Effects of cardiorespiratory physiotherapy on cardiovascular fitness in patients with stroke: A systematic review and network meta-analysis

Abstract

Background:

This study aimed to investigate the effectiveness and ranking of different cardiorespiratory-physiotherapy interventions on cardiovascular fitness in patients with stroke.

Methods:

In total, 21 randomized controlled trials published between 2000 and 2024 were retrieved from PubMed, EMBASE, Cochrane Library, Web of Science, and CINAHL and analyzed. Outcome measures were resting heart rate (RHR), rating of perceived exertion (RPE), and peak oxygen uptake (VO2peak). A network meta-analysis was conducted using the R netmeta package (version 4.3.2), with rankings based on probability scores (p-scores) representing the likelihood of each intervention being the most effective.

Results:

RHR rankings were as follows: combined inspiratory and expiratory training (CIET) (p-score: 94.02%), conventional training (CT) (68.16%), robot-assisted training (RAT) (61.27%), aquatic training (AT) (48.90%), ground-based aerobic training (GBAT) (31.61%), combined aerobic and resistance training (CART) (28.01%), and resistance training (RT) (18.03%). However, CIET did not show statistically significant differences in effect compared to other interventions. RPE rankings were as follows: CIET (89.12%), GBAT (72.23%), CT (66.19%), AT (66.01%), CART (30.20%), RAT (19.70%), and inspiratory training (IT) (6.55%). VO2peak rankings were as follows: RT (88.26%), GBAT (73.22%), AT (64.73%), RAT (45.10%), CART (43.34%), sham training (42.88%), IT (39.61%), CIET (28.29%), and CT (24.56%). The evidence quality ranked from very low to moderate.

Conclusion:

CIET ranked highest in reducing RHR and RPE; RT ranked highest in improving VO2peak. Although CIET did not show significant superiority in RHR reduction, its consistent high ranking in RHR and RPE outcomes indicates its potential clinical utility. These findings suggest incorporating targeted respiratory and strength training into stroke-rehabilitation programs to optimize cardiovascular-fitness outcomes.

1. Introduction

Patients with stroke have a high prevalence of cardiovascular disease, diabetes, and obesity, primarily driven by physical inactivity.^[1] Sedentary behavior significantly reduces musculoskeletal health and cardiorespiratory fitness, elevating cardiometabolic risk.^[2–4] Even in healthy participants, 10 days of inactive bed rest has been reported to reduce the cross-sectional area of the quadriceps femoris muscle and peak oxygen uptake (VO2peak), which is a marker of cardiorespiratory fitness, by approximately 3.2% and 6.4%, respectively.^[5] Therefore, personalized rehabilitation programs emphasizing improved limb function and cardiopulmonary capacity are recommended by current clinical guidelines.^[6]

The respiratory and cardiovascular systems are intimately connected by the same circulatory system, through which oxygen and carbon dioxide are transported.^[7] This means that respiration and cardiovascular function work together as cardiopulmonary function, and interventions to improve cardiovascular function are also necessary. The cardiovascular health assessment of such interventions incorporates factors such as resting heart rate (RHR), rating of perceived exertion (RPE), and VO2peak.^[8]

Therefore, to maintain cardiovascular health, it is necessary to manage vascular elasticity and blood pressure through appropriate cardiovascular stress by applying aerobic training and maintaining an appropriate target heart rate. However, the amount of time spent on target heart rate in traditional physical and occupational therapies for patients with stroke is very low.^[9] This indicates that patients with stroke have limited opportunities to improve their cardiovascular function,^[10] despite being at potential risk for vascular impairments. In addition, patients with stroke have been found to experience decreased cardiorespiratory endurance for 3 to 12 months after stroke,^[11] which is 26 to 87% less than that in healthy individuals of the same age and sex.^[12] Therefore, employing cardiac-respiratory physical therapy rather than traditional physical therapy may be more beneficial to enhancing cardiopulmonary function in patients with stroke. However, the comparative effectiveness of different cardiac-respiratory physical therapy interventions remains unclear.

Consequently, network meta-analysis (NMA) has gained prominence as a method for estimating and comparing the relative effects of interventions directly or indirectly, thereby determining rankings. NMA allows the simultaneous comparison of multiple intervention types using direct and indirect evidence, overcoming limitations associated with traditional pairwise meta-analyses.^[13] NMA additionally provides intervention rankings by calculating probability-based metrics, such as the frequentist p-score, which estimates the likelihood of each intervention being most effective for a given outcome.^[13] This approach allows a more comprehensive evaluation of the relative efficacy of different interventions. These benefits have been capitalized on in studies that have highlighted the benefits of core stability training in promoting balance recovery in patients with stroke,^[14] and studies that have ranked different training modalities for upper extremity dysfunction.^[15] Therefore, NMAs can help identify the most effective interventions among various options by considering multiple components and their interactions. They also provide a valuable tool for healthcare professionals to provide optimal patient care. However, although systematic literature reviews of interventions for cardiovascular health in patients with stroke are being conducted,^[16,17] no NMA study has investigated the effectiveness and evidence of cardiac-respiratory physical therapy. This gap indicates that the most effective overall training option remains unclear. Therefore, applying the NMA approach to establish an intervention hierarchy that includes all possible types of cardiorespiratory-physiotherapy interventions is necessary.

