The ExTraMATCH II project is a meta-analysis of IPD from HF patients recruited to RCTs conducted worldwide, which sought to determine which HF patient subgroups benefit most from ExCR and assess the suitability of exercise capacity as a surrogate end point.
Summary of findings
Of the 37 eligible trials, 19 contributed data to the IPD meta-analysis: 18 trials (3912 patients) to the clinical events (mortality and hospitalisation) analysis; 13 trials (3332 patients) to the exercise capacity and HRQoL analysis; and 10 trials (2656 patients) to the exercise capacity mediational/surrogate end-point analysis.
Patient characteristics at baseline were well balanced between patients in the ExCR and control groups. The majority of patients were male (74%), had a mean age of 61 years, had experienced HFrEF (mean left ventricular ejection fraction of 26.9%) and were in NYHA functional class II (59%) or III (38%). Trials from Europe and North America were published between 1990 and 2012. Sample sizes ranged from 50 to 2130 patients. All trials evaluated an aerobic exercise intervention, which was most commonly delivered in either an exclusively centre-based setting or a centre-based setting in combination with some home exercise sessions. The dose of exercise training ranged widely across trials. ExCR was delivered over a period of 15–90 weeks, with between two and seven sessions per week (median session duration was between 4 and 120 minutes, including warm-up and cool-down). The intensity of exercise ranged between 50% and 85% VO2peak. The overall quality of included trials was judged to be moderate to good, with a median TESTEX score31 of 11 (range 9–14) out of a maximum score of 15.
Compared with no exercise control, ExCR did not have a statistically significant effect on the risk of mortality and hospitalisation. However, uncertainty around effect estimates precludes drawing definitive conclusions for these event outcomes. In contrast, ExCR was found to significantly improve both exercise capacity and HRQoL, the improvement in MLHFQ score being also clinically important (i.e. a mean reduction of ≥ 5 points).74 We found no consistent differences in ExCR effects across patient subgroups (i.e. age, sex, ethnicity, NYHA functional class, ischaemic aetiology, ejection fraction and baseline exercise capacity) on mortality, hospitalisation, exercise capacity or HRQoL. The validation of exercise capacity as a putative surrogate end point for patient-relevant outcomes (i.e. mortality, hospitalisation and HRQoL) was limited by access to only a small number of trials that were able to contribute suitable patient-level data. Although subject to considerable statistical uncertainty, the results provide indicative evidence that VO2peak and 6MWT may be suitable surrogate end points for the treatment effect of ExCR on final outcomes in patients with HF.
Comparison to existing evidence
The finding of a lack of consistent evidence for HF patient subgroup effects of ExCR agrees with both the previous ExTraMATCH18 and Cochrane analyses.10 However, these two previous studies had major limitations that are likely to have limited their ability to detect subgroup effects. ExTraMATCH18 included data on 801 HF patients and observed 88 deaths and 300 patients with a composite outcome of death or hospitalisation and, therefore, lacked statistical power. Using meta-regression analysis, the 2014 Cochrane review10 found no association between trial-level patient characteristics and ExCR. However, meta-regression analysis is highly prone to study-level confounding (ecological fallacy) and should be interpreted with great caution.75
The findings are also consistent with the IPD subgroup analyses from the multicentre HF-ACTION study.19 The HF-ACTION19 investigators reported no significant interaction effect of exercise training intervention on their composite primary outcome (i.e all-cause mortality or hospitalisation) and subgroups of age (≤ 70 vs. > 70 years), sex, race (white vs. non-white), HF aetiology (ischaemic vs. non ischaemic), ejection fraction (≤ 25% vs. > 25%) or NHYA class (II vs. III/IV).19 A post hoc analysis by HF-ACTION19 investigators found a significant (adjusted p = 0.02) interaction between ExCR and the change in 6MWT with ExCR and ethnicity (+26 m in black patients vs. +11 m in white patients), consistent with the current study.76
The validation study results of the suitability of exercise as surrogate outcome, albeit uncertain, are broadly in agreement with this research team’s recent study based on a trial-level meta-analysis.77
Strengths and limitations
The ExTraMATCH II project has a number of strengths. The IPD meta-analysis is the largest to date and has greater power to detect any differential treatment effect across groups than single trials or aggregate meta-analysis. We were able to standardise the handling and analysis of time-to-event outcomes and continuous outcomes across trials. We found no evidence of publication bias. The project was conducted and reported in accordance with current IPD guidance and the PRISMA IPD statement.21,78
