quote (from report): "Review of the patient's exercise logs showed that they uniformly exercised more than the supervised sessions, with an average total of approximately four times per week. Hence, our comparison of training modalities represented only a portion of the patients exercise sessions since the others were done at home" Comment: Participants performed two supervised sessions per week according to group allocation (unassisted, with heliox or with non‐invasive ventilation). The unsupervised, unassisted training at home (without non‐invasive ventilation or without Heliox) may have confounded the results (contamination) Comment: baseline difference between groups for key outcomes (non‐invasive ventilation group significantly lower compared with unassisted group for unassisted exercise time and unassisted maximum workload)

Reuveny 2005.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: diagnosis of severe COPD based on smoking history, clinical findings and pulmonary function testing; FEV1 < 40% predicted and < 12% reversibility; ventilatory limitation to exercise, defined as exercise breathing reserve < 5 L/min Exclusion criteria: cardiovascular, orthopaedic or neuromuscular disease that could limit exercise capacity Statistical analysis: within groups: paired t‐tests; between groups: ANOVA; linear regression for differences in breathing pattern; P value < 0.05 significance
Participants	Participants recruited from: no information provided Number screened: no information provided Sample size: 9 NIV group; 10 control group Age mean (SD), years: 64 (9) NIV group; 63 (9) control group Gender: 1 female participant; 18 male participants FEV1 mean (SD) % predicted: 32 (4) NIV group; 33 (9) control group FEV1/FVC mean (SD) %: 59 (16) NIV group; 58 (16) control group RV mean (SD) % predicted: 194 (34) NIV group; 215 (56) control group RV/TLC mean (SD) % predicted: 164 (19) NIV group; 165 (17) control group Domiciliary NIV: no information provided Long‐term oxygen therapy: nocturnal oxygen used in four participants from the NIV group and two participants from the control group
Interventions	NIV: bilevel support: IPAP 7 to 10 cmH2O, EPAP 2 cmH2O during exercise training Control: unassisted exercise training Supplemental oxygen during exercise training: yes. Supplemental oxygen was provided to 8/10 participants in the control group and 9/9 participants in the NIV group to maintain oxygen saturation of at least 92% during training Type of exercise training: treadmill Intensity of training: initially started at > 2 km/h based on what the participant could maintain for 15 minutes. Speed increased at 0.2 km/h each week after participants could maintain the target intensity for 45 minutes, with the aim of achieving 65% to 70% of the initial maximum walking speed. Zero slope Number of sessions per week: 2 Duration of each session: 45 minutes Total duration of training: 8 weeks Number of training sessions attended (/16): not reported Programme setting: outpatient, centre‐based
Outcomes	Maximal cycle ergometry: unassisted, load increased 15 watts/min Evaluation: Testing was performed within one week before and at the end of the programme
Dropouts	A total of 24 participants were recruited. Of the 12 randomly assigned to NIV, 9 completed the protocol (25% dropout rate). Reasons for dropout: failed to adjust to the mask (N = 3). Of the 12 randomly assigned to the control group, 10 completed the protocol (17% dropout rate). Reasons for dropout: lung transplant (N = 1); back pain (N = 1)
Funding	Supported in part by a grant from the Israel Lung Association, Tel Aviv
Notes	Country: Israel Study author contacted: no response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Comment: no information provided
Allocation concealment (selection bias)	Unclear risk	Quote (from report): "sealed envelope." Unsure if opaque etc
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote (from report): "Although it was not feasible to blind the patients or the trainers to the condition of their exercise training"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote (from report): "...the investigators conducting exercise testing and interpreting the exercise data did so without knowing to which group the patients were assigned"
Incomplete outcome data (attrition bias) All outcomes	High risk	Quote (from report): "Of the 24 patients 5 did not complete the study: 3 from the BiPAP group, who failed to adjust to the mask, and 2 from the control group—one due to lung transplant and one because of back pain" Comment: Total number of participants recruited, number of dropouts from each group and reasons for dropping out are reported. Numbers of dropouts were reasonably well balanced between groups. However, all dropouts in the non‐invasive ventilation group were due to mask intolerance, which may have introduced bias, as only participants who could tolerate non‐invasive ventilation were included in the analysis, as opposed to those who were allocated. An intention‐to‐treat analysis was not reported
Selective reporting (reporting bias)	High risk	Comment: Study report fails to include results for a key outcome that would be expected to have been reported for such a study Between‐group results were not reported for post‐training exercise capacity or for most physiological outcomes. Primary study outcome(s) were not stated. Instead, the study authors report that certain significant changes (when compared with baseline) were detected in the group that trained with NIV only (not the control group). The only between‐groups comparisons reported were for training intensity and change in peak exercise tidal volume
Other bias	Unclear risk	Quote (from report): "...no improvement was found in the control group. This may have been due to the relatively short twice‐weekly training, while effective protocols reported in the literature utilized between three and five training sessions per week" Comment: As the control group did not improve with pulmonary rehabilitation, the efficacy of the rehabilitation programme appears questionable. The group that trained with non‐invasive ventilation did improve. However, as both trainers and participants were not blinded to the intervention, bias cannot be ruled out, although the progression of training intensity was standardised; this should have helped to ensure that participants were exposed to the same type of training programme