Thus, this study aimed to quantitatively analyze and rank various cardiorespiratory-physiotherapy interventions using NMA to determine the most effective strategies for improving cardiovascular fitness in patients with stroke, which could guide clinicians in designing targeted rehabilitation programs to enhance patient outcomes.

2. Methods

2.1. Ethics statement

This study was reported following the PRISMA checklist^[18] and was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under a predefined protocol (CRD42023387282). The protocol submitted during registration is available on the PROSPERO website. Although the original protocol registered in PROSPERO specified a literature search period up to 2023 and did not include the CINAHL database, the search was extended to December 2024 and CINAHL was additionally included to ensure the comprehensive coverage of recent and relevant literature; the original PROSPERO record was not updated. The study was approved by the Institutional Review Board of Nambu University (HBU-IRB-1041478-2023-HR-001).

2.2. Search strategy

The literature search was conducted using international academic databases such as PubMed, EMBASE, Cochrane Library, Web of Science, and CINAHL. To ensure rigorous literature selection, only randomized controlled trials (RCTs) were included, with a time frame limited to 25 years between January 1, 2000, and December 31, 2024. This time frame was established to encompass recent trends and evidence-based interventions in the field. Studies published before 2000 were excluded because they were deemed less likely to reflect interventions currently adopted in clinical practice. On the basis of previous research, we developed a search strategy for stroke and cardiorespiratory physiotherapy.^[19] The detailed search strategy can be found in Table S1A to E (Supplemental Digital Content, https://links.lww.com/MD/Q456). Duplicate removal was performed using Endnote X9 after the initial search and before article selection.

2.3. Eligibility criteria and data extraction

The study participants included patients with stroke, and the intervention types included ground-based aerobic training (GBAT), combined inspiratory and expiratory training (CIET), resistance training (RT), aquatic training (AT), robot-assisted training (RAT), inspiratory training (IT), and combined aerobic and resistance training (CART). Control groups included conventional training (CT) and sham intervention (S). Based on the characteristics of each intervention, all studies were categorized into predefined groups, and this classification was consistently applied throughout the analysis. The classification framework is summarized in Table 1, and the corresponding categorization of each study is presented in Table S2 (Supplemental Digital Content, https://links.lww.com/MD/Q456). The outcome variables were RHR, RPE, and VO2peak.

Table 1.

Abbreviations and descriptions of interventions.

Abbreviations	Description
CIET	Combined inspiratory and expiratory breathing exercises
RAT	Robot-assisted training involving mechanically supported gait
IT	Inspiratory muscle training exercises
GBAT	Ground-based aerobic training such as treadmill walking
RT	Resistance training focusing on strength and power exercises
AT	Aquatic-based aerobic exercises
CART	Combined aerobic and resistance training
CT	Conventional physiotherapy treatments (stretching, etc)
S	Sham intervention

Studies were excluded if the participant did not have a stroke and did not undergo conventional cardiorespiratory physiotherapy. Studies that were not RCTs and those with irrelevant outcome variables, unclear original language, or conducted in a language other than English were also excluded. Two authors independently extracted the following information for the selected studies: author, year of publication, country, sample size, sex, mean age, disease duration, stroke stage, stroke severity, stroke type, and statistical significance.

2.4. Risk of bias and quality of evidence

Quality assessment for selected RCTs was conducted using Cochrane risk of bias in randomized trial 2.0.^[20] Two authors independently assessed bias areas and attempted to reach a consensus, with disagreements resolved in consultation with a third expert. The quality of the evidence for individual outcomes was assessed using the grading of recommendations, assessment, development and evaluations framework.^[21] Publication bias was visualized using comparison-adjusted funnel plots, and further sensitivity analyses were conducted to determine the impact of varying levels of risk of bias on the results.

2.5. Subgroup analysis

Subgroup analyses were conducted to explore potential effect modifiers and to assess whether effects of interventions differed across subgroups. Subgroups were categorized according to participant-level variables, including age (≤60 vs >60 years) and stroke stage (acute, subacute, and chronic). Additionally, to enhance clinical interpretability and minimize heterogeneity, a separate subgroup analysis was conducted using CT as a common comparator across studies.

2.6. Statistical methods

This study used the comprehensive Meta-Analysis program to estimate Hedges g-value for effect size.^[22] The 25%, 50%, and 75% threshold values represented small, medium, and large heterogeneity (I²), respectively.^[23] Considering the potential for clinical and methodological heterogeneity across studies, a random-effects model was applied to accommodate variability in study design, participant characteristics, and intervention types. Cochrane risk of bias in randomized trial 2.0 was visualized using the risk of bias visualization,^[24] and NMA was conducted using the R studio program (version 2023.09.1. build 494, Posit Software, MA) with the netmeta package.^[25] The treatment ranking was performed using a function in netrank,^[26] and effect sizes were presented using league tables and forest plots.

Interpretation of effect sizes was based on established benchmarks: small (0.20), medium (0.50), and large (0.80), following previous research.^[27] The study presented a frequentist p-score, quantifying the probability that an intervention is more effective than others in the network. The p-score represents the relative ranking of interventions, whereas statistical significance should be interpreted in conjunction with effect sizes and confidence intervals presented in the league table. Additionally, local node splitting and Wald testing consistency analyses were conducted to validate our model.

