Exercise-based cardiac rehabilitation for coronary heart disease: a meta-analysis

1See the editorial comment for this article ‘Evidence is indisputable that cardiac rehabilitation provides health benefits and event reduction: time for policy action’, by S. L. Grace, https://doi.org/10.1093/eurheartj/ehac690.

2Coronary heart disease (CHD) is the most common cause of death globally., With increasing numbers of people living longer with CHD, accessible and effective health services for the management of CHD are crucial. Exercise-based cardiac rehabilitation (CR) is recognized as a key component of comprehensive CHD management and is a Class I Grade A recommendation in international guidelines.,

3Although meta-analyses of randomized controlled trials (RCTs) have shown the beneficial effect of CR in patients with CHD, this evidence base has been questioned on the grounds of: (i) uncertainty in the impact on mortality; (ii) lack of data on health-related quality of life (HRQoL); (iii) inclusion of RCTs limited to low-risk patients and conducted in high-income country settings, and (iv) lack of trials conducted during the era of modern CHD therapy.

4To address these uncertainties, we undertook a contemporary update of the Cochrane systematic review and meta-analyses of RCTs to assess the effects of exercise-based CR in patients with CHD on mortality, clinical events, HRQoL, and cost-effectiveness. We also sought to explore whether intervention effects varied with patient case mix, study and intervention characteristics, and CR delivery settings.

5We conducted and reported this meta-analysis in accordance with the Cochrane Handbook for Interventional Reviews and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses and the synthesis without meta-analysis statements, respectively.

6We undertook update literature searches of Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase, CINAHL, and Science Citation Index Expanded from June 2014 (the search end date of the Cochrane 2016 review) to September 2020 (strategy provided in Supplementary material online, File S1). We also searched two clinical trials registers (World Health Organization’s International Clinical Trials Registry Platform and ClinicalTrials.gov), and hand-searched reference lists of retrieved articles and recent systematic reviews. Records collected from trial registry searches were used to identify trials not picked up in database searches, as well as ongoing studies. We sought RCTs of exercise-based CR (exercise training alone or in combination with psychosocial or educational interventions) compared with no-exercise or usual care control, with at least 6-month post-baseline follow-up outcome measures. All patients in both the intervention and control groups were generally reported to receive (local or national) guideline recommended medical treatment.

7Two reviewers (G.O.D. and J.F.) independently confirmed trial eligibility. Disagreements were resolved by discussion or by a third reviewer (R.S.T.), if necessary.

8We included adults (≥18 years), in either hospital-or community-based settings, who had a myocardial infarction (MI), who had undergone revascularization [coronary artery bypass grafting (CABG) or percutaneous coronary intervention (PCI)], or who had angina pectoris or coronary artery disease defined by angiography.

9Two reviewers (G.O.D. and J.F.) independently completed data extraction and assessed study quality using the Cochrane Risk of Bias (ROB) tool, which was checked by a third reviewer (R.S.T.). Trials were assessed based on random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, and selective reporting. Information regarding study methods (country, design, follow-up, and setting), participant characteristics (numbers randomized, age, sex, diagnosis, and inclusion/exclusion criteria), intervention (exercise mode, duration, frequency, intensity), and control (description, i.e. usual care, no exercise), outcomes, funding sources, and notable author conflicts of interest were obtained.

10Clinical event outcomes included overall and cardiovascular (CV) mortality, fatal and/or non-fatal MI (as reported by studies), CABG, PCI, overall hospitalization, and CV hospitalization. Other outcomes included HRQoL and CR costs, and cost-effectiveness per quality-adjusted life year (QALY). One reviewer (G.O.D.) assessed the certainty of the evidence using Grading of Recommendations Assessment, Development, and Evaluation (GRADE),, and had it checked by a second reviewer (R.S.T.). GRADE assessment was applied to clinical event outcomes (overall and CV mortality, fatal and/or non-fatal MI, CABG, PCI, overall hospitalization, and CV hospitalization) at 6–12 months follow-up, the most frequently reported follow-up time point across trials. Evidence was downgraded from high certainty by one level based on the following domains: limitations in study design or execution (ROB), inconsistency of results, indirectness of evidence, imprecision, and publication bias.

