Exercise-based cardiac rehabilitation for chronic heart failure: the EXTRAMATCH II individual participant data meta-analysis

Study design

Randomised controlled trials (RCTs) with a follow-up period of ≥ 6 months (in accordance with the 2014 Cochrane review10).

Target population

Adult patients, aged ≥ 18 years, with a diagnosis of heart failure with reduced ejection fraction (HFrEF) or heart failure with preserved ejection fraction (HFpEF), based on an objective assessment of the left ventricular ejection fraction and on clinical findings.

Setting/context

Patients managed in any setting (i.e. hospital, community facility or patient’s home).

ExCR intervention

An ExCR intervention that included at least an aerobic exercise training component performed by the lower limbs, lasting a minimum of 3 weeks,24 either alone or as part of a comprehensive cardiac rehabilitation programme, which may also include health education and/or a psychological intervention.

Comparator

A non-exercise group receiving standard medical care or an attention placebo.

Sample size

A sample size of > 50 patients to ensure that the logistical effort in obtaining, cleaning and organising the data were commensurate with the contribution of the data set to the analysis. 25,26

Identified RCTs meeting the inclusion criteria are shown in Appendix 3. Study selection for the 2014 Cochrane review and ExTraMATCH IPD meta-analysis was performed by the original research teams that performed these studies. 10,18 For the purposes of this project, a single researcher (RST) compared the included studies from these two previous studies and applied the above inclusion criteria.

Investigator requests

The principal investigators of eligible studies were invited (collaboration invitation) to participate in this IPD meta-analysis and share their anonymised trial data. The list of variables that principal investigators were asked to provide was reported in the study protocol27 (see Appendix 4).

Exclusion of trials from individual participant data analysis

Trials were excluded if:

• authors did not respond to the invitation to provide IPD for the ExTraMATCH II analysis in spite of repeated contact attempts being made

• authors were unable to provide IPD, because the data had been either lost or destroyed

• patients were included in the trial who may also have appeared in another IPD data set.

Ethics approval

The ethics of obtaining data were carefully considered and advice was sought from NHS Digital in April 2016. The original trials had each obtained ethics/institutional review board committee approval and obtained individual patient consent. Given the fully anonymised nature of all the trial data sets (i.e. no inclusion of data, such as patient name or date of birth, that would allow individual patients to be identified), NHS Digital confirmed that there was no further legal/ethics or contractual requirements for use of these data for the purpose of this project. A revision of the HF-ACTION19 data was obtained via the NIH data portal, which required that we obtain a letter of approval from the University of Exeter Medical School Research Ethics Committee. A letter was received on 13 November 2017.

Data management

Data files were received in a variety of formats depending on the security concerns of the host institutions. In most cases, data transfer was by e-mail of a password-protected file, with a separate e-mail containing the password. Each raw data file was saved in its original format on receipt and then converted to a Stata^® file (version 14.2, StataCorp LP, College Station, TX, USA). Data cleaning was carried out in each pseudonymised data set prior to being combined in a master data set. Within the individual data sets, data for each variable (at the patient level) was checked for accuracy in range, extreme values, internal consistency, missing values and consistency with published reports. Data discrepancies or missing information were discussed with trial investigators and corrected, if appropriate.

All data files were stored on a secure password-protected computer server managed in accordance with the data management standard operating procedures of the UK Clinical Research Collaboration-registered Exeter Clinical Trials Unit. Access to data at all stages of cleaning and analysis was restricted to core members of the research team (OC, RST, FW and SW).

Main outcomes

In accordance with the study research objectives, it was sought, from eligible trials, IPD for the following outcomes:

• Mortality – incidence and time-to-event data for all deaths (we also sought to obtain data on the cause of death).

• Hospital admission – incidence, time to event and duration of hospitalisation (we also sought to obtain data on the cause of the hospitalisation).

• Disease-specific HRQoL (as assessed by the MLHFQ and other validated HRQoL outcomes) – value at baseline (pre randomisation) and at 6, 12, 24 and > 24 months post randomisation.

• Exercise capacity [as assessed by peak oxygen uptake (VO₂peak) and other validated exercise capacity measures] – as measured at baseline and at 6, 12, 24 and > 24 months post randomisation.