3. Results

3.1. Characteristics of included studies

The initial database search yielded 4322 articles; 162 were screened, and 141 were excluded on the basis of the exclusion criteria. Therefore, 21 articles were included in the analysis (Table S3, Supplemental Digital Content, https://links.lww.com/MD/Q456).

Characteristics of the included studies are presented in Table S2 (Supplemental Digital Content, https://links.lww.com/MD/Q456), and publication years of the included studies ranged from 2008 to 2023. The 21 selected studies were conducted in the following countries: Germany, Jamaica, Taiwan, Canada, Ghana, India, and the United Kingdom each contributed one study; Turkiye, China, and Australia each contributed 2 studies; the United States contributed 3 studies; and South Korea contributed 5 studies. Two studies did not specify sex, and 19 included mixed sexes. The mean age ranged from 54.1 ± 18.9 to 76.29 years (SD not reported) across studies. The stages of stroke were chronic (50%), subacute (23%), and acute (4%). Although stroke severity was not reported in most studies, among the 7 studies that did report severity, 6 involved patients with mild stroke, and only one involved patients with moderate-to-severe stroke.

Among all intervention arms included in the NMA, the distribution was as follows: GBAT (29.55%), AT (6.82%), CART (11.36%), CIET (4.55%), RT (4.55%), RAT (4.55%), and IT (2.27%). Additionally, CT (34.09%) and S (2.27%) were classified as control interventions. The intervention duration ranged from 2 to 120 weeks, with a mean of 13.71 ± 22.69 weeks. The session durations varied, ranging from 15 to 60 minutes, with a mean of 40.11 ± 13.69 minutes. The total number of intervention sessions ranged from 12 to 120, with a mean of 39.39 ± 24.01 sessions. The proportion of studies with no information on intervention intensity was 36.96%, while the proportion of those classified as low, moderate, and high intensity were 13.04%, 4.35%, and 10.87%, respectively. Additionally, the proportions of studies reporting low-to-high intensity, low-to-moderate intensity, moderate-to-high intensity, and progressive intensity increase were 4.35%, 6.52%, 13.04%, and 10.87%, respectively. Furthermore, 2 studies implemented intermittent exercise, which were classified as low-to-moderate intensity and progressive-to-high intensity, respectively.

3.2. NMA

The consistency analysis demonstrated that the model was consistent and suitable for analysis, with P values > .05 for RHR (P = .114), RPE (P = .803), and VO2peak (P = .331) (Table S4, Supplemental Digital Content, https://links.lww.com/MD/Q456). Therefore, models for NMA were developed for RHR (15 treatment arms across 10 studies), RPE (13 treatment arms across 6 studies), and VO2peak (16 treatment arms across 14 studies) (Fig. 1). Each connecting line in the model represents an intervention comparison, with a thicker line indicating a higher number of studies for that intervention comparison.

Figure 1.

Network meta-analysis of eligible comparisons for (A) resting heart rate, (B) rate of perceived exertion, and (C) peak oxygen uptake. AT = aquatic training; CART = combined aerobic and resistance training, CIET = combined inspiratory and expiratory training; CT = conventional training, GBAT = ground-based aerobic training, IT = inspiratory training, RT = resistance training, RAT = robot-assistance training, S = sham intervention.

3.3. NMA of RHR

Overall, 15 paired comparisons between interventions and 553 participants were included. In RHR analysis, a small effect size was observed in CT (standardized mean difference [SMD] = 0.221, 95% confidence interval [95% CI] = 0.016–0.425) compared with GBAT (Table 2). The p-score rankings indicate that CIET (94.02%) ranked highest, followed by CT (68.16%), RAT (61.27%), AT (48.90%), GBAT (31.61%), CART (28.01%), and RT (18.03%) (Fig. 2, Table 3).

Table 2.

Network meta-analysis matrix of the RHR.

GBAT	.	−0.219 (−0.860; 0.421)	−0.063 (−0.438; 0.311)	.	−0.072 (−0.543; 0.398)	0.275 (0.060; 0.491)
0.820 (−0.047; 1.686)	CIET	.	.	.	.	−0.599 (−1.441; 0.243)
−0.219 (−0.860; 0.421)	−1.039 (−2.117; 0.038)	RT	.	.	.	.
0.102 (−0.211; 0.415)	−0.718 (−1.624; 0.188)	0.322 (−0.392; 1.035)	AT	.	.	−0.213 (−0.743; 0.317)
0.221 (−0.336; 0.777)	−0.599 (−1.587; 0.389)	0.440 (−0.409; 1.289)	0.119 (−0.498; 0.735)	RAT	.	0.000 (−0.518; 0.518)
−0.072 (−0.543; 0.398)	−0.892 (−1.878; 0.094)	0.147 (−0.648; 0.942)	−0.174 (−0.739; 0.391)	−0.293 (−1.022; 0.436)	CART	.
0.221 (0.016; 0.425)	−0.599 (−1.441; 0.243)	0.440 (−0.232; 1.113)	0.119 (−0.216; 0.454)	−0.000 (−0.518; 0.518)	0.293 (−0.220; 0.806)	CT

The lower-left triangle presents the findings (Hedges g [g] with 95% confidence intervals [CI]) of the network meta-analysis; the upper-right triangle presents the findings (g with 95% CI) of the pairwise meta-analysis. Each value represents the standardized mean difference (g) between 2 interventions. A positive value indicates that the intervention in the row is more effective than the intervention in the column; a negative value indicates that the intervention in the column is more effective than the one in the row. The values in parentheses indicate the 95% CIs. If the interval does not include 0, the result is considered statistically significant. Bold values indicate statistical significance at P < .05 (95% CI does not include 0).