11Outcome data were pooled at the longest reported follow-up and at three separate time periods: ‘short-term’ (6–12 months), ‘medium-term’ (13–36 months), and ‘long-term’ (>36 months) follow-up. Given the level of clinical heterogeneity (variation in CR interventions and populations), we purposively undertook random-effects meta-analyses, using the DerSimonian and Laird random-effects meta-analysis method, assuming that each study estimates a different underlying intervention effect. Dichotomous outcomes (overall and CV mortality, MI, CABG, PCI, and all-cause hospitalization, and CV hospitalization) are expressed as risk ratios (RRs) with 95% confidence intervals (CIs). For those clinical event outcomes with significant risk reductions, we calculated the number needed to treat for an additional beneficial outcome (NNT). Where ≥2 trials reported the same validated HRQoL measures and domains [i.e. Short-Form-36 (SF-36), EuroQol-5D (EQ-5D)], continuous outcomes were pooled separately by each scale and reported as the mean difference (MD) and 95% CI. Given the heterogeneity in HRQoL outcome measures and reporting, for comprehensiveness, we used a vote-counting approach to synthesis in addition to meta-analyses, where the number of positive, negative, and non-significant results was summed. Cost-effectiveness data were synthesized narratively. Statistical heterogeneity was considered substantial where I2 statistic >50%. For outcomes with ≥10 trials included in the meta-analysis, we used the funnel plot and Egger’s test to examine small study bias. The two-sided P-values <0.05 were considered statistically significant. A univariate random-effects meta-regression was used to explore heterogeneity and examine the following pre-defined treatment effect modifiers across clinical event outcomes only: (i) case mix (patients percentage presenting with MI), (ii) ‘dose’ of exercise [dose (units) = number of weeks of exercise training × average sessions per week × average duration of each session in min], (iii) type of CR (exercise only vs. comprehensive CR), (iv) length of follow-up (longest follow-up used where multiple time points are assessed), (v) publication year, (vi) sample size, (vii) CR setting (home or centre based), (viii) ROB (low in <3 of 5 domains), (ix) study continent (Europe, North America, Australia/Asia, or other), and (x) study country status [low-middle-income countries (LMICs) or high-income countries] according to the World Bank Group. Given the number of statistical comparisons performed in this review, the results interpretation was primarily based on 95% CIs rather than P-values. Statistical analyses were performed in RevMan Web version 3.12.1 and STATA version 16.1.

12The search selection process is summarized in Figure 1. Updated database and trial registry searches resulted in a total of 13 783 hits, of which 11 056 unique records were identified, and 244 were selected for full-text review. The main reasons for exclusion were study design (e.g. non-RCT, <6-month follow-up), or use of exercise comparators. The 22 new RCTs (7795 participants; 43 publications), identified in this update, provide a total evidence base of 85 RCTs (145 publications, 23 430 participants) comparing exercise-based CR with a no-exercise control group in patients with CHD. The participants in the newly included trials represent about one-third of all participants included in this study (33%). A complete list of primary and associated supplementary references for included studies is provided in Supplementary material online, File S2.

13A summary of the study, participant, intervention, and comparator characteristics of the 85 included studies is presented in Table 1. Seventy-nine (93%) of the 85 studies were two-arm parallel RCTs, with four studies comparing more than two arms, (two types of CR vs. control),,,, one study using quasi randomization methods, and one cluster RCT. Sixteen of the 22 new trials identified were undertaken in LMICs,,,,, resulting in a total of 21 RCTs in LMICs. Three large multicentre trials contributed a total of 8956 participants (∼40% overall).,, The median age of participants across studies was 56 years, and over the last decade, the percentage of female patients included in trials increased from 11% to 17%. The median CR intervention duration and trial follow-up were 6 and 12 months, respectively. Thirty-eight of the 85 (45%) interventions were exercise only,,,,,,,,,,,,,,,,,,,, with 47 (55%) involving multiple components including education (20 trials),,,,,,,,,,,,,,,,,, psychosocial (seven trials),,,,,,, or a combination of both (16 trials),,,,,,,,,,,,,, or other components (i.e. controlled diet, risk factor management, smoking cessation, relaxation; four trials).,,, Exercise was typically aerobic, with the inclusion of resistance training reported in 27% trials (23 out of 85).,,,,,,,,,,,,,,,,,,,, The dose of exercise interventions varied widely, with frequency ranging between 1 and 7 sessions per week, length of sessions ranging between 20 and 90 min, and intensity ranging between 50% and 90% of maximal or peak heart rate, 50%–95% of aerobic capacity, or at a rating of perceived exertion between 11 and 16. Of the 21 home-based exercise programmes,,,,,,,,,,,,,,,,, four were delivered electronically via mobile phones or the internet.,,,