Patient subgroups

We also requested individual patient demographic and clinical data, including age, sex, ejection fraction, NYHA class, HF aetiology (ischaemic vs. non-ischaemic), race/ethnicity and exercise capacity at baseline. Details of exercise training prescription (i.e. session frequency, duration, intensity and overall programme duration) was collected as part of the 2014 Cochrane review. 10

Statistical analysis plans

A detailed statistical analysis plan was produced for each of the three analyses described below:

1. the impact of ExCR on mortality and hospitalisation outcomes

2. the impact of ExCR on HRQoL and exercise capacity outcomes

3. the validation of exercise capacity as a surrogate outcome.

Descriptive statistics

For each analysis, patient-level characteristics were compared for those patients in the ExCR and control groups of the included studies. Descriptives of trial-level characteristics by group are also reported.

Assessment of study quality and risk of bias

We checked for potential small-study bias by visually assessing funnel plot asymmetry and using Egger’s test. 30 Study quality and risk of bias was assessed using the Tool for the assEssment of Study qualiTy and reporting in EXercise (TESTEX). 31 Statistical heterogeneity was assessed using the I²-statistic. 29

Inclusion of trials

Trials were included in the mortality and hospitalisation analyses if IPD was provided for the one or more of the outcomes of interest detailed below.

Outcomes of interest

The final patient-relevant outcomes of interest in this study were time to event to:

• all-cause mortality

• HF-related mortality

• all-cause hospital admission

• HF-related hospital admission.

Owing to the inconsistency of reporting in IPD sets, we were able to consider only time-to-event outcomes and not incidence or duration of events. Insufficient data were made available to allow analyses on ‘sudden death’ to be carried out.

Each of the outcomes described above was analysed separately. Each trial contributed to between one and four analyses.

Primary analysis

In the primary analysis, a two-stage IPD meta-analysis approach was taken. A Cox regression model was performed on the data from each trial individually and the resulting HRs used in a random-effects meta-analysis. For the meta-analysis of treatment–covariate interactions, the same approach was used, with a Cox regression model applied to the data from each trial and the resulting HR for the interaction effect used to inform the meta-analysis. A random-effects model was used to account for the high degree of clinical heterogeneity across the individual studies due to differences in population, ExCR intervention and comparator. 29 An overall estimate of the effect of ExCR for each outcome, both by trial and as a pooled estimate, was presented as a HR and 95% CI. Additionally, the and I² and τ² statistics were reported alongside the associated p-value for the results of the main analyses. 29,32 The Cochrane handbook advises that using specific threshold values for the interpretation of the I²-statistic can be misleading. 33

Secondary analysis

The secondary analysis used a one-stage IPD meta-analysis Cox regression model, stratified by trial. Stratification allowed the baseline hazard to vary between studies, rather than forcing the hazard in individual studies to be proportionate to each other. 34 No distributional assumptions about this baseline hazard were made. Owing to failure of convergence in the one-stage random-effect models, probably due to the low level of heterogeneity between studies, a fixed-effect approach was used.

The within-trials interaction term used here identifies any patient characteristics that influence the effectiveness of ExCR on an individual level, necessary for making inferences for stratified medicine, as recommended by Riley et al. 35 The within-trial interaction effect is fixed across trials. Continuous covariates were centred on the mean value within each trial; binary covariates were centred on the proportion within each trial.

Sensitivity analyses

To test the robustness of the primary and secondary analyses, we undertook a number of prespecified sensitivity analyses: we excluded the largest trial (HF-ACTION19); truncated outcomes at 1, 2 and 5 years’ follow-up; and included trial-level outcome data for studies that could not provide IPD. 26

Inclusion of trials

Trials were included in the HRQoL and exercise capacity analyses if IPD was provided for the one or more of the outcomes of interest detailed below.