Toledo 2007.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: clinical and spirometric diagnosis of COPD; FEV1 < 60% predicted and FEV1/FVC < 70%; clinically stable for a minimum of six months   Exclusion criteria: no associated cardiovascular, orthopaedic or neuromuscular disorders or reactive hypertension related to effort that would impede involvement in the programme Statistical analysis: within groups: Wilcoxon test; between‐group comparisons: Mann‐Whitney U‐test; P value < 0.05 significance
Participants	Participants recruited from: referred to respiratory physical therapy service at a university hospital Number screened: no information provided Sample size: 9 NIV group; 9 control group Age mean (SD), years: 67 (11) NIV group; 67 (9) control group Gender: numbers not stated FEV1 mean (SD) % predicted: 33 (10) NIV group; 34 (8) control group FVC mean (SD) % predicted: 63 (18) NIV group; 55 (11) control group Domiciliary NIV: no information provided Long‐term oxygen therapy: no information provided
Interventions	NIV: bilevel support: IPAP 10 to 15 cmH2O, EPAP 4 to 6 cmH2O during exercise training Control: unassisted exercise training Supplemental oxygen during exercise training: no Type of exercise training: treadmill Intensity of training: 70% of maximum baseline speed Number of sessions per week: 3 Duration of each session: 30 minutes Total duration of training: 12 weeks Number of training sessions attended (/36): not reported Programme setting: outpatient, centre‐based
Outcomes	Incremental treadmill test, increasing 0.5 km/h every 2 minutes; blood lactate; respiratory muscle strength Evaluation: no information provided
Dropouts	A total of 18 participants completed the protocol (NIV group N = 9, control group N = 9). No information was given regarding dropouts
Funding	No information provided
Notes	Country: Brazil Study author contacted: partial response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Comment: no information provided
Allocation concealment (selection bias)	Low risk	Quote (from correspondence): "...randomised into 2 groups by opaque and sealed envelopes"
Blinding of participants and personnel (performance bias) All outcomes	High risk	Comment: It would not have been possible to blind participants or trainers, as sham non‐invasive ventilation was not used
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote (from correspondence): "Tests were made with 3 assessors and they did not know the group allocation"
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Comment: Number of dropouts from each group (if any) and reasons for dropping out are not reported. Total number of participants recruited was stated (N = 18), and all of the 18 participants were included in the analysis
Selective reporting (reporting bias)	High risk	Comment: Study report fails to include results for a key outcome that would be expected to have been reported for such a study Between‐group results were not reported for post‐training exercise capacity or for most of the physiological outcomes. Primary study outcome(s) were not stated. It is not clear whether the study was powered to detect a change between groups. Instead, the study authors report that certain significant changes (when compared with baseline) were detected in the group that trained with non‐invasive ventilation only (not the control group). The only between‐group comparison reported was isoload lactate
Other bias	Low risk	None evident

van 't Hul 2006.