14The overall ROB of included trials was judged to be low or unclear (see Supplementary material online, Figure S1), and the quality of reporting improved since 2010 (80% of studies had <3 low-ROB domains pre-2010 vs. 55% post-2010). The 30 (35%) trials reported sufficient and appropriate details of random sequence generation,,,,,,,,,,,,,,,,,,, and 23 (27%) reported appropriate allocation concealment,,,,,,,,,,,,,,,,, with 24 (28%) reporting sufficient details of outcome assessment blinding.,,,,,,,,,,,,,,, The 38 (44%) trials were assessed to have low-ROB for incomplete outcome data,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, and 62 (73%) had low-ROB for selective reporting.,,,,,,,,,,,, GRADE assessments for the clinical event outcomes at short-term follow-up ranged from low to high (Table 2), downgrading for imprecision (wide CIs), evidence of publication bias, or substantial statistical heterogeneity.

15A summary of pooled clinical events across all four follow-up time points [longest reported follow-up, short-term (6–12 months), medium-term (13–36 months), and long-term (>36 months)] is presented in Table 2. GRADE assessments for certainty of evidence at short-term (6–12 months) follow-up across clinical event outcomes ranged from low-to-high certainty. We downgraded overall mortality, CV mortality, PCI, and CV hospitalization by one level for imprecision, due to wide CIs that overlapped the boundary with no effect. We downgraded MI and all-cause hospitalization by one level due to evidence of publication bias. We downgraded CV hospitalization by an additional level due to evidence of substantial heterogeneity.

16Of the 60 trials (61 comparisons) that reported overall mortality, 13 trials reported zero events in both arms. There was no difference in risk of overall mortality at short-term follow-up (6–12 months; RR: 0.87, 95% CI: 0.73–1.04, I2 = 0%; moderate certainty evidence) or longest follow-up (47 trials, RR: 0.96, 95% CI: 0.89–1.04, I2 = 0%; Figure 2).

17Across 33 trials (35 comparisons) reporting CV mortality, seven trials reported zero events in both arms. A 26% reduction in risk of CV mortality was seen at longest reported follow-up (26 trials, RR: 0.74, 95% CI: 0.64–0.86, I2 = 0%; Figure 3) with an NNT of 37. At short-term (6–12 months) follow-up, there was no significant difference in CV mortality (RR: 0.88, 95% CI: 0.68–1.14, I2 = 0%, moderate certainty).

18Across 42 trials (44 comparisons) reporting fatal and non-fatal MI, three trials reported zero events in both arms. An 18% reduction in risk was shown at longest follow-up (39 trials, RR: 0.82, 95% CI: 0.70–0.96, I2 = 9%; Figure 4) with an NNT of 100. The overall risk was driven by significant reductions in the short-term (6–12 months; RR: 0.72, 95% CI: 0.55– 0.93, I2 = 7%, high certainty evidence) and long-term (>36 months; RR: 0.67, 95% CI: 0.50–0.90, I2 = 0%) with no difference in the medium-term follow-up (13–36 months; RR: 1.07, 95% CI: 0.91–1.27, I2 = 0%).