Outcomes of interest

The final patient-relevant outcomes of interest in this study were:

• HRQoL measured using the MLHFQ score

• HRQoL measured through any validated scale

• exercise capacity measured using VO₂peak (ml/kg/minute)

• exercise capacity measured using the 6-minute walk test (6MWT) (m)

• exercise capacity measured using a standardised exercise capacity score calculated from any of the four validated exercise capacity measures [i.e. VO₂peak, 6MWT, incremental shuttle walk test (ISWT) and workload on cycle ergometer].

Health-related quality of life: scales of measurement

Health-related quality of life measured using one of the following three validated measures was included in this analysis:

1. the MLHFQ36

2. the Kansas City Cardiomyopathy Questionnaire37

3. the Guyatt et al. 38 Chronic Heart Failure Questionnaire scale.

The first HRQoL analysis was carried out for trials providing the MLHFQ data; the second analysis used a standardised score calculated from any of the three measures above.

Exercise capacity: scales of measurement

Exercise capacity measured using one of four validated measures was included in this analysis:

1. VO₂peak (ml/kg/minute)

2. distance (m) walked on the 6MWT

3. distance (m) walked in an ISWT

4. workload on cycle ergometer (watts).

Exercise capacity analysis was carried out for:

• trials providing VO₂peak

• trials providing the 6MWT

• a standardised exercise capacity score, calculated from any of the validated exercise capacity measures listed above.

One study, HF-ACTION,19 provided data on both VO₂peak and the 6MWT, and was included in all analyses, with the VO₂peak measure taking precedence for the standardised exercise capacity analysis.

Primary analysis

The primary analyses included one- and two-stage IPD meta-analyses carried out at 6 and 12 months. At each time point, we used the observation closest to and prior to the time point. All one-stage IPD models used a hierarchical random-effects regression model, adjusted for the baseline value of the outcome measure. All two-stage models used random treatment effects. We performed a series of models to estimate the overall treatment effect and to investigate potential interactions between ExCR and predefined patient subgroups (i.e. age, sex, left ventricular ejection fraction, HF aetiology, NYHA class and baseline exercise capacity23,27). Each model investigated one interaction effect only. The I² and τ² statistics were reported alongside the associated p-value for the results of the main analyses. 29,32

Secondary analysis

The secondary analyses used a random-effects hierarchical model for repeated measures at multiple time points. These models utilised HRQoL and exercise capacity outcome data at all available time points. Adjustments for baseline values of the outcome measure were made; no other covariates were included in the model. This model included a time-by-treatment interaction term.

Sensitivity analyses

To test the robustness of the primary analyses, prespecified sensitivity analyses were carried out:

• the primary analysis was repeated after exclusion of the largest trial (HF-ACTION19)

• data was added from studies that did not provide IPD.

Inclusion of trials

All studies in the ExTraMATCH II meta-analysis were eligible for inclusion in the surrogate analyses, dependent on the availability of data on exercise capacity and final patient-relevant outcomes, as explained below.

Outcomes of interest

The final patient-relevant outcomes of interest in this validation study were:

• HRQoL measured by MLHFQ score

• HRQoL measured through any validated scale

• time to all-cause mortality

• time to all-cause hospital admission.

For this study, three approaches to exercise capacity definition were used:

1. direct assessed VO₂peak

2. 6MWT

3. direct and indirect VO₂peak (conversion from the 6MWT and the ISWT; no conversion was possible for watts as it is dependent on the body weight of individual patients).

Distances recorded as either 6MWT or ISWT at baseline were converted to VO₂peak using previously reported methods. 39–43 Details can be found in Appendix 5.

Follow-up time considerations

The following outcome follow-up times were considered: ≤ 6 months for exercise capacity outcomes, ≤ 12 months for HRQoL outcomes and all available follow-up time for mortality and hospitalisation. This approach was consistent with the assumption of temporal antecedence for a causal relationship between the surrogate end point and the final outcomes.

