Designing clinical trials in heart failure with preserved ejection fraction: quality over quantity?

This article refers to ‘Defining changes in physical limitation from the patient perspective: insights from the VITALITY-HFpEF randomized trial’ by J. Butler et al., published in this issue on pages 843–850. Guideline-directed medical therapy for heart failure with reduced ejection fraction (HFrEF) consists of four classes of medications proven in randomised controlled trials to reduce cardiovascular mortality.1 Widespread implementation of these therapies over time has reduced event rates, such that, to improve statistical power, study designs now routinely adopt composite endpoints, usually comprising of cardiovascular mortality and adjudicated heart failure hospitalisation. However, time-to-event analyses of death and hospitalisation have not always been the principal focus of heart failure trials, with earlier work also recognising heart failure as a disease of physical limitation, commonly employing objective markers of exercise capacity as trial endpoints.2 Sadly, such an approach was greatly confounded by being limited to patients who could complete the study protocol, whilst those who became too unwell or died were excluded, limiting generalisability and potentially exaggerating treatment effects. Additionally, participants were often required to achieve a highly reproducible pre-treatment value as variation between tests was often greater than the effect of the study drug, reducing sample sizes and the generalisability of findings.3 Subsequently, the discovery that therapies which improved acute haemodynamics and exercise capacity in the short term could increase the risk of death,4 whilst therapies which had neutral effects on exercise capacity could dramatically improve survival5 resulted in a moral imperative that safety be a primary outcome of cardiovascular trials. For heart failure with preserved ejection fraction (HFpEF) no therapies are yet proven to improve survival and even the positive result of EMPEROR-Preserved (Empagliflozin Outcome Trial in Patients with Chronic Heart Failure with Preserved Ejection Fraction) was driven primarily by reductions in first hospitalisations for heart failure (hazard ratio [HR] 0.71; 95% confidence interval [CI] 0.60–0.83)6 whilst all-cause mortality was unaffected (HR 1.00, 95% CI 0.87–1.15). Whether a pharmacological therapy targeting only the perturbations of the heart failure syndrome could reduce the risk of all-cause mortality in this heterogeneous disease is questionable. In EMPEROR-Preserved around half of deaths were due to non-cardiovascular causes and during a median follow-up of 26.2 months, only 1.5% died of heart failure. This was despite enrolling a symptomatic population with elevated natriuretic peptides, many of whom had been hospitalised for heart failure within the previous 12 months (23%) or had left ventricular ejection fraction <50% (33%).6 Given these challenges, the scope of clinical trials in HFpEF has broadened to incorporate patient-reported outcome measures including health-related quality of life, which is commonly impaired in chronic heart failure and for many patients every bit as important as prognosis.7 Health status has usually been reported separately from ‘hard’ endpoints (Table 1). For example, in EMPORER-Preserved, in which receipt of empagliflozin was associated with modest improvements in mean Kansas City Cardiomyopathy Questionnaire (KCCQ) overall summary score and greater odds of improvement by ≥5 points, this intriguing result was an exploratory outcome analysed separately from death and hospitlisation.8 Exercise capacity (VO2peak) Echocardiography (E/e') Exploratory endpoint (KCCQ-23) Secondary endpoint (KCCQ-TSS) Primary and secondary endpoints (KCCQ-TSS) Time-to-first and total HF events Quality of life (KCCQ-TSS) Estimating change in health status appears simple but becomes challenging where there are missing data, particularly relevant for conditions such as chronic heart failure where censorship due to death is common. Quality of life assessments are only available for patients who survived the study protocol and were well enough to complete the questionnaires, possibly exaggerating treatment effect estimates since quality of life and survival are closely correlated.9 To address this limitation, analyses which incorporate health status and adverse outcomes into a single metric have been proposed including hierarchical clinical composite scores10 as utilised recently in EMPULSE11 (EMPagliflozin 10 mg Compared to Placebo, Initiated in Patients Hospitalised for acUte Heart failUre Who Had Been Stabilised). This approach tackles the issue of mixing endpoints with different clinical (death is more important than hospitalisation) and statistical (earlier deaths are worse than later deaths) weights, incorporates all the information likely to be relevant to patients, and creates a simple overall result determined by ‘win-ratio’. An alternative approach is joint modelling which has the advantage of direct utilisation of the raw data from the trial offering the opportunity of considering the dependencies between time-to-event data and repeated