19Of 31 trials (33 comparisons) reporting CABG, two trials reported zero events in both arms. There was no difference in risk of CABG at longest follow-up (29 trials, RR: 0.96, 95% CI: 0.80–1.15, I2 = 0%; Figure 5). Of the 20 trials (21 comparisons) reporting PCI, three trials reported zero events in both arms. There was no significant difference in risk of PCI (17 trials, RR: 0.84, 95% CI: 0.69–1.02, I2 = 0%; Figure 6).

20Across 22 trials (24 comparisons) that reported overall hospitalization, one trial reported zero events in both arms. A 23% reduction in overall hospitalization risk with participation in exercise-based CR was shown at the longest follow-up (21 trials, RR: 0.77, 95% CI: 0.67–0.89, I2 = 32%; Figure 7) with an NNT of 37. Nine trials reported CV hospitalizations and one trial reported zero events in both arms. There was no significant difference in CV hospitalization at longest follow-up (eight trials, RR: 0.85, 95% CI: 0.67–1.08, I2 = 12%; Figure 8).

21Six trials reported SF-36 summary component scores with up to 12-month follow-up (Figure 9). There was evidence of increases in both the mental component score (MD: 2.14, 95% CI: 1.07–3.22, I2 = 21%) and the physical component score (MD: 1.70, 95% CI: −0.08–3.47, I2 = 73%) with exercise-based CR. These findings were supported by improvements in selected SF-36 individual domain scores (Figure 10) that included physical functioning, physical performance, general health, vitality, social functioning, and mental health. There was no evidence of an improvement in pooled EQ-5D visual analogue scores (VASs; MD 0.05, 95% CI −0.01–0.10, I2 = 69%; Figure 11).

22Vote-counting across the 32 trials that assessed HRQoL using a range of validated generic or disease-specific outcome measures confirmed the benefit of CR, with 20 (63%) trials reporting higher levels of HRQoL with exercise-based CR compared with control in one or more subscales and 12 (38%) trials reporting higher levels of HRQoL in >50% of the subscales (see Supplementary material online, Table S1).

23Only 8 of the 85 studies reported data on healthcare costs of CR with 5 studies reporting overall healthcare costs in both groups (Table 3). Total healthcare costs were lower with exercise-based CR than usual care in three studies (mean US$2378, €1083, and US$415 less per patient), higher healthcare costs were reported for exercise-based CR than usual care in three studies (mean US$395, US$4,839, and US$480 more per patient), and no difference was reported in one study. However, the difference was significant in only one (mean US$2378/patient; P < 0.001). Acceptable cost-effectiveness ratios per QALY in favour of exercise-based CR were reported in three trials (US$42,535, €15,247, and US$9,200).

24Egger’s tests and visual inspection of funnel plots indicated there was no evidence of small study bias for overall mortality (Egger’s test: P = 0.05; Supplementary material online, Figure S2), CV mortality (Egger’s test: P = 0.20; Supplementary material online, Figure S3), CABG (Egger’s test: P = 0.12; Supplementary material online, Figure S4), and PCI (Egger’s test: P = 0.39; Supplementary material online, Figure S5). However, there was evidence of small study bias with funnel plot asymmetry and significant Egger’s tests for MI (Egger’s test: P = 0.001; Supplementary material online, Figure S6) and all-cause hospitalization (Egger’s test: P < 0.001; Supplementary material online, Figure S7).

25There was no evidence of significant differences in treatment effects across patient, intervention, and study characteristics for all clinical event outcomes (see Supplementary material online, Table S2).

26This updated Cochrane review and meta-analysis of RCTs incorporated data from >23 000 CHD patients and confirmed the benefits of participation in exercise-based CR that include reductions in risk of CV mortality, MI, and all-cause hospitalization at a median follow-up of 12 months (Structured graphical abstract). No significant differences in effect were found across patient case mix, the type or set of CR programme, the dose of exercise prescribed, study sample size, location, length of follow-up, year of publication, and ROB. Reduced hospitalizations are likely to have benefits for both healthcare services as well as for patients in terms of health resource usage and associated costs, and early return home to families and community support networks. Importantly, this updated review demonstrates that the benefits of CR extend across recent trials that are more representative of the modern therapeutic approach in CHD, the expanded CHD population, and low- and middle-income settings (21 trials undertaken in LMICs with 7851 participants), where the prevalence of CHD continues to rise.