Mediation analysis

Mediation is known as the phenomenon whereby a cause affects an intermediate variable (also called mediator), and the change in the intermediate variable goes on to affect the outcome. 44,45 The effect of the cause on the outcome that operates through the intermediate of interest is sometimes referred to as an indirect or mediated effect. Mediation analysis usually refers to the set of techniques by which a researcher assesses the relative magnitude of these direct and indirect effects. The product method specification of this approach was used to determine whether or not a change in VO₂peak (ΔVO₂peak) or a change in the 6MWT result (Δ6MWT) mediate the relationship between treatment assignment (i.e. ExCR vs. no ExCR), and each of the final outcomes of interest. Linear or Cox regression analyses were conducted to evaluate the following four hypotheses:

1. Treatment assignment (i.e. ExCR vs. control) has a significant effect on ΔVO₂peak or Δ6MWT from baseline to 6 months’ follow-up.

2. ΔVO₂peak or Δ6MWT have a significant effect on ΔMLHFQ or ΔHRQoL, or on the hazards of developing a clinical event.

3. Treatment assignment (i.e. ExCR vs. control) has a significant effect on ΔMLHFQ or ΔHRQoL, or on the hazards of developing a clinical event.

4. The effect of treatment assignment (i.e. ExCR vs. control) on ΔMLHFQ or ΔHRQoL, or on the hazards of developing a clinical event, is attenuated when ΔVO₂peak or Δ6MWT is added to the model.

All regression models took into account the clustering within trials to allow for study-level differences in treatment effect and unstructured covariance between random intercept and random slope. Regression models were adjusted for baseline of either exercise capacity values or baseline HRQoL values. No other adjustments were made, because patients were randomly assigned to intervention or control arm. In testing hypothesis (2), no adjustment is made for potential confounding variables.

It was assumed necessary to reject the null for at least the first of these hypotheses (i.e. the treatment assignment is associated with the mediator), to support the validation of ΔVO₂peak or Δ6MWT as mediator end points and proceed further with the estimation of proportion explained or proportion mediated.

Meta-analytic approach: R² and surrogate threshold effect

Although mediation analysis considers pathways by which treatment effects may arise, surrogacy principally concerns whether or not we are able to predict the effect of treatment on the final end point by using the effect of treatment on the surrogate.

Given the issues described with the proportion explained and indirect effects approaches in identifying consistent surrogates, the meta-analytic approach may offer the most promise for assessing surrogate outcomes and for making policy and treatment decisions. 46,47 This approach requires multiple studies, or at least multiple subgroups (e.g. centres within a trial), which we have through the ExTraMATCH II IPD meta-analysis. As a true and strong association between the treatment effect on the final end point and the treatment effect on the surrogate is considered to be the hallmark of surrogacy,47 this approach proceeds as follows. Let ϕ_j denote the estimate of the effect of treatment on the final outcome in the jth study, let θ_j denote the estimate of the effect of treatment on the surrogate outcome in the jth study, both derived from RCTs. For a good surrogate, a monotonic relationship would exist between ϕ_j and θ_j and, in a regression of ϕ_j on θ_j, there would be limited variability around the regression line. If the relationship between ϕ_j and θ_j is approximately linear, a reasonable measure of surrogacy is the R²_trial of the regression of ϕ_j on θ_j. Another intuitive measure recommended as a surrogacy metric is the surrogate threshold effect (STE), which takes into account the variability around the regression line and represents the intercept of the prediction band of the regression line with the zero effect line on the final outcome. 46 For each trial, we estimated study-level treatment effects by conducting linear regression or Cox proportional hazards regression models. Adjustment was made for baseline exercise capacity or HRQoL values. Then we conducted linear meta-regressions to relate estimated difference in exercise capacity to the estimated effect on change in HRQoL: log(HR) of all-cause mortality or log(HR) of all-cause hospitalisation events. The square of the inverse standard error was used as a weight to account for uncertainty in the estimated patient-relevant outcomes effect. We calculated commonly reported indicators of surrogate validation. 48 The correlation coefficient (p-value) and the R² for the relationship between treatment effect difference on exercise capacity and each of the final outcomes was estimated individually using weighting by the inverse of the variance (for the treatment effect on final outcomes). In order to estimate STE, prediction bands were calculated based on approximate prediction intervals. 48,49