Methods	Study design: parallel randomised controlled trial Inclusion criteria: diagnosis of COPD based on GOLD criteria; FEV1 < 60% predicted; breathing reserve at maximal exercise < 20% of maximum voluntary ventilation; peak minute ventilation < 50 L/min; resting PaO2 > 60 mmHg; SpO2 at maximal exercise > 85%; 40 to 75 years of age Statistical analysis: within groups: paired t‐tests or Wilcoxon test; between groups: independent samples  t‐tests or Mann‐Whitney U‐test; strength of associations: Pearson’s correlation coefficient; P < 0.05 significance
Participants	Participants recruited from: information not available from trial report Number screened: information not available from trial report Sample size: 15 NIV group; 14 control group Age mean (SD), years: 70 (5) NIV group; 71 (4) control group Gender: 4 female participants NIV group; 1 female participant control group FEV1 mean (SD) % predicted: 41 (10) NIV group; 38 (9) control group FVC mean (SD) % predicted: 87 (14) NIV group; 76 (14) control group FEV1/FVC mean (SD) %: 34 (6) NIV group; 36 (8) control group RV mean (SD) % predicted: 161 (18) NIV group; 164 (50) control group PImax cmH2O: 58 (20) NIV group; 59 (21) control group PEmax cmH2O: 100 (44) NIV group; 111 (28) control group Domiciliary NIV: nil Long‐term oxygen therapy: no information provided
Interventions	NIV: inspiratory pressure support 10 cmH2O during exercise training Control: sham IPS 5 cmH2O during exercise training Supplemental oxygen during exercise training: no Type of exercise training: cycle ergometry Intensity of training: initially 65% of peak work capacity, then increased progressively by 5% of peak work rate when able to cycle for > 15 minutes Number of sessions per week: 3 Duration of each session: 45 minutes, supervised Total duration of training: 8 weeks Number of training sessions attended (/24): Participants were allowed to miss up to 4 training sessions, provided that the training period was extended (from 8 weeks to a maximum of 9 to 10 weeks) to ensure that each participant completed all 24 training sessions Program setting: outpatient, centre‐based
Outcomes	Incremental shuttle walk test; constant work rate cycle endurance time (75% of baseline peak work rate); St. George's Resiratory Questionnaire Evaluation: baseline measurements performed within 2 weeks before the start of the training period. Post‐training measurements took place within 2 weeks following the final training session
Dropouts	A total of 37 participants were recruited. Of the 19 randomly assigned to NIV, 15 completed the protocol (21% dropout rate). Reasons for dropout: acute exacerbation of COPD (N = 3); worsening general fatigue (N = 1). Of the 18 randomly assigned to the control group, 14 completed the protocol (22% dropout rate). Reasons for dropout: exacerbation of COPD (N = 3); cerebrovascular accident (N = 1)
Funding	The study was supported by a grant from the Dutch Lung Foundation
Notes	Country: The Netherlands Study author contacted: full response
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote (from correspondence): "...concealed envelopes and block randomisation. Each block (envelope) contained four cards. Two cards with 'experimental condition' and two cards with 'control condition' written on it. When a patient got to be randomised an independent observer drew one of the cards out of this envelope. After four procedures the envelope was empty and the next envelope was used, and so on"
Allocation concealment (selection bias)	Low risk	Quote (from report): "Patients were randomly allocated (concealed envelopes)"
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote (from report): "...not possible to blind the three physiotherapists to the inspiratory pressure support (IPS) intensity patients were training with, but patients were kept naive with respect to randomisation outcome"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote (from report): "All measurements were performed by one independent investigator (A. van't Hul), who was not involved in the training and who was kept blinded to randomisation outcome"
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: Total number of participants recruited, number of dropouts from each group and reasons for dropping out are reported. Both intention‐to‐treat analyses and per‐protocol analyses were performed for key outcomes
Selective reporting (reporting bias)	Low risk	Comment: Published report contains all expected outcomes, including those that were prespecified
Other bias	Low risk	None evident

ANOVA: analysis of variance; COPD: chronic obstructive pulmonary disease; EPAP: expiratory positive airway pressure; FEV₁: forced expiratory volume in one second; FVC: forced vital capacity; GOLD: Global Initiative for Chronic Obstructive Lung Disease; IPAP: inspiratory positive airway pressure; METS: metabolic equivalents; NIV: non‐invasive ventilation; PaO₂: arterial partial pressure of oxygen; PaCO₂: arterial partial pressure of carbon dioxide; PE_max: maximal expiratory pressure; PI_max: maximal inspiratory pressure; RCT: randomised controlled trial; RV: residual volume; SD: standard deviation; SpO₂: oxygen saturation; TLC: total lung capacity