AT = aquatic training, CART = combined aerobic and resistance training, CIET = combined inspiratory and expiratory training, CT = conventional training, GBAT = ground-based aerobic training, RAT = robot-assistance training, RHR = resting heart rate, RT = resistance training.

Figure 2.

Forest plot of (A) resting heart rate, (B) rate of perceived exertion, and (C) peak oxygen uptake. AT = aquatic training, CART = combined aerobic and resistance training, CI = confidence interval, CIET = combined inspiratory and expiratory training, CT = conventional training, GBAT = ground-based aerobic training, IT = inspiratory training, RAT = robot-assistance training, RPE = rate of perceived exertion, RT = resistance training, S = sham intervention, SMD = standardized mean difference, VO2peak = peak oxygen uptake.

Table 3.

Probability ranking results of different therapy on RHR, RPE, and VO2peak.

Intervention measure	RHR (Rank)	RPE (Rank)	VO2peak (Rank)
AT	48.90 (4)	66.01 (4)	64.73 (3)
CART	28.01 (6)	30.20 (5)	43.34 (5)
CIET	94.02 (1)	89.12 (1)	28.29 (8)
CT	68.16 (2)	66.19 (3)	24.56 (9)
GBAT	31.61 (5)	72.23 (2)	73.22 (2)
IT	–	6.55 (7)	39.61 (7)
RT	18.03 (7)	–	88.26 (1)
RAT	61.27 (3)	19.70 (6)	45.10 (4)
S	–	–	42.88 (6)

This table presents the probability ranking results (p-scores) of each intervention for 3 outcomes: resting heart rate (RHR), rating of perceived exertion (RPE), and peak oxygen uptake (VO2peak). Higher p-scores indicate a higher probability of being the most effective intervention for the respective outcome. The rank in parentheses indicates the relative position among compared interventions, with 1 representing the highest rank.

AT = aquatic training, CART = combined aerobic and resistance training, CIET = combined inspiratory and expiratory training, CT = conventional training, GBAT = ground-based aerobic training, IT = inspiratory training, RAT = robot-assistance training, RHR = resting heart rate, RPE = rate of perceived exertion, RT = resistance training, S = sham intervention, VO2peak = peak oxygen uptake.

3.4. NMA of RPE

Overall, 13 paired comparisons between interventions and 173 participants were included. In RPE analysis, CIET (SMD = −1.793, 95% CI = −2.511 to −1.075), GBAT (SMD = −1.533, 95% CI = −2.370 to −0.697), CT (SMD = 1.483, 95% CI = 0.755–2.192), and AT (SMD = −1.473, 95% CI = −2.333 to −0.612) showed statistically significant effects compared to IT. Additionally, CIET showed statistically significant effects compared to RAT (SMD = −1.393, 95% CI = −2.755 to −0.031) and CART (SMD = −1.033, 95% CI = −2.065 to −0.001) (Table 4). The ranking results of the p-score indicated that CIET (89.12%) had the highest ranking, followed by GBAT (72.23%), CT (66.19%), AT (66.01%), CART (30.20%), RAT (19.70%), and IT (6.55%) (Fig. 2, Table 3).

Table 4.

Network meta‐analysis matrix of the RPE.

GBAT	.	0.083 (−0.532; 0.698)	−1.133 (−2.290; 0.024)	.	.	−0.151 (−0.669; 0.367)
0.260 (−0.458; 0.978)	CIET	.	.	−1.818 (−2.766; −0.871)	.	−0.298 (−0.923; 0.326)
−0.060 (−0.549; 0.428)	−0.320 (−1.066; 0.426)	AT	.	.	.	0.154 (−0.461; 0.769)
−1.133 (−2.290; 0.024)	−1.393 (−2.755; −0.031)	−1.073 (−2.329; 0.183)	RAT	.	.	.
−1.533 (−2.370; −0.697)	−1.793 (−2.511; −1.075)	−1.473 (−2.333; −0.612)	−0.400 (−1.828; 1.028)	IT	.	1.460 (0.554; 2.367)
−0.773 (−1.745; 0.199)	−1.033 (−2.065; −0.001)	−0.712 (−1.705; 0.280)	0.360 (−1.151; 1.871)	0.760 (−0.357; 1.878)	CART	0.723 (−0.141; 1.587)
−0.050 (−0.495; 0.395)	−0.310 (−0.873; 0.254)	0.011 (−0.478; 0.499)	1.083 (−0.156; 2.323)	1.484 (0.775; 2.192)	0.723 (−0.141; 1.587)	CT

The lower-left triangle presents the findings (Hedges g [g] with 95% confidence intervals [CI]) of the network meta-analysis; the upper-right triangle presents the findings (g with 95% CI) of the pairwise meta-analysis. Each value represents the standardized mean difference (g) between 2 interventions. A positive value indicates that the intervention in the row is more effective than the intervention in the column; a negative value indicates that the intervention in the column is more effective than the one in the row. The values in parentheses indicate the 95% CIs. If the interval does not include 0, the result is considered statistically significant. Bold values indicate statistical significance at P < .05 (95% CI does not include 0).

