Heart failure (HF) is a clinical syndrome characterised by high morbidity and mortality, with a median incidence of 3.20 cases per 1000 person-years and a prevalence of 1–2% of adults in Europe.1–3 HF also significantly impairs physical function and quality of life, with an annual healthcare costs reaching $284.17 billion globally.4 Traditionally it has been classified by left ventricle ejection fraction (LVEF), with worse pharmacological response in patients with >50% LVEF.5 However, differences have also been observed in mortality, sex, cardiac remodelling and systemic biomarkers.6

With an ageing population, HF with preserved ejection fraction (HFpEF) has become the most common type of HF, and its incidence is expected to increase. Its typical multimorbid nature and lack of established pharmacological treatments present significant challenges for healthcare systems.5 Current clinical Practice Guidelines.5,7 recommend focusing on comorbidity management in patients with HFpEF. Among non-pharmacological interventions, exercise programmes effectively control cardiovascular risk factors and key comorbidities such as obesity, diabetes, obstructive pulmonary disease, anaemia.5,7,8 and atrial fibrillation,9,10 making them a valuable adjunctive option.

Previous reviews.11–13 have analysed the comparative effectivity of different exercise modalities on HFpEF, showing positive results especially for peak VO2, which is correlated with fewer recurrent hospitalisations.14 However, some exercise modalities studied were not included in the assessment of effectiveness across all exercise performance, echocardiographic parameters, and quality of life (QoL) variables. In addition, they all included studies in which the baseline LVEF mean with its standard deviation exceeded the 50% threshold, which could imply a selection bias, affecting the implementation of the results.

In light of the prognostic differences among phenotypes, the studied interventions, and the growing body of scientific literature, we conducted a network meta-analysis (NMA). The objective was to analyse the effectiveness of different exercise modalities on exercise performance, echocardiographic parameters and quality of life in HFpEF patients. An additional objective was to compare the exercise modalities to analyse which ones were superior for each outcome.

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA-P) extension for network meta-analyses.15 and the Cochrane Handbook for Systematic Reviews of Interventions.16 The protocol of this review was registered in PROSPERO. Literature search, studies selection, data extraction, the risk of bias and quality of the evidence assessment were conducted by two independent authors and discrepancies were solved by a third author.

A total of 12 studies were included in the quantitative analysis.23–34 The selection process is summarised in the PRISMA flow diagram in Supplementary material F.

The included studies were carried out between 2011 and 2021, reporting data on 602 patients, with a mean age of 64.23 years, of whom 397 were women (65.94%). The type of exercise interventions found were HIIT,23,26,31 IMT,32,33 MIAT,23–26,28,30,31 CT27,29,34 and Tai Chi.34 The weekly frequency of the treatment was 3 days for HIIT, 2–5 days for MIAT, 2–7 days for CT, 7 days for IMT, and 2 days for Tai Chi. Other characteristics of the included articles are summarised in Table 1.

With respect to exercise reporting, CERT complementation ranged between 26 and 74%. Only one study reported >60% (13 items) of the checklist. Fig. 1 shows the compliance for each item per individual study.

Fig. 1.

CERT evaluation for each study.

Colored cells represent reported items. Darker colored cells represent a higher percentage of CERT completion.

Exercise performance variables

Compared to UC: CT showed an effect on peak VO2, but only in the Frequentist model for VO2 at VT; IMT showed an effect on peak VO2, but only in the Frequentist model for VE/VCO2 slope, RER and 6MWT; MIAT showed an effect on peak CPET time, PPO, and 6MWT, but only in the Frequentist approach for peak VO2. Tai Chi showed an effect on peak VO2. HIIT only reached statistical significance in the Frequentist model for peak VO2, VE/VCO2 slope and CPET. No intervention reached had a significant effect on peak HR. According to the rank table, CT was most probable first-line treatment for VO2 at VT (P-score=0.77) and PPO (P-score=0.75), HIIT was the most probable first-line treatment for CPET time (P-score=0.81), IMT was the most probable first-line treatment for VE/VCO2 slope (P-score=0.99), RER (P-score=0.96) and 6MWT (P-score=0.80), and Tai Chi was the most probable first-line treatment for VO2 peak (P-score=0.92). Regarding between-intervention comparisons, HIIT was superior to IMT for VE/VCO2 slope. Only the Frequentist model supported that: for peak VO2 CT was superior to MIAT and to HIIT, and Tai Chi was superior to MIAT. For VE/VCO2 slope, MIAT was superior to IMT. Finally, IMT was superior to MIAT for RER. Table 2 shows the effect estimates across all comparisons. Fig. 2 shows the forest plot for each exercise modality on peak VO2. Forest plots for the rest of variables can be found in Supplementary material I.

Table 2.

Effect estimates for every possible treatment comparison.

