The win ratio: a valuable addition to the toolbox of cardiovascular trialists

This editorial refers to ‘Effect of intensive blood pressure and blood glucose control on cardiovascular outcomes driven by reductions in cardiovascular death and nephropathy: win ratio analysis of ADVANCE trial’, by S. Bompoint  et al., https://doi.org/10.1093/ehjcvp/pvag026.

The win ratio was introduced in 2012 as an approach for the analysis of composite endpoints, motivated by cardiovascular trials, which typically compare treatment efficacy through a combination of clinical events, such as cardiovascular death, stroke, and myocardial infarction.¹ In only a short time, the win ratio has come into common use, both for the reanalysis of already published trials and as the primary analysis in ongoing and future trials.^2–5 In an article presented in this issue of the EHJ-CVP, Bompoint and colleagues reanalyse the ADVANCE trial with the win ratio approach.⁶ The purpose of the reanalysis is to ‘better understand the relative contributions of individual outcomes in a composite endpoint’.

The ADVANCE trial was a 2 × 2 factorial trial comparing intensive lowering of blood pressure (perindopril and indapamide) and intensive glucose control (gliclazide modified release) with placebo for patients with type 2 diabetes.^7,8 The primary endpoint was a composite of major macrovascular events (cardiovascular death, myocardial infarction, and stroke) and major microvascular events (nephropathy and retinopathy). The blood pressure intervention reduced the composite endpoint with a relative reduction of 9% (hazard ratio [HR] = 0.91, 95% confidence interval [CI] 0.83–1.00), and the glucose control intervention reduced the composite endpoint with a relative reduction of 10% (HR = 0.90, 95% CI 0.82–0.98). The individual components that contributed the most to the composite endpoint appeared to be cardiovascular death for the blood pressure intervention and nephropathy for the glucose control intervention.

What can the win ratio approach add to this analysis? In the original analysis of ADVANCE, the composite endpoint was analysed in a time-to-first-event fashion, ignoring the severity of the event (and all potential future and recurrent events), thereby treating all components of the primary endpoint as equally important. This can be problematic, for instance, if one or more less important events commonly occur earlier than the more important events. Then, a time-to-first event analysis can be unduly influenced by the less severe events, and serious long-term effects may be ignored. The win ratio approach overcomes this problem by explicitly defining a hierarchy of outcomes based on their clinical importance.¹ This ensures that the more important outcomes are assessed before the less important outcomes. Setting up the hierarchy of outcomes is, however, not always straightforward. In Bompoint et al., it was (i) cardiovascular death, (ii) non-fatal stroke, (iii) non-fatal myocardial infarction, (iv) new or worsening nephropathy, and (v) new or worsening retinopathy.⁶ It is not self-evident that non-fatal stroke should rank above non-fatal myocardial infarction, as the clinical impact depends on the severity and sequelae of the individual event. This nuance is not fully captured by the win ratio approach, which relies on a fixed ordering of outcomes.

In the analysis, the win ratio approach considers all possible pairs of patients consisting of one patient from each treatment group. Within each pair, intervention and control is compared, starting with cardiovascular death and continuing down the hierarchy of outcomes if the two patients are tied. When all pairs have been compared, the win ratio is estimated as the number of wins for the intervention group divided by the number of wins for the control group.

What happened in the reanalysis of ADVANCE? The main results were confirmed (Figure 1). The win ratio for the blood pressure intervention was estimated as 1.11 (95% CI 1.01–1.22), and the win ratio for the glucose control intervention was estimated as 1.10 (95% CI 1.01–1.20). The interpretation of a win ratio of 1.11 is that for any given pair of patients that are not tied for all components of the composite endpoint, the odds that the intervention patient is the ‘winner’ is 1.11. This odds can be expressed as a probability: 1.11/(1.11 + 1) = 0.53. How does this compare with a more familiar metric such as the hazard ratio? While the win ratio is only related to the hazard ratio under specific circumstances (single outcome, proportional hazards, and no censoring),⁹ we can roughly estimate the corresponding hazard ratio as 1/win ratio, giving approximate HRs of 0.90 and 0.91, which are almost equal to those obtained from the original analysis. The confidence intervals are also similar.

Figure 1.

Comparison of the original ADVANCE analysis and the win ratio reanalysis.

What else did the reanalysis reveal? Figures 1 and 2 in Bompoint et al.⁶ show the number (and percentage) of wins, ties, and losses of blood pressure lowering and blood glucose control over placebo for each component of the composite endpoint. For blood pressure lowering, cardiovascular death and nephropathy have win differences (percentage wins-percentage losses) of 0.8% and 0.6%, respectively, and are the biggest contributors to the overall win difference of 1.5%. For blood glucose control, the biggest contributors are cardiovascular death (0.5%), nephropathy (0.8%), and retinopathy (0.4%), with an overall win difference of 1.6%. The two figures also show the order in which the comparisons of events are performed, with details on the number of pairs available and the number of wins, ties, and losses at each step, providing a transparent look into the win ratio approach.

So, did we learn anything new? The overall results and the components that contributed the most to the composite endpoint were the same in the two analyses; however, the win ratio quantified the impact of each component more directly and provided more details. There is also value in confirming a result, which adds to the robustness of the original analysis. Yet, the usefulness of the win ratio might be the greatest in the prospective use of the approach, to design new trials with the win ratio as the main analysis approach. Forcing researchers to make explicit a priori priorities among clinical endpoints would be one of the benefits. The win ratio can also include other types of outcomes than binary or time to event outcomes, including ordinal and continuous outcomes such as quality of life.⁴

It should be mentioned that the win ratio is a debated topic,^10,11 which also has its critics.^11,12 The limitations of the win ratio include an effect measure that may not be intuitively clear, sensitivity to misspecification of the hierarchy of outcomes, a potential for overreliance on outcomes of minor importance, and sample size calculations that are not straightforward.^2,5 Also, the win ratio is no solution to the problem of lack of power in trials with few major events. What is clear, however, is that the win ratio has become a valuable addition to the toolbox of cardiovascular trialists.

Data availability

There are no new data associated with this article.