AT = aquatic training, CIET = combined inspiratory and expiratory training, CT = conventional training, GBAT = ground-based aerobic training, IT = inspiratory training, RAT = robot-assistance training, RPE = rating of perceived exertion.

3.5. NMA of VO2peak

Overall, 16 paired comparisons between interventions and 772 participants were included. The VO2peak analysis indicated that RT (SMD = 2.305, 95% CI = 0.401–4.208) and GBAT (SMD = 1.489, 95% CI = 0.378–2.600) showed statistically significant effects compared to CT (Table 5). The ranking results of the p-score indicated that RT (88.26%) had the highest ranking, followed by GBAT (73.22%), AT (64.73%), RAT (45.10%), CART (43.34%), S (42.88%), IT (39.61%), CIET (28.29%), and CT (24.56%) (Fig. 2, Table 3).

Table 5.

Network meta‐analysis matrix of the VO2peak.

GBAT	.	−1.961 (−4.539; 0.617)	0.174 (−1.647; 1.995)	.	.	−0.043 (−2.544; 2.457)	.	2.015 (0.754; 3.275)
1.529 (−0.834; 3.891)	CIET	.	.	.	−0.387 (−2.942; 2.168)	.	.	−0.046 (−2.599; 2.507)
−0.816 (−2.720; 1.089)	−2.344 (−5.168; 0.479)	RT	.	.	.	.	.	1.163 (−1.410; 3.736)
0.174 (−1.647; 1.995)	−1.355 (−4.338; 1.628)	0.990 (−1.645; 3.624)	AT	.	.	.	.	.
0.926 (−1.821; 3.673)	−0.603 (−3.867; 2.662)	1.742 (−1.410; 4.894)	0.752 (−2.544; 4.048)	RAT	.	.	.	0.563 (−1.949; 3.075)
1.135 (−1.228; 3.499)	−0.393 (−2.479; 1.692)	1.951 (−0.873; 4.775)	0.962 (−2.022; 3.945)	0.210 (−3.055; 3.474)	IT	.	.	0.360 (−2.195; 2.914)
0.949 (−0.689; 2.587)	−0.580 (−3.157; 1.998)	1.765 (−0.580; 4.109)	0.775 (−1.674; 3.224)	0.023 (−2.911; 2.957)	−0.186 (−2.765; 2.392)	CART	0.077 (−2.498; 2.651)	0.019 (−1.791; 1.830)
1.026 (−2.026; 4.077)	−0.503 (−4.146; 3.140)	1.842 (−1.641; 5.324)	0.852 (−2.702; 4.405)	0.100 (−3.804; 4.003)	−0.110 (−3.753; 3.534)	0.077 (−2.498; 2.651)	S	.
1.489 (0.378; 2.600)	−0.040 (−2.125; 2.045)	2.305 (0.401; 4.208)	1.315 (−0.818; 3.448)	0.563 (−1.949; 3.075)	0.353 (−1.732; 2.439)	0.540 (−0.976; 2.055)	0.463 (−2.525; 3.451)	CT

The lower-left triangle presents the findings (Hedges g [g] with 95% confidence intervals [CI]) of the network meta-analysis; the upper-right triangle presents the findings (g with 95% CI) of the pairwise meta-analysis. Each value represents the standardized mean difference (g) between 2 interventions. A positive value indicates that the intervention in the column is more effective than the intervention in the row; a negative value indicates that the intervention in the row is more effective than the one in the column. The values in parentheses indicate the 95% CIs. If the interval does not include 0, the result is considered statistically significant. Bold values indicate statistical significance at P < .05 (95% CI does not include 0).

AT = aquatic training, CART = combined aerobic and resistance training, CIET = combined inspiratory and expiratory training, CT = conventional training, GBAT = ground-based aerobic training, IT = inspiratory training, RAT = robot-assistance training, RT = resistance training, S = sham intervention, VO2peak = peak oxygen uptake.

3.6. Subgroup analysis

Subgroup analysis revealed a significant difference in RPE according to stroke stage (P < .05). RPE significantly decreased in the acute stroke group, whereas it significantly increased in the chronic stroke group, although the effect size was very small.

Subgroup analysis of VO2peak revealed significant differences according to age and stroke stage (P < .05). Furthermore, significant improvements in VO2peak were observed in patients aged ≤ 60 years and those in the chronic phase of stroke. Notably, in an additional subgroup analysis restricted to comparisons using CT as the control group, significant improvements were observed in the chronic stroke subgroup (Table S5A and B, Supplemental Digital Content, https://links.lww.com/MD/Q456).

3.7. Risk of bias and quality of evidence

One study exhibited a low risk of bias, ten raised some concerns, and ten had a high risk of bias (Fig. S6a and b, Supplemental Digital Content, https://links.lww.com/MD/Q456). Sensitivity analyses, excluding studies with a high risk of bias, revealed differences in heterogeneity compared with the main analysis. However, relative rankings and effects mostly remained consistent without including interventions from the excluded studies, demonstrating good consistency (Table S7A–C, Supplemental Digital Content, https://links.lww.com/MD/Q456; Fig. S8A–C, Supplemental Digital Content, https://links.lww.com/MD/Q456; and Table S9A–C, Supplemental Digital Content, https://links.lww.com/MD/Q456). Notably, the deflation of various intervening variables led to differences in heterogeneity for RHR and VO2peak, excluding 6 (40%) and 8 (50%) cases, respectively, during paired comparisons. Additionally, RHR showed no significant difference between interventions, whereas VO2peak demonstrated a moderate effect of RAT over CT. This effect was originally inconclusive because RT and AT were excluded as interventions. However, the p-score ranking for each intervention was similar to that in the main analysis. Furthermore, the studies were mostly symmetrical in the comparison-adjusted funnel plot, showing no bias (Fig. S10, Supplemental Digital Content, https://links.lww.com/MD/Q456). Therefore, the overall quality according to grading of recommendations, assessment, development and evaluations was very low or moderate (Table S11, Supplemental Digital Content, https://links.lww.com/MD/Q456).