Primary analysis

Compared with control, all time-to-event mean treatment effects from the two-stage random-effects IPD meta-analysis were in favour of ExCR, but had wide CIs and were not statistically significant [all-cause mortality: HR 0.83 (95% CI 0.67 to 1.04; p = 0.107, 17 studies,19,50–55,57,58,60–67 3782 patients, τ² = 0.04, I² = 26%); HF-specific mortality: HR 0.84 (95% CI 0.48 to 1.46; p = 0.527, 9 studies,51–54,58,60,64,65,67 915 patients, τ² = 0.00, I² = 0%); all-cause hospitalisation: HR 0.90 (95% CI 0.76 to 1.06; p = 0.210, 11 studies,19,51,54,55,57,58,60,61,64–66 3190 patients, τ² = 0.01, I² = 12.4%); and HF-specific hospitalisation: HR 0.98 (95% CI 0.72 to 1.35; p = 0.902, 13 studies,19,50,51,52,56–58,60,61,64–67 3494 patients, τ² = 0.10, I² = 45%)] (Figure 4 and Tables 7–10).

FIGURE 4.

Effect of ExCR on mortality and hospitalisation across patient subgroups: two-stage IPD meta-analysis. (a) All-cause mortality; (b) HF-specific mortality; (c) all-cause hospitalisation; and (d) HF-specific hospitalisation. DANREHAB, DANish Cardiac ReHABilitation trial.

Interaction analyses for the two-stage model revealed no consistent interaction between the effect of ExCR and any of the predefined subgroups (age, sex, ejection fraction, NYHA class, HF aetiology, ethnicity or baseline exercise capacity) for all-cause mortality, HF-related mortality, all-cause hospitalisation or HF-related hospitalisation (see Tables 7–10). In order to make further comparisons of mortality and hospitalisation rates within each subgroup, the HR and associated 95% CI from individual subgroup one-stage IPD meta-analyses are shown in Figure 5. The p-values from the interaction test in the two-stage IPD meta-analyses are presented alongside these estimates.

FIGURE 5.

Effect of ExCR on mortality and hospitalisation across patient subgroups: individual subgroup one-stage IPD meta-analyses. (a) All-cause mortality; (b) HF-related mortality; (c) all-cause hospitalisation; and (d) HF-related hospitalisation. a, Although stratified meta-analyses are shown, the interaction p-values are calculated based on continuous distribution of age, ejection fraction and baseline exercise capacity.

Secondary analyses

These primary analysis results were broadly consistent across secondary analyses.

Sensitivity analyses

In sensitivity analyses, four weak interaction effects (p < 0.05) were seen:

1. Age compared with all-cause mortality (p = 0.034) in the two-stage 2-year truncation model (i.e. larger reduction in all-cause mortality with ExCR in older patients).

2. Age compared with HF-related mortality (p = 0.017) in the two-stage 2-year truncation model (i.e. larger reduction in HF-related mortality with ExCR in older patients).

3. Ischaemic status compared with HF-related mortality (p = 0.047) in the one-stage model (i.e. larger reduction in HF-related mortality with ExCR in ischaemic patients).

4. Standardised baseline exercise capacity compared with all-cause hospitalisation (p = 0.027) in the two-stage 1-year truncation model (i.e. larger reduction in all-cause hospitalisation with ExCR in patients with lower than average baseline exercise capacity) (see Tables 7–10).

Inferences did not change following the addition of trial-level data from trials that met the study inclusion criteria but did not contribute IPD (data are not shown here, but are available from the authors of this report).

Primary analysis

Compared with control, treatment effects from the one-stage meta-analysis at the 12-month follow-up showed a significant improvement with ExCR in exercise capacity as assessed by the 6MWT (mean difference 21.0 m, 95% CI 1.57 to 40.4 m; p = 0.034, τ² = 491, I² = 78%) and standardised exercise capacity score (mean difference 0.27 SD units, 95% CI 0.11 to 0.43; p = 0.001, τ² = 0.08, I² = 91%). No significant difference in VO₂peak at 12 months was observed (1.01 ml/kg/minute, 95% CI –0.42 to 2.44 ml/kg/minute; p = 0.168, τ² = 2.17, I² = 94%).