The upper part of each table contains the effect estimate for direct comparisons, the lower part contains the effect estimate from the Network-Meta-Analysis. For each comparison, the upper cell presents the Frequentist estimate and the lower cell the Bayesian estimate. Green cells represent statistical significance for that specific comparison. A: late mitral flow, CT: combined training, CPET: cardiopulmonary exercise testing, E: early mitral flow, e´: early diastolic velocity of the mitral annulus, HR: Heart rate, HIIT: high intensity interval training, IMT: inspiratory muscle training, LAVI: left atrial volume index, LVEF: left ventricle ejection fraction, MIAT: moderate intensity aerobic training, MLHFQ: Minnesota living with heart failure questionnaire, RER: respiratory exchange ratio, RT: resistance training, VE: minute ventilation, VCO2: carbon dioxide production, VO2 peak: peak oxygen uptake, VT1: oxygen uptake at first ventilatory threshold, 6MWT: 6-minute walking test.

Fig. 2.

Frequentist and Bayesian Forest plots for peak VO2.

Upper Forest plot represents Frequentist NMA, lower Forest represents Bayesian NMA. HIIT: high-intensity aerobic training, IMT: inspiratory muscle training, MIAT: moderate-intensity aerobic training.

Heterogeneity was found for peak VO2, VO2 at VT, HR peak, and especially for 6MWT. Meanwhile, heterogeneity was lower for RER, PPO and VE/VCO2. Inconsistency was not significant for peak VO2 and VE/VCO2 and could not be calculated for the rest of the variables (Supplementary material J). Net and node-splitting functions revealed no inconsistency for VO₂ peak and the VE/VCO₂ slope; for the remaining variables, inconsistency could not be assessed due to the absence of loops. (Supplementary material K).

The leverage versus residual deviance plots indicated good model fit for all variables, except for VO₂ peak, where one study (16) was identified as contributing to a poorer model fit (Supplementary material L). Sensitivity analysis for peak VO2 showed that, when compared to UC: CT, HIIT, IMT, MIAT and Tai Chi were still effective (Supplementary material M). Heterogeneity significantly decreased and inconsistency remained non-significant (Supplementary material J).

Meta-regression for assessing if age, sex differences or weeks of treatment could influence the magnitude of the estimates was non-significant for peak VO2.

Comparison-adjusted funnel plot and Egger test suggested there is no publication bias. (Supplementary material N).

Our results align with previous research for the study of peak VO2, as we found the effectiveness of CT,11,12 HIIT,11–13 MIAT.11,12 and IMT.12 Contrary to Gomes-Neto,12 we observed an effect of Tai Chi on this outcome, possibly due to the inclusion of only one study meeting the LVEF ≥ 50% criterion, resulting in low precision of the observed estimate.29 For CPET time and PPO, we found MIAT to be effective, consistent with the findings of the only previous NMA that addressed these variables.11 Conversely, we did not observe any effect on peak HR, nor HIIT across the outcomes. This could be due to our inclusion criteria, different studies and more comparators included in this NMA. For 6MWT, our results coincide with Cavero-Redondo.11 for the effectiveness of MIAT, with the novel finding of an effect on IMT on this outcome. Regarding RER, our NMA is the first to obtain an effect of IMT. Only one previous systematic review addressed this outcome, finding no differences when comparing HIIT to MIAT, which is also in line with our findings.35

Concerning the echocardiographic variables, our findings are in line with Cavero-Redondo.11 for the effectiveness of HIIT on E/é ratio and the absence of statistical significance on E/A; however we did not find any effect on é. Surprisingly, we found a negative effect for IMT on LVEF. As only one study reported information for this exercise modality, this should be further studied in future trials.32 Regarding LAVI, this is the first time an NMA has analysed it. Although we did not find a significant effect, consistent with a previous RS comparing HIIT and MIAT,35 there is a tendency for HIIT to reduce LAVI; future investigations may corroborate or refute this. The lack of an observed effect on é, E and A, may be attributed to differences in the included studies, low statistical power to detect differences, or the absence of it. The better results for the exercise performance variables in contrast to the echocardiographic parameters could be explained as myocardial muscle has a limited capacity for improvement when compared to skeletal muscle,36 furthermore, it seems that improvements in exercise performance are driven primarily by peripheral (arteriovenous O2 difference) rather than central adaptations.37,38 Longer interventions may be required to achieve better results for these outcomes.

Regarding QoL, our results contrast with those of Walters,13 as we found MIAT but not HIIT to be effective on MLHFQ. The more conservative nature of our findings compared to previous reviews may be attributed to: (1) the inclusion of only studies where the intervention was exercise in isolation, (2) the exclusion of articles involving HF patients with LVEF < 50%, and (3) the exclusion of duplicate studies with the same NCT, retaining only the most recent one, (19) thus lowering the precision of the effect estimates and not reaching statistical significance.