4. Discussion

4.1. Summary of main finding

To our knowledge, this is one of the first analyses using NMA to examine cardiorespiratory-physiotherapy interventions aimed at improving RHR, RPE, and VO2peak in patients with stroke. CIET ranked highest for reducing RHR, but it did not show significant effects compared to other interventions. CT showed only a minimal effect on RHR reduction compared to GBAT. Additionally, CIET ranked highest and showed effectiveness in improving RPE; likewise, RT was ranked highest and was effective in improving VO₂peak.

4.2. Effectiveness of interventions on RHR

Based on the study findings, CIET had the highest ranking for improving RHR, followed by CT, RAT, AT, GBAT, CART, and RT in descending order. The high ranking of CIET may be explained by an indirect relationship between respiration and heart rate reduction. This phenomenon suggests that, since the heart rate shares neural pathways with respiration, the voluntary regulation of slow breathing may influence the activity of the parasympathetic nervous system.^[28] Supporting this interpretation, a previous meta-analysis investigating the effects of respiratory training in individuals with cardiovascular diseases – a major contributing factor to stroke onset – found that respiratory training reduced RHR.^[29] However, CIET did not show statistically significant effects compared to other interventions. Although CIET did not demonstrate statistically significant superiority over other interventions in direct or indirect comparisons, its top ranking in the NMA and its plausible physiological rationale suggest that it has potential for reducing RHR in patients with stroke. Nonetheless, given the limited number of studies and wide confidence intervals, further research is warranted to confirm its clinical efficacy.

In contrast, CT demonstrated a statistically significant effect compared to GBAT, although the magnitude of this effect was minimal. The modest yet significant effectiveness of CT compared with GBAT observed herein might be attributable to CT components such as direct stretching, muscle relaxation, and joint range-of-motion enhancement training. However, mechanistic evidence to support this hypothesis remains limited; further research is required. These activities, associated with heart rate, indirectly reduce arterial stiffness, decreasing RHR.^[30] This interpretation supports the findings of a meta-analysis assessing cardiac autonomic function in multiple populations. The study discovered stretching as a useful therapeutic intervention for heart rate variability^[31] and observed that it reduced arterial stiffness and significantly lowered RHR when applied to older adults.^[32] However, its physiological regulatory mechanisms on heart rate remain unclear. Furthermore, researchers hypothesized that baroreflex sensitization, vasodilation, and increased nitric oxide bioavailability contribute to regulating heart rate and blood pressure.^[31] Taken together, minimal differences in intervention effects were observed in this study, and further in-depth evaluation is needed to determine whether the actual effects of the interventions can lead to clinically meaningful changes for patients. Furthermore, the observed effect between interventions is small, and the subdomain of CT comprises various conventional treatments. Therefore, it is unclear if the effect is specific to stretching. Future research is required to directly compare the RHR between patients with stroke who have undergone CT and GBAT.

4.3. Impact of RT and aerobic training on RHR

In contrast, compared with other intervention types, RT ranked lowest, with RT failing to reduce arterial stiffness significantly compared with aerobic training.^[33] Additionally, a meta-analysis of multiple interventions in a population-based study found that yoga, which includes stretching and breathing exercises, was highly effective in reducing RHR, whereas RT was ineffective.^[34] Heart rate variability-based training analysis studies, including endurance and RT, have also shown no benefit for RHR.^[35] Furthermore, a meta-analysis study examining the effects of training on cardiovascular and autonomic parameters in patients with stroke found that CART, including aerobic training, effectively improved RHR, which was inconsistent with our findings.^[17] However, other studies have shown that aerobic training is more effective on the cardiovascular system than other training modalities. Dynamic RT was ineffective in reducing arterial stiffness, whereas combined training and isometric RT were effective.^[36] Therefore, further research on the potential effects of aerobic training, RT, and CART on RHR is needed. Additionally, few systematic reviews of RHR have been conducted in patients with stroke, which complicates intuitive comparisons and generalizes this study’s results. However, studies have indicated that an RHR of ≥77 beats per min*** is significantly associated with increased vascular mortality compared with an RHR of ≤66 beats per min***.^[37] Additionally, lower heart rates are correlated with better functional outcomes in patients with stroke.^[37] Therefore, this study, which presents a comprehensive analysis of various intervention training methods to lower RHR in patients with stroke, could provide valuable insights into intervention selection during cardiorespiratory physiotherapy for patients with stroke.