One-stage meta-analysis showed a significant improvement in HRQoL as assessed by the MLHFQ (mean difference –5.94 points, 95% CI –1.0 to –10.9 points; p = 0.018, τ² = 77, I² = 88%) and standardised HRQoL score (mean difference SD 0.20, 95% CI 0.03 to 0.37; p = 0.020, τ² = 0.07, I² = 85%), at the 12-month follow-up. Similar results were seen at the 6-month follow-up (Figures 8 and 9). Marked statistical heterogeneity (I² > 70%) was seen for all exercise capacity and HRQoL outcomes.

FIGURE 8.

Effect of ExCR on HRQoL and exercise capacity at 6 months: two-stage IPD meta-analysis. (a) MLHFQ score (points) (mean difference); (b) all HRQoL measures (standardised mean difference); (c) VO₂peak directly reported (mean difference); (d) 6MWT (m) directly reported (mean difference); and (e) all exercise capacity measures (standardised mean difference). Note that weights are from random effects, DerSimonian–Laird estimator. CBT, cognitive–behavioural therapy.

FIGURE 9.

Effect of ExCR on HRQoL and exercise capacity at 12 months: two-stage IPD meta-analysis. (a) MLHFQ score (points) (mean difference); (b) all HRQoL measures (standardised mean difference); (c) VO₂peak directly reported (mean difference); (d) 6MWT (m) directly reported (mean difference); and (e) all exercise capacity measures (standardised mean difference). Note that weights are from random effects, DerSimonian–Laird estimator. CBT, cognitive–behavioural therapy.

Analyses revealed no consistent interaction between the effect of ExCR and the predefined subgroups (sex, ejection fraction, NYHA class, HF aetiology, ethnicity and baseline exercise capacity), for either exercise or HRQoL.

A differential effect of ExCR across ages was observed in the standardised HRQoL analysis, with a reduction in HRQoL score (i.e. an increase in standardised HRQoL score) as age increased (SD 0.006, 95% 0.002 to 0.011; p = 0.006). To put this into context, based on a MLHFQ SD of 24 points, this equates to a decrease of 1.4 points in the treatment effect on the MLHFQ score for every 10-year increase in patient age.

Interaction analyses for the one-stage model at the 12-month follow-up showed differential effects of ExCR by sex, with women showing greater benefit than men for VO₂peak (0.57 ml/kg/minute, 95% CI 0.04 to 1.11 ml/kg/minute; p = 0.036) and the 6MWT (14.9 m, 95% CI 1.2 to 28.7 m; p = 0.034). Differential effects of ExCR were also seen between ethnic groups, with white patients showing a greater improvement in 6MWT distance than non-white patients (14.2 m, 95% CI 0.40 to 28.0 m; p = 0.044).

Secondary analysis

In the repeated measures analyses for each HRQoL and exercise capacity outcome, a significant interaction between ExCR and time was observed (Figure 10). To visualise comparisons of changes in HRQoL and exercise capacity in each subgroup, the effect estimates and associated 95% CI from individual subgroup one-stage IPD meta-analyses are shown in Figure 11. The p-values from the interaction test in the two-stage IPD meta-analyses are presented alongside these estimates.

FIGURE 10.

Effect of ExCR on HRQoL and exercise capacity. (a) MLHFQ score (points) (mean difference); (b) all HRQoL measures (standardised mean difference); (c) VO₂peak directly reported (mean difference); (d) 6MWT directly reported (mean difference); and (e) all exercise capacity measures (standardised mean difference).

FIGURE 11.

Effect of ExCR on HRQoL and exercise capacity across patient subgroups (individual subgroup one-stage IPD meta-analyses). (a) MLHFQ score (points); (b) all HRQoL measures (standardised score); (c) VO₂peak directly reported; (d) 6MWT (m) directly reported; and (e) all exercise capacity measures (standardised score). a, Although stratified meta-analyses are shown, the interaction p-values are calculated based on continuous distribution of age, ejection fraction and baseline exercise capacity.

Sensitivity analyses

In sensitivity analyses, the results of the analyses excluding HF-ACTION,19 were broadly consistent with the overall results (Tables 14–18). Similar results were found with the addition of the study-level aggregate data to the two-stage model at 12 months’ follow-up.