measurements of longitudinal data such as health status.12 Joint models have been applied successfully in the fields of human immunodeficiency virus and oncology, and explored post hoc in the setting of transcatheter aortic valve implantation.13 Regardless of the approach taken, results of clinical trials need to be readily understood by non-statisticians, whilst ensuring observed changes in quality of life are ‘real’ and meaningful to patients. Our current understanding of minimal importance differences in health-related quality of life has been largely derived from studies of HFrEF and whether these thresholds are relevant for HFpEF is unknown. In the current issue of the Journal, Butler and colleagues address this important question leveraging data from VITALITY-HFpEF,14 a randomised trial assessing the effect of vericiguat on quality of life in patients with chronic heart failure and left ventricular ejection fraction ≥45%, which uniquely used change in KCCQ physical limitation score (PLS) as the primary endpoint.15 The authors should be congratulated not only for contributing important data which will help the design and interpretation of future clinical trials, but also for having the foresight to include the present analysis as a pre-specified aim of the trial. The KCCQ-PLS, a measure of patient-reported impairments in physical limitations due to heart failure, was assessed at baseline, 12 and 24 weeks. Firstly, meaningful within-patient changes in KCCQ-PLS at 12 weeks were estimated using anchor-based approaches, determined for each Patient Global Impression of Change (PGIC) category (‘much worse’, ‘a little worse’, ‘the same’, ‘a little better’, ‘much better’), with additional analyses using change in Patient Global Impression of Severity (PGIS) scores (‘no limitations’, ‘mild’, ‘moderate’, ‘severe’, ‘very severe’). Secondly, distribution-based approaches were used to support these observations and determine the proportion of patients who experienced different levels of change in KCCQ-PLS across PGIC categories. With an anchor-based approach, minimally important differences for improvement of KCCQ-PLS were ≥8.33 points (when transformed to a 100-point scale) corresponding to an improvement of ≥2 points in PGIC, whereas a meaningful deterioration corresponded to ≥4.17 points, or any worsening in PGIC. Distribution-based approaches supported these observations with minimal meaningful within-patient changes in KCCQ-PLS corresponding to 12.2 points for 0.5 standard deviations at baseline (corresponding to a ‘moderately large change’). These observations are plausible, being well-aligned to the prior estimates of a 5-point improvement in KCCQ-PLS raw score (which approximates to 8.33 points on the transformed 0–100 scale) being clinically important in HFrEF. Some limitations should be noted. Firstly, the Spearman's rank correlation coefficient between the PGIC and change in KCCQ-PLS at 12 weeks was 0.28, slightly below the generally accepted threshold of 0.30, suggesting PGIC may not be a reliable anchor upon which to quantify change in KCCQ-PLS. Secondly, the sensitivity and specificity for the proposed cut-point were only 0.49 and 0.64, respectively, suggesting around half of patients who improved may be miscategorised using a threshold of ≥8.33 points. Thirdly, that any decline in physical functioning was considered meaningful to patients is interesting, but whether this remains the case outside the artificial setting of a randomised controlled trial remains to be seen. Mean KCCQ-PLS improved for both study groups, with most patients reporting at least some improvement in PGIC. This observation may be the result of a placebo effect, modification of behaviour as the result of being observed (Hawthorne effect) or simply due to intensification of background therapies consequent to being enrolled in a clinical trial with more frequent clinical assessment. This finding is not unique to VITALITY-HFpEF, with improvements in health status also observed in patients randomised to placebo in EMPEROR-Preserved (although improvements were more marked in those randomised to empagliflozin).8 Finally, changes in KCCQ-PLS, PGIC and PGIS are by their very nature, somewhat subjective, and although are ‘real’ to patients, we lack 6-min walk distance or another objective measure of workload to determine whether these changes correlate to prognostically relevant measures of functional capacity. Accepting these limitations, these data should help inform the design and interpretation of future clinical trials in HFpEF; firstly, by allowing researchers to compare between groups the proportions who improved, were unchanged or deteriorated, and secondly, by facilitating estimates of the number needed to treat to achieve meaningful improvements in quality of life. The past two decades have taught us that HFpEF is a heterogeneous condition, for which it is unlikely any single pharmacological therapy will substantially improve prognosis. Perhaps to make progress we should shift our focus to what matters most to patients and ensure that symptoms and quality of life are mandated trial endpoints. Conflict of interest: S.S. declares no competing interests.