4.4. Analysis of exercise tolerance and fatigue (RPE)

In addition, stroke can weaken cardiovascular health and exercise capacity, resulting in symptoms such as shortness of breath and fatigue.^[2,38] Studies measuring RPE, an indicator of physical effort and fatigue, have reported that interventions that reduce fatigue can improve exercise tolerance and positively impact physical activity participation.^[39] CIET was identified as the most effective intervention for improving the RPE, with GBAT, CT, and AT emerging as viable alternatives to CIET. Conversely, CART, RAT, and IT had no effect. This suggests that other intervention types, besides IT and RAT, effectively improve exercise tolerance physiologically and subjectively by reducing RPE. Previous studies have shown that RAT failed to reduce RPE owing to the high level of assistance from the robot, and insufficient exercise intensity applied to patients with stroke did not adequately stress the cardiopulmonary system.^[40] Additionally, CIET may be more effective than CART in reducing RPE because it exerts more direct effects on cardiopulmonary function, including enhanced respiratory muscle performance.^[41] However, the effects of the intervention may vary depending on stroke severity and intervention intensity; therefore, future studies should carefully consider these factors.

4.5. CIET for reducing RHR and RPE

Another finding from RPE analysis is that IT did not demonstrate effectiveness, highlighting the importance of expiratory training in patients with stroke. Previous studies have suggested that combining inspiratory and expiratory muscle training in patients with stroke is more effective for cardiopulmonary function.^[41] Patients with stroke experience a 34% reduction in heart rate compared with those without stroke, with a 41% decrease in minute ventilation capacity and a continued slower pace of ventilation and carbon dioxide production even after maximal exercise.^[42] However, the slow carbon dioxide production in the body may result in inadequate carbon dioxide elimination during respiration, leading to oxygen deprivation within muscles during energy generation processes and lactic acid production.^[43] Therefore, appropriate respiratory training can enhance oxygenation in the blood and increase muscle pH levels, thereby reducing muscle fatigue.

The favorable ranking of CIET in reducing RHR and RPE suggests that direct inspiratory and expiratory training supports the body’s cardiopulmonary adaptation. This efficient acquisition of respiratory exchange methods is believed to improve cardiopulmonary function without incorporating aerobic training. Furthermore, given the lack of clear guidelines or methods for conventional treatment in patients with stroke, emphasizing the importance of respiratory education for cardiopulmonary function is essential over simply applying conventional or aerobic training. Therefore, effective respiratory pattern recognition and enhancement must be prioritized, as this foundational training is believed to enhance functional movement, daily activities, and efficient exercise performance in patients with stroke.

4.6. VO2peak enhancement and intervention priorities

Sensitivity analysis of VO2peak showed a significant effect of RAT compared with CT in the absence of RT and AT. This may be because RAT intensity maintained a cardiovascular burden of 50 to 55% of VO2peak in the RAT studies included in our analysis, suggesting that moderate aerobic activity enhanced VO2peak.^[44] This means that the effectiveness of RAT regarding cardiopulmonary stress varies across studies and requires further confirmation through in-depth research. These results also suggest that RAT with moderate aerobic activity intensity may be the next best alternative to CT when RT or GBAT is not feasible regarding VO2peak.

Additionally, RT and GBAT demonstrated significant improvements in VO2peak compared with CT, with the rankings as follows: RT, GBAT, AT, RAT, CART, S, IT, CIET, and CT. Previous studies have shown that VO2 kinetics in patients with stroke were 29% slower and were associated with VO2peak and maximum cardiac output.^[42] Typically, slow VO2 kinetics imply issues with oxygen uptake rate, suggesting problems with cardiovascular or muscle functions. Essentially, slower oxygen delivery after exercise onset may lead to secondary effects such as muscle fatigue, reduced endurance, and overall negative impacts on exercise performance. Previous studies on VO2peak in patients with stroke have shown significant improvements after high-intensity treadmill training.^[8] Subgroup analyses have suggested high-quality evidence for high-intensity treadmill training (≥70% heart rate reserve/VO2peak) in achieving these improvements.^[8] Therefore, interventions, including aerobic training, yield a 10 to 15% improvement in VO2peak.^[45]

4.7. VO2peak improvement by RT

However, the results of this study indicate a higher prioritization of RT over GBAT. VO2peak is determined by physiological factors such as heart rate, stroke volume, arterio-venous oxygen difference, and pulmonary ventilation and diffusion capacity. Heart rate and stroke volume reflect cardiovascular function, whereas arterio-venous oxygen difference indicates the amount of oxygen extracted from the blood during muscle activity. Therefore, cardiovascular, muscular, and pulmonary functions all contribute to determining VO2peak.^[46] This process reflects the overall capacity to transport atmospheric oxygen to the mitochondria for performing work.^[46] Muscle atrophy, fat accumulation, and reduced capillary density in paralyzed muscles of patients with stroke may contribute to decreased levels of VO2peak, potentially impairing functional mobility and independence.^[47] A meta-analysis study found that RT applied to patients with stroke improves muscle strength, power, endurance, and other related factors compared with high-intensity aerobic exercise. Even low-intensity RT showed benefits in muscle strengthening.^[48]

In cardiovascular and cardiac rehabilitation, the importance of RT is emphasized alongside aerobic exercise.^[49] RT can be applied as an alternative to aerobic exercise when individuals are unable or unwilling to participate in aerobic activities. Previous research supports that RT can improve muscle and cardiovascular health in individuals with low initial fitness levels.^[50] Furthermore, it contributes to improvements in capillary-to-fiber ratio, mitochondrial enzyme activity, and blood flow, leading to enhancements similar to VO2peak in VO2max.^[50] RT ranked high in our study; however, there was no significant difference in effectiveness between RT and GBAT. Therefore, understanding the patient’s condition and symptom severity is crucial for appreciating the benefits of RT and appropriately selecting between RT and aerobic training for application in patients with stroke. However, research on cardiovascular function in patients with stroke remains limited, and future studies should comprehensively investigate and evaluate the effects based on the benefits of cardiovascular function in patients with stroke.