Although systematic reviews and meta-analyses of IPD from randomised trials are recognised as the gold standard for assessing intervention effects,79 the study has a number of limitations. First, there was a lack of consistency in how included trials with IPD in the analyses defined and collected the outcomes of interest (i.e. time to event for death and hospitalisation, exercise capacity and HRQoL). We made considerable efforts to contact study authors to clarify issues around the definition of outcomes, especially HF-related mortality and hospitalisations. Although we were able to resolve data issues in many cases, we recognise that a lack of consistency in outcome definition across included trials may exist, weakening the strength of these conclusions. Second, we were not able to obtain IPD from all includable trials for all outcomes; not all investigators for the trials that met the inclusion criteria were able to provide IPD and, of the trials that did provide IPD, not all collected the outcomes of interest. For example, the large NIH-funded US multicentre HF-ACTION study19 did not collect HF-specific hospitalisation data,19 thus reducing the statistical power for this outcome. Third, we did not seek patient-level data on ‘ExCR dose’ (i.e. adherence according to exercise training duration, frequency and intensity undertaken by an individual patient). Using IPD from HF-ACTION,19 Keteyian et al.80 found exercise volume (defined as metabolic equivalent of task hours per week) to be a predictor for the composite outcome of all-cause mortality or hospitalisation (p = 0.03). Fourth, there were high levels of statistical heterogeneity for both exercise capacity and HRQoL outcomes. This heterogeneity may well have reflected the variation in ExCR interventions across the included trials. Fifth, the analysis is based on randomised trials identified by literature searches up to 2013 and, therefore, did not include IPD from more recent trials that may have met the inclusion criteria of this study.
Finally, in terms of the surrogate validation analysis, a particular limitation was the proportion of included trials that provided patient-level data on both exercise capacity and patient-relevant outcomes. Of the 19 trials (3990 patients) that met the inclusion criteria, only 10 trials (2656 patients) provided paired data on exercise capacity and mortality, hospitalisation or HRQoL. This has a number of implications for the interpretation of the findings. First, the statistical power of the analysis was low, evidenced by the wide CIs in pooled analysis and, although all outcomes were in direction of benefit of ExCR, none reached a level of formal statistical significance at the 5% level. Second, and relatedly, we had limited statistical power to detect an association between changes in exercise capacity and the final patient-related outcomes. Last, the results are likely to be subject to selection bias and, therefore, may not be representative of all RCT evidence.
Relevance to clinical practice
The observed improvements in patient exercise capacity and HRQoL with ExCR participation support the class I recommendation of current international clinical guidelines that ExCR should be offered to HF patients.3,13,15 The findings do not endorse limiting ExCR interventions to subgroups of HF patients.
Research recommendations
In spite of the comprehensiveness of this IPD meta-analysis, the findings of this study demonstrate that further evidence is still required to definitively assess the impact of ExCR on mortality and hospitalisation in patients with HFrEF; in particular, to increase the power to examine whether or not the effect of ExCR varies according to patient characteristics. To more reliably quantify the impact of ExCR on clinical outcomes and examine how these effects may vary across HF patients, there is an urgent need for trial investigators to more consistently collect, report and share patient-level data in the future.
Two central aspects of future data collection are a consensus on the definition, collection and reporting of clinical event data, especially hospitalisation, and the capture of data on patient-level adherence to the amount of exercise training during the ExCR intervention period. More generally, the research community should continue to implement policies that encourage primary study authors to make their data sets available, either by depositing in publicly available repositories or by sharing with IPD meta-analysis collaborations when directly requested.81
Given that the vast majority of IPD in this study was from HFrEF patients, future trials including HFpEF patients are needed to assess the effectiveness of ExCR and whether or not there are differential effects of ExCR in this patient group.
Future IPD meta-analyses of RCTs for interventions in HF are needed to confirm the tentative conclusion that VO2peak and 6MWT may be suitable surrogate end points for the final patient-related outcomes. Such future IPD meta-analyses also need to consider individual patient adherence to exercise training.
