Hierarchical Endpoints in Cardiology: How Much Treatment Benefit, and From Where

For decades, many cardiovascular clinical trials were designed around mortality, from thrombolysis after myocardial infarction to ACE inhibitors in heart failure. When death rates were higher and treatment options more limited, reducing mortality was the clearest demonstration of clinical benefit. As death rates fell and background therapy improved, trials powered on mortality alone became impractically large, and composite endpoints combining death with nonfatal events became the standard.

These composites are usually analyzed as the time to the first event, an approach whose limitations grow as therapies change. A modern treatment may prolong survival, reduce hospitalizations, improve physical function, and relieve symptoms, while carrying its own safety risks. A time-to-first-event analysis ignores events after the first, weighs a hospitalization like a death, and cannot readily include function or quality of life.¹

A simple case illustrates the problem. One patient has a nonfatal myocardial infarction at 2 months and survives; another dies at 3.5 months without a prior event. A time-to-first-event analysis treats the first patient's outcome as the worst one, because the event came earlier. Recurrent-event analyses recover events after the first, but still count a death and a hospitalization as one event each. What hierarchical endpoints add is the ordering of outcomes by clinical importance, and the combination of clinical events with continuous measures such as function or quality of life.

Cardiovascular research as a testing ground

One of the most important developments in recent years has been the growing adoption of hierarchical composite endpoints analyzed through Generalized Pairwise Comparisons (GPC). The approach traces back to Finkelstein and Schoenfeld,² was generalized to any number and type of prioritized outcomes by Buyse³ and entered cardiology through the win ratio.⁴

Rather than focusing solely on the first event experienced by a patient, GPC compares patients in the experimental and control arms according to a predefined prioritization of outcomes. Outcomes considered most clinically important are evaluated first, with lower-priority outcomes used only when patient pairs cannot be distinguished on higher-priority measures.

This allows multiple clinically relevant outcomes to contribute to a single analysis while preserving their relative importance. Mortality, hospitalization, functional capacity, biomarkers, symptoms, and patient-reported outcomes can all be considered within the same framework.¹

Cardiovascular medicine has become the leading field for applying these methods. A 2026 scoping review identified 92 randomized trials using hierarchical composite primary endpoints, of which 43.5% were conducted in cardiology. Among published studies, GPC was the most commonly used analytical approach.⁵

Importantly, the methodology itself is not inherently patient-centered. Its relevance depends on which outcomes are selected, how they are prioritized, and whether those decisions reflect patient and clinical priorities. The same review found that more than 80% of trials with available documentation did not report how the order of outcomes was established, and that patients were involved in setting it in 2 of 73.⁵

Nevertheless, these prioritized approaches may represent an important step toward evaluating treatment benefit in a way that more closely reflects how patients experience disease.

Lessons from transthyretin amyloid cardiomyopathy

Some of the most influential applications of hierarchical analyses have come from trials in transthyretin amyloid cardiomyopathy. Tafamidis and acoramidis were both approved on primary analyses that compared patient pairs on prioritized outcomes using GPC, with the win ratio reported as the supporting estimate of treatment effect. ATTR-ACT ranked all-cause mortality above the frequency of cardiovascular hospitalization (win ratio 1.70).⁶ ATTRibute-CM extended the list to change in NT-proBNP and change in 6-minute walk distance (win ratio 1.8).⁷

The published pair counts for ATTRibute-CM also illustrate the Net Treatment Benefit. Across all acoramidis–placebo pairs, the acoramidis patient came out ahead in 63.7% and the placebo patient in 35.9%. The difference, 27.8%, is the net treatment benefit (NTB). Because it is a difference, it can be split by outcome, and the parts add up to the total. Every outcome favored acoramidis, and the benefit on clinical events held on its own: restricted to mortality and cardiovascular hospitalization, the prespecified win ratio was 1.5 (95% CI, 1.1 to 2.0).⁷

Win ratio and net treatment benefit

The win ratio is the most commonly reported summary of these analyses, and it helped establish them in cardiovascular medicine.⁵ Both measures come from the same comparisons. Every patient in the experimental arm is compared with every patient in the control arm to determine, outcome by outcome in order of priority, who is better. The win ratio divides how often the experimental patient comes out ahead by how often the control patient does. The Net Treatment Benefit (NTB) subtracts the second from the first. Both test the same null hypothesis, so reporting the NTB requires no change to the design, the endpoint, or the primary test.

