IPM Take
A positive cardiovascular trial does not always mean more patients survived.
That distinction is becoming increasingly important as clinical research moves beyond traditional mortality endpoints toward measures that also capture repeated hospitalisations, physical functioning and quality of life.
In an analysis published by Applied Clinical Trials on 2 October, Samuel Salvaggio examines the growing use of hierarchical composite endpoints, which prioritise outcomes according to clinical importance and evaluate them through pairwise comparisons. Unlike conventional composite endpoints, which often focus on the first event experienced by a patient, hierarchical methods can distinguish between outcomes with very different consequences.
The approach can produce a more informative picture of treatment benefit. But it introduces a question that sponsors, regulators and clinicians cannot afford to overlook.
When a trial reports a favourable win ratio, what actually drove that result?
Evidence from major heart failure and structural heart disease trials shows why this matters. An overall treatment benefit may reflect improvements in symptoms or quality of life rather than a demonstrated reduction in mortality or hospitalisation. Both outcomes can be clinically meaningful, but they are not interchangeable.
The challenge is not simply developing better statistical methods. It is making sure the hierarchy reflects meaningful clinical priorities, patients have a voice in those priorities, and the final result remains understandable to decision-makers.
Executive Summary
Hierarchical composite endpoints are attracting growing interest in cardiovascular trials because conventional time-to-first-event composites can overlook recurrent events and treat outcomes with markedly different clinical importance as components of a single statistical measure.
The win ratio addresses some of these limitations by comparing patients receiving different treatments according to a predefined order of outcomes, typically beginning with death before considering hospitalisation, symptoms or functional status. Each comparison is resolved at the highest-priority outcome that distinguishes the two patients.
In the EMPULSE trial, involving 530 patients hospitalised for acute heart failure, empagliflozin achieved a stratified win ratio of 1.36 compared with placebo. However, subsequent methodological analysis showed that changes in patient-reported symptoms contributed substantially to the overall result.
In the original TRILUMINATE Pivotal analysis, tricuspid valve repair achieved a win ratio of 1.48, with the observed advantage driven primarily by improvement in health status rather than a demonstrated reduction in death or heart failure hospitalisation at one year.
These examples reinforce the importance of reporting Net Treatment Benefit, also called the win difference, alongside the win ratio, and showing the contributions of individual endpoint components.
Why it matters
- HTA bodies: A positive hierarchical composite does not automatically demonstrate improvement in every component outcome. Assessments should examine the clinical relevance of the hierarchy, the absolute differences in wins and losses, and which outcomes actually drive the result.
- Payers: Treatment value depends on the nature and magnitude of benefit. Fewer cardiovascular admissions, improved symptoms and increased survival may each justify investment, but their implications for reimbursement and cost-effectiveness differ.
- Industry / innovation partners: Hierarchical endpoints can capture a broader range of treatment effects, potentially improving the relevance of cardiovascular trial designs. But choices about endpoint ranking, statistical thresholds, missing data and patient involvement need to be justified before results are known.
Cardiovascular trials have a measurement problem.
As treatments improve and mortality rates decline, demonstrating a survival benefit alone often requires larger populations and longer follow-up. To make trials more feasible, researchers increasingly combine several clinically important outcomes into composite endpoints, including cardiovascular death, myocardial infarction, hospitalisation and worsening symptoms.
But conventional composites have an important limitation.
In a typical time-to-first-event analysis, only the first qualifying event determines when a patient reaches the endpoint. A patient who experiences a relatively minor event followed by a major complication may contribute no additional information to the primary analysis after that first event. Repeated hospitalisations can also be overlooked.
When outcomes are not equally important
Hierarchical composite endpoints attempt to address this problem by establishing an explicit order of clinical importance.
Instead of analysing only the first event, researchers compare patients receiving different treatments, beginning with the most important outcome. Death might come first, followed by recurrent hospitalisations and then measures of physical function or quality of life.
If the first comparison produces no winner, the analysis moves to the next endpoint in the hierarchy.
The resulting win ratio compares the number of favourable treatment comparisons with the number favouring the control group. A value above 1 indicates more wins for the treatment under the specified hierarchy.