4.8. Subgroup analysis: exploratory interpretation

In the subgroup analysis, RPE significantly decreased in the acute stroke group and significantly increased in the chronic stroke group. Because the acute stroke group included only one study and the effect size in the chronic stroke group was minimal, these findings should be interpreted at an exploratory level. Furthermore, significant improvements in VO2peak were observed in patients aged ≤60 years and in the chronic stroke subgroup. In an additional subgroup analysis limited to CT as the control group, VO2peak also significantly increased in the chronic stroke subgroup. These findings suggest that cardiorespiratory physiotherapy is relatively more effective in improving VO2peak among patients aged ≤60 years and those with chronic stroke. Patients in the chronic phase may have achieved functional stability, resulting in relatively higher exercise performance, while those aged ≤60 years may have exhibited greater physiological responsiveness, leading to more pronounced intervention effects. Nevertheless, given the small number of included studies and the high heterogeneity observed in some analyses, these results should be interpreted with caution.

4.9. Limitations and future directions

This study has some limitations. First, few NMA studies can be compared to our results, making it difficult to generalize our findings. Second, future studies should consider and report key patient characteristics such as stroke severity, disability, and details of treatment intensity, frequency, and duration. In the present study, we attempted to reflect these variables as much as possible in the general characteristics; however, owing to inconsistencies in reporting methods and insufficient information in many studies, further subgroup analyses were limited. Moreover, there is a lack of RCTs targeting patients with stroke that also examine other cardiovascular physiological indicators such as heart rate variability and blood pressure, which limited the ability to conduct an integrated analysis including these variables. Third, the small number of included studies made it difficult to conduct NMAs of sub-domains, and the limited number of articles prevented analyses of other treatment modalities. Fourth, the literature analyzed in this study was restricted to studies from limited databases, so further research should be conducted to supplement the literature, including gray literature. Fifth, some included studies applied 2 or more interventions simultaneously. In such cases, the difference between comparison groups that shared a common intervention was interpreted as the effect of the additional component. However, this approach cannot completely rule out the possibility of interaction between interventions, limiting the ability to clearly isolate the unique effect of each component. Therefore, the results should be interpreted with caution. Sixth, many studies were assessed to be highly biased, and the overall quality of evidence from the included studies was rated very low to moderate. Further analyses are needed to synthesize more studies with low heterogeneity to draw clearer conclusions. To reduce this risk, we evaluated the potential for bias in each study and conducted sensitivity analyses to remove those with a high risk of bias. Despite efforts to minimize bias, it cannot be entirely eliminated, and thus, caution is required when interpreting results.

5. Conclusion

CIET had the highest ranking in RHR and RPE, and RT had the highest ranking in VO2peak. CIET did not show a statistically significant effect in reducing RHR compared to other interventions, whereas CT significantly reduced RHR compared to GBAT. However, the effect size was very small, and thus, caution is needed when interpreting its clinical significance. Given that CIET was ranked relatively higher than other interventions, it may be a promising approach for improving RHR if its effectiveness is confirmed in future studies.

In RPE, other intervention types were more effective than IT, which only included inhalation training. Specifically, CIET with exhalation training was more effective than CART, RAT, and IT. This suggests that breathing exercises that include exhalation may be relatively more effective in improving cardiovascular health in patients with stroke.

Additionally, RT was identified as the most effective intervention for improving VO2peak, with GBAT being a viable alternative. This suggests that, to enhance cardiopulmonary function, it is necessary to consider the volume of aerobic exercise and the roles of RT and breathing patterns. In addition, relatively greater improvements in VO2peak were observed in patients under 60 years of age or in the chronic stage of stroke, suggesting that cardiorespiratory-physiotherapy interventions are more effective in specific populations. The overall quality of evidence for these findings was very low to moderate. These findings highlight the clinical importance of cardiorespiratory physiotherapy for patients with stroke and may support evidence-based clinical decision-making.

Acknowledgments

This work was supported by the National Research Foundation of Korea grant that was funded by the Korean government (Ministry of Science and ICT) [grant number 2022R1F1A1067604]. This study was supported by research funds from Nambu University, 2025. The authors declare no potential conflicts of interest concerning this article’s research, authorship, and publication.

Author contributions

Conceptualization: So Hyun Kim, Sung Hyoun Cho.

Data curation: So Hyun Kim, Sung Hyoun Cho.

Formal analysis: So Hyun Kim, Sung Hyoun Cho.

Funding acquisition: Sung Hyoun Cho.

Investigation: So Hyun Kim, Sung Hyoun Cho.

Methodology: So Hyun Kim, Sung Hyoun Cho.

Project administration: Sung Hyoun Cho.

Resources: Sung Hyoun Cho.

Software: So Hyun Kim, Sung Hyoun Cho.

Supervision: Sung Hyoun Cho.

Validation: So Hyun Kim, Sung Hyoun Cho.

Visualization: So Hyun Kim.

Writing – original draft: So Hyun Kim.

Writing – review & editing: So Hyun Kim, Sung Hyoun Cho.