Regarding transitivity assumption, we did not find age, sex or duration of the intervention to influence on some estimates. In addition, patients ‘characteristics and exclusion criteria were similarly distributed across studies (Table 1), except for one study reporting lower BMI,29 other including patients with cancer.34 and two including patients with NYHA class IV.32,33 Furthermore, for several comparisons we could not perform inconsistency tests to assess incoherence, or there was substantial heterogeneity (Supplementary material J). Thus, we decided to downgrade for indirectness and inconsistency in some of the comparisons.21,39,40

Current guidelines.5,7 recommend exercise to improve exercise capacity, functional status, QoL and reduce hospitalisations with a grade IA recommendation for HF patients in general, although there are no exact recommendations regarding which kind of exercise modality is the best for HFpEF.41 The main finding of this NMA was the effectiveness of CT, IMT, MIAT, and Tai Chi, particularly in improving exercise performance, as its impairment is a hallmark of this population. Our findings support the implementation of MIAT in clinical practice, as it appears to have the greatest overall impact across the evaluated outcomes, while incorporating resistance training yields superior improvements in peak VO2 (Table 2). From a practical point of view, progressing from lower to higher intensities could have an impact across the continuum of metabolic adaptations such as mitochondrial oxidative function.42–45 and neurohormonal effects.46,47 Adding IMT seems plausible to further improve these results.

To our knowledge, this is the first review to present results from both models on the comparative effectiveness of all exercise interventions in isolation across the spectrum of exercise capacity, echocardiography and QoL. Notably, this is the first NMA providing information about the effect of CT on VO2 at VT, HIIT on VE/VCO2 slope, IMT on 6MWT, RER and LVEF, MIAT on MLHFQ and Tai Chi for peak VO2. Our findings offer a guide for selecting specific exercise interventions to target each parameter in HFpEF. Additionally, this review presents a characterisation of the FITT and exercise monitoring parameters summarised in the in Table 1, which are also relevant for clinicians. Another insight is that exercise interventions in RCTs including HFpEF patients are moderately reported according to CERT, hence limiting the replicability of the interventions. This underscores the need for scientific journals to mandate more detailed reporting of exercise interventions in future RCTs, particularly for the less frequently reported items outlined in Fig. 1.

Although our findings have clinical relevance, they are not without limitations. First, the quality of the evidence was very low for all the outcomes evaluated, partially due to the high risk of bias for almost all the included studies. Second, the heterogeneity observed in some of the estimates limits the applicability of the findings. Third, the low number of studies included in the NMA impacts the precision of the results, reduces the power for some analyses, and prevents the testing of overall consistency within the networks. Fourth, we were only able to perform meta-regressions for a limited number of variables. Although no significant effects modifiers were identified, it may have impacted the transitivity assumption. Fifth, the large number of variables analysed for exercise performance and echocardiography parameters implies more likelihood of false positives, which is why we highly recommend taking into account the convergence of both approaches. Sixth, in one study we had to estimate the mean and SD.32 Seventh, although we intended to minimise the presence of patients with an EF < 50% in the analysis, we cannot assume their absence because we didn’t have individual data.

Considering both approaches very low quality of the evidence indicates an effect of CT, IMT, and Tai Chi on peak VO2, as well as an impact of MIAT on other exercise performance parameters. Furthermore, CT is superior to MIAT on peak VO2, and HIIT is superior to IMT on VE/VCO2 slope. Only the Frequentist model showed an effect for CT, HIIT and MIAT on exercise performance and echocardiography parameters, and an effect of MIAT on quality of life. Only the Frequentist model supported the superiority of CT over HIIT for peak VO2 and Tai Chi over MIAT; for VE/VCO2 slope, MIAT is superior to IMT, and for RER IMT is superior to MIAT. Future RCTs should aim to improve methodological quality and include bigger sample sizes (especially for HIIT and Tai-Chi) to enhance precision and overall quality of the evidence.

Funding for open access charge: Universidad de Málaga / CBUA

Supplementary material A. Search strategy in the bibliographical search.

Supplementary material B. Consensus on Exercise Reporting Template description.

Supplementary material C. Risk of bias assessment description.

Supplementary material D. Grading recommendations Assessment, Development, and Evaluation (GRADE) tool assessment description.

Supplementary material E. Sensitivity analysis, meta-regression and publication bias analyses.

Supplementary material F. PRISMA flow diagram of the bibliographical search.

Supplementary material G. Risk of bias assessment.

Supplementary material H. GRADE assessment for each Network Meta-Analysis.

Supplementary material I. Frequentist and Bayesian Forest plots for each Network Meta-Analysis.

Supplementary material J. Heterogeneity and Inconsistency for each Network Meta-Analysis

Supplementary material K. Network graph for various Network Meta-Analysis

Supplementary material L. Leverage versus residual deviance plots for each Network Meta-Analysis.

Supplementary material M. Sensitivity analysis for peak VO2.

Supplementary material N. Comparison adjusted funnel plots for each Network Meta-Analysis.

Concept and design (AICV, JMHJ), acquisition (JMHJ, IJFA), analysis or interpretation of the data (JMHJ, IVJFA, CGC), drafting of the manuscript (JMHJ, AICV, IJFA, CGC), critical revision of the manuscript (IJFA, AICV), statistical analysis (JMHJ), and supervision (AICV).