Importantly, a win ratio of 1.36 does not mean that 36% more patients survived or benefited. It expresses the ratio of winning to losing pairwise comparisons, excluding ties from that ratio. Its interpretation depends on the outcomes selected and how comparisons are resolved.
This distinction becomes particularly important when a trial combines hard clinical events with patient-reported outcomes.
The EMPULSE example: where did the benefit come from?
The EMPULSE trial evaluated empagliflozin against placebo in 530 patients hospitalised for acute heart failure.
Its primary endpoint followed four levels: death, number of heart failure events, time to first heart failure event and change in Kansas City Cardiomyopathy Questionnaire Total Symptom Score (KCCQ-TSS) at 90 days.
The trial demonstrated a favourable result, with a stratified win ratio of 1.36.
However, a subsequent methodological analysis revealed that the largest contribution to the unstratified result came from improvements in KCCQ symptom scores. The overall unstratified win difference was 14.9 percentage points, of which 8.8 percentage points were attributable to the KCCQ component.
The findings did not invalidate the trial’s result.
Rather, they illustrated the importance of distinguishing between a treatment improving survival, reducing clinical events and making patients feel better. These effects can all matter, but their respective contributions should remain visible.
TRILUMINATE raises the same question
The TRILUMINATE Pivotal trial provides another revealing example.
In its original randomised analysis involving 350 patients with severe tricuspid regurgitation, transcatheter edge-to-edge repair achieved a win ratio of 1.48 compared with medical therapy.
The endpoint prioritised death or tricuspid valve surgery, followed by hospitalisation for heart failure and improvement in quality of life.
Subsequent analysis showed that the positive primary result was driven by KCCQ improvement, without evidence of a corresponding advantage in death or hospitalisation at one year. Because the study was not blinded, interpreting patient-reported improvements also required consideration of potential reporting bias.
That illustrates the central issue with hierarchical composites.
A positive result can be both statistically valid and clinically meaningful without demonstrating an effect on mortality.
The problem arises when the headline result obscures that distinction.
Beyond the win ratio: reporting absolute benefit
One solution is to report Net Treatment Benefit (NTB), also referred to as the win difference.
Unlike the win ratio, which divides winning comparisons by losing comparisons, NTB calculates the difference between their proportions across eligible pairwise comparisons.
It can help explain the absolute balance of wins and losses, including the contribution of each component. It should not, however, be interpreted as the proportion of individual patients helped by treatment.
A 2024 European Heart Journal methodological review recommended presenting the win difference alongside the win ratio and identifying the contribution of individual outcomes to the overall result.
Even this approach requires caution. Research published in Statistics in Medicine in May 2026 showed that hierarchical comparisons can produce counterintuitive results under certain conditions, including scenarios in which a treatment improves the marginal probability of each component outcome but produces an unfavourable overall win ratio.
That reinforces the need for component-level reporting, transparent statistical assumptions and sensitivity analyses.
Who decides what matters most?
Another unresolved question concerns who determines the hierarchy.
Clinical researchers traditionally prioritise mortality and major cardiovascular events. That is often appropriate, but patients may place different values on avoiding hospitalisation, preserving independence, reducing symptoms or maintaining daily functioning.
Research into cardiovascular endpoint preferences has shown that patients distinguish substantially between different clinical outcomes and may value involvement in selecting the endpoints used in trials.
More recent methodological work has also explored incorporating individual patient preferences into composite outcome assessments.
The order of endpoints is therefore not merely a statistical decision. It reflects assumptions about the relative importance of different outcomes.
Making those assumptions explicit is essential if hierarchical composites are to support genuinely patient-centred evidence generation.
A more demanding standard for cardiovascular evidence
Hierarchical endpoints represent an important development in cardiovascular trial methodology.
They can capture treatment effects that conventional first-event analyses miss and bring together outcomes that more closely reflect how patients experience disease.
But they are not inherently superior for every trial. Their value depends on the scientific justification for the hierarchy, how repeated events and missing information are handled, and whether the resulting statistics remain clinically interpretable.
For regulators, HTA bodies, clinicians and patients, the next standard should go beyond establishing that a trial achieved statistical significance.
The critical question is not simply whether a treatment won, but what it won on, by how much and whether those gains matter to patients.
