Study selection and characteristics of included studies
Our systematic search identified 5739 potentially relevant studies, including 22 studies identified from bibliographies of reports of relevant systematic reviews and meta-analysis (figure 1). After full-text screening, we included 139 RCTs, with 18 670 participants conducted between 1996 and 2021. The studies were conducted in 28 countries spread across Europe (eg, UK, Germany, Netherlands, Switzerland), North and South America (eg, USA, Canada, Brazil, Uruguay), Africa (Nigeria), Asia (eg, China, Taiwan) and Australia.
The PRISMA 2020 flow diagram of study selection.19 6MWD, 6-min walk distance; HF, heart failure; HRQoL, health-related quality of life; ISWD, incremental shuttle walk distance; KCCQ, Kansas City Cardiomyopathy Questionnaire; MLHFQ, Minnesota Living with Heart Failure Questionnaire; NMA, network meta-analysis; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses; SF-36 MCS, Short Form Survey 36 Mental Component Score; SF-36 PCS, SF 36 Physical Component Score; VO2peak, peak oxygen uptake.
All four ExCR delivery modes were represented. Of the 139 trials, 80 were centred-based vs UC, and 35 were HB vs UC, followed by 9 hybrid vs UC, 7 TE vs UC, 4 centre vs HB, 3 CB vs TE and 1 hybrid vs HB. Detailed information about trial treatments is available in online supplemental file 1, and comparisons are summarised in the network plot figure. A small number of studies reported exercise intensity (n=15) and exercise training compliance (n=18). Aerobic (n=84) and aerobic +resistance (n=27) were the most common training modes, followed by flexibility (n=9), resistance (n=8), aerobic +resistance + flexibility (n=7), aerobic +flexibility (n=2) and resistance +flexibility (n=2). Characteristics of included RCTs28–166 are summarised in online supplemental file 1.
The median sample size was 50 participants (range: 10–2331), median participant age was 61.1 years (range: 44–81) and 71.4% of the pooled sample population were male. The median exercise programme duration was 12 weeks (IQR: 12–24 weeks). One study delivered a 10-year programme; however, this comprised three supervised sessions per week for 2 months followed by only two supervised sessions per year.90 The median length of study follow-up was 16 weeks (IQR: 12–26 weeks).
Included studies assessed exercise capacity via peak oxygen uptake (VO2peak, mL/kg/min) or proxy measures including 6 min walk distance (6MWD, m) and incremental shuttle walk distance (ISWD, m). HRQoL was assessed with the Minnesota Living with Heart Failure Questionnaire (MLHFQ), Kansas City Cardiomyopathy Questionnaire (KCCQ) and Short Survey Form 36 (SF-36) mental and physical components (figure 1). HF-related hospitalisations and HF-related mortalities were reported in absolute numbers.
Of the 139 RCTs, 12 trials reported adverse events that occurred during or immediately after exercise training.40 43 53 56 74 90 104 110 132 145 156 158 The reported adverse events were: worsening of HF, hospitalisation due to myocardial infarction, acute coronary syndrome, musculoskeletal injury, shortness of breath, hypoglycaemia, palpitations, angina, arrhythmia, presyncope or syncope, occlusion of peripheral bypass, ectopic heartbeats, hypotension and back pain. No exercise-induced fatal events were reported.
Risk of bias assessment
Sixty-nine (49.6%) of the 139 RCTs had high overall risk of bias (figure 2); 33 (23.7%) studies had high risk of bias due to the randomization process, 25 (18.8%) due to missing outcome data, 27 (19.4%) due to measurement of the outcome, and one due to selection of the reported result. Two studies31 120 had a high risk of bias due to deviations from the intended interventions, where 23 participants crossed over from control to intervention. Of the 139 RCTs, 66 (47.5%) had some concerns about their overall risk of bias: 122 (87.8%) RCTs had some concerns due to the selection of the reported result—studies did not report if they followed a prespecified analysis plan; 72 (51.8%) due to bias in the measurement of the outcome—studies did not report if outcome assessors were blind; and 60 (43.2%) due to the randomization process—studies did not clearly describe allocation concealment. One hundred and thirty-five (97.1%) RCTs had a low risk of bias due to deviations from intended interventions, and 101 (72.7%) due to missing outcome data (figure 2).
The Cochrane risk of bias graph for the included studies.
Exercise capacity
Six-min walk distance
Among 66 comparisons of effects on 6MWD, 32 were between CB ExCR and UC followed by 21 HB ExCR and UC (figure 3). Only hybrid, CB and HB ExCR were associated with increases in 6MWD relative to UC (MD (95% CrI)=84.78 (31.64 to 138.32) m: moderate evidence, MD=50.35 (30.15 to 70.56) m: high evidence and MD=36.77 (12.47 to 61.29) m: moderate evidence, respectively). There were no statistically significant differences between delivery modes (online supplemental files 2 and 3).
SUCRA showed that hybrid ExCR had the highest probability of being ranked first (94.6%), followed by CB ExCR (68.8%) and HB ExCR (46.9%) (online supplemental file 4). There was evidence of network heterogeneity (I2=97.67%) but not incoherence (p>0.1).
Incremental shuttle walk distance
Among six comparisons of effects on ISWD, five were between HB ExCR and UC (figure 3). Neither home or CB programmes improved ISWD compared with UC (HB MD=23.28 (−16.62 to 60.40) m; moderate evidence, and CB MD=9.05 (−70.20 to 88.29) m; low evidence). There was no statistically significant difference between the two ExCR modes (online supplemental files 2 and 3).