27Additionally, we found gains in HRQoL with increased scores across six of the eight SF-36 domains, mental component scores, EQ-5D VAS, and synthesis without meta-analysis across 32 trials reporting HRQoL data. Based on the minimally important clinical differences, the increases in the individual domain scores were not clinically important, but increases in EQ-5D VAS scores could be clinically meaningful. Minimally important clinical differences for the summary component scores are yet to be published for CHD patients. Although HRQoL is important to patients and improvements have been demonstrated in generic measures, this finding might have been more convincing if a generic measure had been accompanied by the additional use of a CHD disease-specific HRQoL measure. To provide more persuasive evidence, we recommend that future trials consider routinely incorporating both types of HRQoL outcome measures for at least 12 months to delineate which, if any, aspects of HRQoL may yield an improvement. Trial-based economic evaluations showed that CR is a cost-effective use of healthcare resources compared with usual care.

28Coronary heart disease is clinically changing from a life-threatening disease to a chronic disease trajectory, as reflected in the terminology of current clinical guidelines on chronic coronary syndromes. This crucial shift strongly calls for interventions that contribute to improvements in the rehospitalization rate and the well-being and HRQoL of people living with chronic diseases. Thus, this latest Cochrane review of RCTs still reinforces the importance of exercise-based CR as part of integrated CHD care alongside modern invasive and pharmacological therapy.

29Our review has a number of potential limitations. First, although we found that the methodological quality and reporting of studies have improved over the last decade and that poor reporting did not appear to alter the review findings, several ROB assessments across trials were judged to be unclear, with many studies inadequately reporting methodologies. Second, this update sought to combine evidence across a range of CHD indications and studies that employed exercise-based CR interventions with varying doses of exercise, delivery settings, and durations of follow-up. However, we applied random-effect meta-analysis to take account of this potential clinical heterogeneity across studies. Furthermore, the GRADE assessment framework also considers heterogeneity in the evidence. For example, the outcomes all-cause mortality, CV mortality, PCI, and CV hospitalization were downgraded in GRADE due to wide CIs that crossed the boundary with no effect. Cardiovascular hospitalization was downgraded due to evidence of statistical heterogeneity (I2 statistic >50%). Thirdly, while studies reported a prescribed dose of exercise, few, if any, reported the actual level of exercise undertaken by participants. So, we were not able to assess the impact of intervention adherence. Fourth, the number of trials reporting follow-up data beyond 12 months has decreased over the last decade, from 48% (between 2000 and 2009) to 23% (between 2010 and 2020). Consequently, the number of deaths and clinical events reported in several trials were low or zero, and these data were often reported within descriptions of trial loss to follow-up rather than as primary or secondary outcomes, which also means that trials would not have been powered for these outcomes. Additionally, hazard ratios were inconsistently reported across trials; therefore, no analyses using these data were possible. Finally, we also found evidence of reporting bias. For example, although 60 trials reported all-cause mortality, only 33 of these same trials reported CV mortality. Sensitivity analysis of the subgroup group of 16 trials that reported both mortality outcomes (see Supplementary material online, Figures S8 and S9) showed improvements in both pooled overall (RR 0.85, 95% CI: 0.74–0.98) and CV mortality (RR 0.79, 95% CI: 0.68–0.92). This sensitivity analysis is in contrast with our main analysis, showing different effects of exercise-based CR on overall mortality and CV mortality.

30The findings of this latest Cochrane review of 85 RCTs in 23 430 CHD patients confirm the clinical outcome benefits of reduced CV mortality, MI, and hospitalization with participation in exercise-based CR and also provide timely evidence that supports the generalizability of these benefits across patients, in the context of contemporary medical management, and across healthcare settings, including LMICs. This updated review also provides meta-analytic evidence that CR participation improves patient quality of life-based on validated HRQoL data. Our findings reinforce the need to improve access to CR for patients with CHD across the globe.