Mediation analysis

The four criteria that must be satisfied to establish that change in exercise capacity is a mediator of mortality, hospitalisation and change in HRQoL are listed in Table 23. First, mean improvements were seen in all exercise capacity metrics of ExCR compared with control, although none reached statistical significance at p < 0.05. Second, greater differences in exercise capacity significantly reduced the risk of mortality and hospitalisation and were associated with a larger gain in HRQoL. Third, although ExCR decreased both the risk of mortality and hospitalisation, and was also associated with a larger gain in HRQoL, there was no statistically significant difference compared with the control. Finally, the effect of ExCR compared with control on final patient-relevant outcomes was attenuated by adding Δ6MWT and ΔVO₂peak (directly and indirectly measured) to the model. No attenuation was seen with the addition of ΔVO₂peak when measured directly.

Meta-analytic regression: R² and surrogate threshold effect

Regression coefficients of determination (R²) and correlation coefficients (p-value) between the change in exercise capacity and hospitalisation were poor (R²_trial < 50% and p < 0.50). Moderate to good levels of correlation (R²_trial > 50% and p > 0.50) between exercise capacity VO₂peak and 6MWT with mortality and HRQoL were seen (Table 24). The STE for MLHFQ score ranged from an increase of 1.6 to 4.6 ml/kg/minute for VO₂peak. The STE was not estimable for the 6MWT. Negative correlation coefficients indicate that larger ExCR effects on exercise capacity are associated with larger ExCR effects on mortality and HRQoL. Figures 13–15 illustrate the results of the meta-regression and STE analyses.

FIGURE 13.

Regression analyses: relationship at the 6-month follow-up between ΔVO₂peak direct and (a) log(HR) of all-cause mortality; (b) ΔHRQoL all outcomes; (c) log(HR) of all-cause hospitalisation; and (d) ΔMLHFQ score. Circles represent trial-level treatment effects, with sizes proportionate to study weights (based on inverse variance weighting). Solid green lines correspond to the bounds of the 95% prediction interval for the true effect on the final outcomes in a new study. SMD, standardised mean difference.

FIGURE 14.

Regression analyses: relationship at the 6-month follow-up between Δ6MWT and (a) log(HR) of all-cause mortality; (b) ΔHRQoL all outcomes; (c) log(HR) of all-cause hospitalisation; and (d) ΔMLHFQ score. Circles represent trial-level treatment effects, with sizes proportionate to study weights (based on inverse variance weighting). Solid green lines correspond to the bounds of the 95% prediction interval for the true effect on the final outcomes in a new study. SMD, standardised mean difference.

FIGURE 15.

Regression analyses: relationship between ΔVO₂peak direct and indirect and (a) log(HR) of all-cause mortality; (b) ΔHRQoL all outcomes; (c) log(HR) of all-cause hospitalisation; and (d) ΔMLHFQ score. Circles represent trial-level treatment effects, with sizes proportionate to study weights (based on inverse variance weighting). Solid green lines correspond to the bounds of the 95% prediction interval for the true effect on the final outcomes in a new study. SMD, standardised mean difference.

Small-study bias

There was no evidence of significant small-study bias, as shown by the funnel plots (Figure 16) or Egger’s test p-values, for any of the exercise capacity outcomes (ΔVO₂peak direct, p = 0.699; Δ6MWT, p = 0.93; ΔVO₂peak direct and indirect, p = 0.553), or for the four patient-relevant final outcomes (ΔMLHFQ score, p = 0.607; ΔHRQoL outcomes, p = 0.659; mortality, p = 0.745; hospitalisation, p = 0.733).

FIGURE 16.

Funnel plots for the surrogate analyses. (a) VO₂peak; (b) 6MWT; (c) converted exercise capacity score; (d) HRQoL; (e) MLHFQ score; (f) mortality; and (g) hospitalisation.

Search strategy

1. xp Myocardial Ischemia/

2. myocard$4 adj5 (ischaemi$2 or ischemi$2)).ti,ab.

3. (ischaemi$2 or ischemi$2) adj5 heart).ti,ab.

4. xp Coronary Artery Bypass/

5. oronary.ti,ab.