Although it did not show statistical significance, SUCRA showed that HB ExCR had the highest probability of being ranked first (76.8%), followed by CB ExCR (48.0%) (online supplemental file 4). There was evidence of network heterogeneity (I2=99.05%).
Peak oxygen uptake
Among 90 comparisons of effects on VO2peak, 59 were between CB ExCR and UC followed by 15 HB ExCR and UC (figure 3). Only CB, HB and TE ExCR were associated with increases in peak oxygen uptake compared with UC (MD=3.10 (2.55 to 3.65) mL/kg/min; high evidence, MD=2.69 (1.67 to 3.70) mL/kg/min; moderate evidence and MD=1.76 (0.26, 3.26) mL/kg/min: low evidence, respectively). There were no statistically significant differences between delivery modes (online supplemental files 2 and 3).
SUCRA showed that CB ExCR had the highest probability of being ranked first (90.5%), followed by HB ExCR (71.8%) and hybrid ExCR (44.1%) (online supplemental file 4). There was evidence of network heterogeneity (I2=94.59%) but not incoherence (p>0.1).
Health-related quality of life
MLHFQ score
Among 52 comparisons of effects on MLHFQ, 29 were between CB ExCR and UC followed by 18 HB ExCR and UC (figure 3). Only centre and HB ExCR showed significant decreases in MLHFQ score compared with UC (MD=−10.38 (−14.15 to –6.46); high evidence, and MD=−8.80 (−13.62 to –4.07); low evidence, respectively). There were no statistically significant differences between delivery modes (online supplemental files 2 and 3).
SUCRA showed that TE ExCR had the highest probability of being ranked first (70.6%), followed by CB ExCR (66.6%) and hybrid ExCR (56.6%) (online supplemental file 4). There was evidence of network heterogeneity (I2=98.05%) but not incoherence (p>0.1).
SF-36 mental component summary score
Among eight comparisons of effects on the SF-36 mental component summary score, six were between CB ExCR and UC (figure 3). Only CB delivery was associated with a statistically significant increase relative to UC (MD (95% CrI)=3.64 (0.30 to 6.14); moderate evidence). There were no statistically significant differences between the CB, HB or TE delivery modes (online supplemental files 2 and 3).
SUCRA showed that CB ExCR had the highest probability of being ranked first (74.7%), followed by TE (70.4%) and HB ExCR (40.2%) (online supplemental file 4). There was evidence of network heterogeneity (I2=83.4%).
SF-36 physical component summary score
Among nine comparisons of effects on SF-36 physical component summary score, six were between CB ExCR and UC (figure 3). No delivery mode improved the SF-36 physical component summary score compared with UC (CB MD=3.24 (−0.37 to 7.35); moderate evidence, HB MD=3.28 (−3.63 to 10.74); high evidence) and TE MD=3.59 (−5.38 to 13.21); moderate evidence). There were no statistically significant differences between the three modes (online supplemental files 2 and 3).
Although it did not show statistical significance, SUCRA showed that CB ExCR had the highest probability of being ranked first (63.8%), followed by TE (62.6%) and HB ExCR (60.4%) (online supplemental file 4). There was evidence of network heterogeneity (I2=98.18%) but not incoherence (p>0.1).
KCCQ score
Among nine comparisons of effects on KCCQ, six were between CB ExCR and UC (figure 3). Only HB ExCR was associated with a significant increase in KCCQ relative to UC (MD=20.61 (4.61 to 36.47); moderate evidence). There were no statistically significant differences between delivery modes (online supplemental files 2 and 3).
SUCRA showed that HB ExCR had the highest probability of being ranked first (95.6%), followed by TE ExCR (56.9%) and CB ExCR (41.5%) (online supplemental file 4). There was evidence of network heterogeneity (I2=98.77%).
HF-related hospitalisation
Among 15 comparisons of effects on HF-related hospitalisation, nine were between CB ExCR and UC, and included relatively short observation periods (4–60 weeks) except for one study with a 520-week treatment period90 (figure 3). CB ExCR was the only delivery mode associated with lower HF-related hospitalisation risk (OR=0.41 (95% CrI 0.17 to 0.76): high evidence), and HF-related hospitalisation risk did not differ between ExCR delivery modes (online supplemental files 2 and 3).
SUCRA showed that hybrid ExCR had the highest probability of being ranked first (75.2%), followed by HB ExCR (71.7%) and CB ExCR (66.2%) (online supplemental file 4). There was evidence of network heterogeneity (I2=87.81%) but not incoherence (p>0.1).
HF-related mortality
Only seven comparisons assessed effects on HF-related mortality; four were between CB ExCR and UC, and included relatively short observation periods (12–60 weeks) except the one study with a 520-week treatment period90 (figure 3). Similar to HF-related hospitalisation, CB ExCR was the only delivery mode associated with lower HF-related mortality risk (OR=0.42 (95% CrI 0.16 to 0.90): moderate evidence), and effects did not differ between ExCR delivery modes (online supplemental files 2 and 3).
SUCRA showed that hybrid ExCR had the highest probability of being ranked first (88.9%), followed by CB ExCR (56.9%) and HB ExCR (45.0%; online supplemental file 4). There was neither network heterogeneity (I2=0) nor incoherence (p>0.1).