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Allan 2009	Not exercise training
Amann 2010	Not exercise training
Ambrosino 2000	Not an RCT
Ambrosino 2004	Not an RCT
Ambrosino 2006	Not an RCT
Ambrosino 2011	Not an RCT
Anonymous 2007	Not an RCT
Arad 1992	Not an RCT
Araujo 2005	Not an RCT
Bach 1992	Not an RCT
Bach 1993	Not an RCT
Baer 1989	Not an RCT
Barakat 2007	Not an RCT
Bianchi 1998	Not exercise training
Borghi‐Silva 2005	Not exercise training
Borghi‐Silva 2008	Not exercise training
Borghi‐Silva 2009	Not exercise training
Borghi‐Silva 2010	Wrong comparison
Boye 1994	Not NIV during exercise
Bullemer 1999	Not an RCT
Carrascossa 2010	Not exercise training
Chaturvedi 2011	Not exercise training
Chen 2012	Not COPD
Chiang 2006a	Not COPD
Chiang 2006b	Not exercise training
Corner 2010	Not an RCT
Costa 2006	Not an RCT
Costes 2003	Not an RCT
De Backer 2010	Not an RCT
den Hartog 2003	Not an RCT
Dieperink 2006	Not an RCT
Dolmage 1997	Not exercise training
Dreher 2007	Not exercise training
Dreher 2008	Not an RCT
Dreher 2009	Not exercise training
Dreher 2010	Not COPD
Duiverman 2008	Not NIV during exercise
Duiverman 2011	Not NIV during exercise
Dyer 2011	Not stable COPD
Gallagher 1989	Not COPD
Garrod 2000	Not NIV during exercise
Hernandes 2012	Not an RCT
Hernandez 2001	Not exercise training
Highcock 2003	Not exercise training
Hussain 2011	Not exercise training
Jackson 1991	Not an RCT
Keilty 1994	Not exercise training
Kleinsasser 2004	Not COPD
Kohnlein 2009	Not NIV during exercise
Kyroussis 2000	Not exercise training
Maltais 1995	Not exercise training
Martin 2005	Not an RCT
Medvedev 2007	Not COPD
Menadue 2009a	Not exercise training
Moga 2012	Not an RCT
Monteiro 2012	Not an RCT
Nicolini 2013	Not exercise training
O'Donnell 1988a	Not exercise training
O'Donnell 1988b	Not exercise training
Oliveira 2010	Not exercise training
Padkao 2010	Not exercise training
Pepin 2010	Not an RCT
Pessoa 2012	Not exercise training
Petrof 1990	Not exercise training
Pires Di Lorenzo 2003	Wrong comparison
Poggi 2006	Not exercise training
Polkey 1996	Not exercise training
Polkey 2000	Not exercise training
Poon 1987	Not an RCT
Porszasz 2013	Not exercise training
Puhan 2004	Not an RCT
Revill 2000	Not exercise training
Ricci 2013	Not an RCT
Rochester 2013	Not an RCT
Rodrigues 2013	Not exercise training
Schmidt 1999	Not an RCT
Schonhofer 2003	Not an RCT
Schonhofer 2008	Not an RCT
Skobel 2011	Not an RCT
Soo Hoo 2003	Not an RCT
Spruit 2007	Not an RCT
van't Hul 2002	Not an RCT
van't Hul 2004	Not exercise training
Vitacca 2006	Not exercise training
Walterspacher 2013	Not exercise training
Wibmer 2013	Not NIV during exercise
Wijkstra 2011	Not an RCT
ZuWallack 2008	Not an RCT

Differences between protocol and review

The following changes were made to the published protocol (Menadue 2009b) during the review process: an addition was made to the criteria for the types of interventions that were eligible for inclusion in the review whereby studies that involved the use of nocturnal NIV were included only if both the actively treated group and the control group received nocturnal NIV; rather than data extraction being performed by one review author and verified by another review author, data from the included studies were extracted independently by two review authors; post training outcome data were only extracted if study participants were evaluated off NIV (e.g. unassisted test of exercise capacity); when dichotomous data were combined, the treatment effect was defined as the OR with 95% CI; if ITT analyses were not reported, data from the per‐protocol analyses were extracted for use in the meta‐analysis; a fixed effect model was used for analyses with I² less than 30%, otherwise a random effects model was used; ventilator settings (including level of ventilatory assistance and mode of ventilation) were added as a factor to be considered for subgroup analyses; sensitivity analyses were performed for outcomes that included data from studies for which baseline differences between groups were accounted for by using change from baseline data rather than post intervention data. In addition, a summary of findings table was presented. Both the summary of findings table and the outcomes presented in the table were not planned a priori.

Contributions of authors

Collette Menadue: initiation and writing of protocol and manuscript, data extraction and analysis.

Amanda Piper: protocol development, data extraction, manuscript review.

Alex van't Hul: protocol development, manuscript review.

Keith Wong: protocol development, data extraction, manuscript review.

Sources of support

Internal sources

• Royal Prince Alfred Hospital, Australia.

External sources

• No sources of support supplied

Declarations of interest

Amanda Piper has received honoraria for educational presentations conducted on behalf of Respironics, Australia; ResMed, Australia; and Weinmann, Germany. She has also received a grant from the ResMed Foundation. The sleep laboratory of Collette Menadue, Amanda Piper and Keith Wong has previously received industry‐sponsored project grants from ResMed, Australia, and positive airway pressure equipment for other research projects from Philips Respironics, Australia; Air Liquide, Australia; and MayoHealthcare, Australia. Alex van't Hul is an author of one of the studies included in the present review.

New