6. xp Coronary Disease/

7. xp Myocardial Revascularization/

8. Myocardial Infarction/

9. myocard$5 adj5 infarct$5).ti,ab.

10. (heart adj5 infarct$5).ti,ab.

11. exp Angina Pectoris/

12. angina.ti,ab.

13. exp Heart Failure/

14. (heart adj5 failure).ti,ab.

15. (HFNEF or HFPEF or HFREF or ‘HF NEF’ or ‘HF PEF’ or ‘HF REF’).ti,ab.

16. or/1-15

17. exp Heart Diseases/

18. (heart adj5 disease$2).ti,ab.

19. myocard$5.ti,ab.

20. cardiac$2.ti,ab.

21. CABG.ti,ab.

22. PTCA.ti,ab.

23. (stent$4 and (heart or cardiac$4)).ti,ab.

24. Heart Bypass, Left/or exp Heart Bypass, Right/

25. or/17-24

26. *Rehabilitation Centers/

27. exp Exercise Therapy/

28. *Rehabilitation/

29. exp Sports/

30. Physical Exertion/or exertion.ti,ab.

31. exp Exercise/

32. rehabilitat$5.ti,ab.

33. (physical$4 adj5 (fit or fitness or train$5 or therap$5 or activit$5)).ti,ab.

34. (train$5 adj5 (strength$3 or aerobic or exercise$4)).ti,ab.

35. ((exercise$4 or fitness) adj5 (treatment or intervent$4 or programs$2 or therapy)).ti,ab.

36. Patient Education as Topic/

37. (patient$2 adj5 educat$4).ti,ab.

38. ((lifestyle or life-style) adj5 (intervent$5 or program$2 or treatment$2)).ti,ab.

39. *Self Care/

40. (self adj5 (manage$5 or care or motivate$5)).ti,ab.

41. *Ambulatory Care/

42. exp Psychotherapy/

43. psychotherap$2.ti,ab.

44. (psycholog$5 adj5 intervent$5).ti,ab.

45. relax$6.ti,ab.

46. exp Relaxation Therapy/or exp Mind-Body Therapies/

47. exp Counseling/

48. (counselling or counseling).ti,ab.

49. exp Cognitive Therapy/

50. exp Behavior Therapy/

51. ((behavior$4 or behaviour$4) adj5 (modify or modificat$4 or therap$2 or change)).ti,ab.

52. *Stress, Psychological/

53. (stress adj5 management).ti,ab.

54. (cognitive adj5 therap$2).ti,ab.

55. meditat$4.ti,ab.

56. *Meditation/

57. exp Anxiety/

58. (manage$5 adj5 (anxiety or depress$5)).ti,ab.

59. CBT.ti,ab.

60. hypnotherap$5.ti,ab.

61. (goal adj5 setting).ti,ab.

62. (goal$2 adj5 setting).ti,ab.

63. (psycho-educat$5 or psychoeducat$5).ti,ab.

64. (motivat$5 adj5 (intervention or interv$3)).ti,ab.

65. Psychopathology/

66. psychopathol$4.ti,ab.

67. psychosocial$4.ti,ab.

68. distress$4.ti,ab.

69. exp Health Education/

70. (health adj5 education).ti,ab.

71. (heart adj5 manual).ti,ab.

72. Autogenic Training/

73. autogenic$5.ti,ab.

74. or/26-39

75. or/40-73

76. 16 or 25

77. 74 or 75

78. 76 and 77

79. randomized controlled trial/

80. randomized controlled trial.pt.

81. controlled clinical trial.pt.

82. controlled clinical trial/

83. Random Allocation/

84. Double-Blind Method/

85. single-blind method/

86. (random$ or placebo$).ti,ab.

87. ((singl$3 or doubl$3 or tripl$3 or trebl$3) adj5 (blind$3 or mask$3)).ti,ab.

88. exp Research Design/

89. Clinical Trial.pt.

90. exp clinical trial/

91. (clinic$3 adj trial$2).ti,ab.

92. or/79-91

93. 78 and 92

94. (Animals not Humans).sh.

95. 93 not 94

96. limit 95 to yr=‘2008 -Current’