They differ in how they treat pairs that cannot be separated. The win ratio sets them aside; the NTB keeps them in the denominator. When ties are frequent, the two can tell different stories, a concern raised by Butler, Stockbridge, and Packer.⁸ PARAGLIDE-HF, which used a four-outcome hierarchical composite as a secondary endpoint, shows this on published data.⁹ In the prespecified subgroup with ejection fraction of 60% or less, the win ratio was 1.46, corresponding to a Net Treatment Benefit of about 15%. When the investigators removed NT-proBNP from the list, the win ratio was essentially unchanged at 1.49, but the win odds, which counts ties, was 1.12 (95% CI, 1.00 to 1.26), equivalent to a Net Treatment Benefit of about 6%.¹⁰ Without the biomarker, most pairs were ties. The win ratio, computed only on the pairs that were separated, could not show it.

Tie handling is not unique to the NTB; the win odds also accounts for ties. What the NTB adds is an absolute scale and additivity. Given the order of outcomes, individual contributions sum to the total, so each outcome's share of the benefit can be read directly, and an outcome favoring the control arm appears as a negative contribution. In the 2026 review, only 6 of 45 published trials reported how much each component weighed on the overall effect, and none reported the win difference, another name for the NTB.⁵ Pocock and colleagues have themselves proposed computing the win difference for each component as a way to quantify absolute benefit.¹¹

Looking ahead

Three conclusions follow. First, hierarchical endpoints analyzed by pairwise comparisons are no longer experimental in cardiology. They underpin two drug approvals, cardiology accounts for the largest share of trials using them as primary endpoints, and their use is growing in other fields.^6,7

Second, the order of outcomes deserves the same justification as any other design choice. Rank order sets priority, not weight: an outcome ranked first but rarely observed decides few pairs, and the result is carried by the outcomes below it. Yet most trials do not report how the order was set, and patients are rarely involved.⁵ The rationale, the expected event rate at each level, and any minimum clinically relevant difference for continuous outcomes belong in the protocol.

Third, results should move beyond the Win Ratio and be reported with the Net Treatment Benefit as the main measure of effect. It comes from the same comparisons and the same test, counts every pair, and adds up across outcomes, so readers can see transparently both how large the benefit is and where it comes from. Like any summary of a hierarchical endpoint, the NTB depends on the outcomes chosen, their order, and follow-up.

Samuel Salvaggio, senior trial design lead, One2Treat

References

1. Verbeeck J, De Backer M, Verwerft J, et al. Generalized Pairwise Comparisons to Assess Treatment Effects: JACC Review Topic of the Week. Journal of the American College of Cardiology. 2023;82(13):1360-1372.
2. Finkelstein DM, Schoenfeld DA. Combining Mortality and Longitudinal Measures in Clinical Trials. Statistics in Medicine. 1999;18(11):1341-1354.
3. Buyse M. Generalized Pairwise Comparisons of Prioritized Outcomes in the Two-Sample Problem. Statistics in Medicine. 2010;29(30):3245-3257.
4. Pocock SJ, Ariti CA, Collier TJ, Wang D. The Win Ratio: A New Approach to the Analysis of Composite Endpoints in Clinical Trials Based on Clinical Priorities. European Heart Journal. 2012;33(2):176-182.
5. Ehrenzeller S, de Jong AJ, Martin Y, et al. Randomized Clinical Trials Using a Hierarchical Composite Primary End Point: A Scoping Review. JAMA Network Open. 2026;9(7):e2625935.
6. Maurer MS, Schwartz JH, Gundapaneni B, et al. Tafamidis Treatment for Patients with Transthyretin Amyloid Cardiomyopathy. New England Journal of Medicine. 2018;379(11):1007-1016.
7. Gillmore JD, Judge DP, Cappelli F, et al. Efficacy and Safety of Acoramidis in Transthyretin Amyloid Cardiomyopathy. New England Journal of Medicine. 2024;390(2):132-142.
8. Butler J, Stockbridge N, Packer M. Win Ratio: A Seductive but Potentially Misleading Method for Evaluating Evidence From Clinical Trials. Circulation. 2024;149:1546-1548.
9. Mentz RJ, Ward JH, Hernandez AF, et al. Angiotensin-Neprilysin Inhibition in Patients With Mildly Reduced or Preserved Ejection Fraction and Worsening Heart Failure. Journal of the American College of Cardiology. 2023;82(1):1-12.
10. Shoji S, Cyr DD, Hernandez AF, et al. Win Ratio Analyses Using a Modified Hierarchical Composite Outcome: Insights From PARAGLIDE-HF. American Heart Journal. 2025;280:70-78.
11. Pocock SJ, Gregson J, Collier TJ, Ferreira JP, Stone GW. The Win Ratio in Cardiology Trials: Lessons Learnt, New Developments, and Wise Future Use. European Heart Journal. 2024;45(44):4684-4699.