The win ratio, introduced by S. Pocock in 2012, is an alternative and practical approach for analyzing composite endpoints. It was originally designed to address challenges faced in cardiovascular (CV) trials, but over the years the win ratio has been utilized in multiple therapeutic areas.
A CV trial typically includes multiple clinical efficacy endpoints (e.g., CV death, myocardial infarction, stroke). One analysis approach may be to focus on the time to occurrence of the first event, regardless of which of the events it is. Traditional methods such as the Cox model or log-rank test are limited to capturing only the time to the first event, ignoring subsequent or recurrent events. For example, if a patient experiences a non-fatal stroke, any subsequent event, including death, is disregarded. As a result, early non-fatal events are given more weight than potentially more severe events, or death, that occur later. Additionally, multiple non-fatal events can occur, but only the first one is counted in a time to first event analysis. Despite these limitations, composite endpoints and conventional methods for analyzing them remain commonly used in CV trials.
The win ratio method improves upon conventional methods for analyzing composite endpoints by considering the relative priorities of each component and allowing the inclusion of different types of outcomes. This innovative approach addresses some of the limitations of traditional methods.
Consider a clinical trial that compares a new treatment to a control, using a composite endpoint of all-cause mortality and hospitalization due to heart failure (HFH). A win ratio analysis is conducted as follows. Every patient from the treatment group is paired with every patient from the control group. Within each pair, the components are evaluated in order of clinical importance to determine the ‘winner.’ In this trial, death is deemed more clinically important than HFH, so it is assessed first. If a patient from the treatment group dies later than his/her counterpart in the control group, the patient from the treatment group registers a ‘win.’ Conversely, the patient from the treatment group registers a ‘loss’ if he/she dies sooner than his/her counterpart in the control group. If both patients die on the same day or one patient is censored earlier than the other, the ‘winner’ cannot be determined by death, and HFH, the less clinically important endpoint, is then evaluated in a similar manner to determine the winner of the pair. This process is repeated for all possible patient pairs.
The win ratio is calculated as the ratio of NW to NL, where NW represents the total number of pairwise wins for the treatment group, and NL represents the total number of pairwise losses. The variance of the win ratio can be calculated using the Finkelstein-Schoenfeld method or asymptotic theory. This allows for the estimation of the treatment effect (the win ratio), along with the calculation of a confidence interval and p-value.
The win ratio approach offers distinct advantages compared to other methods for analyzing composite endpoints. One key benefit is that it prioritizes the component events in a meaningful clinical hierarchy. Unlike conventional composite endpoints, which treat each outcome as equally important, the win ratio method organizes the components hierarchically based on their relative clinical importance. Additionally, the win ratio approach can account for all types of trial-related endpoints and outcomes, not just time-to-event data but also broader measures such as quality of life, biomarkers, and laboratory results. Moreover, the win ratio method is conceptually simple and can be easily extended to handle recurrent events (e.g., hospitalizations) without added statistical complexity. The process of counting winners and losers across all pairwise comparisons is straightforward and easy to understand.
Since the introduction of the win ratio in 2012, its popularity has grown in clinical and medical research due to these distinct advantages. It has been applied across various therapeutic areas, including medical device trials, and has also been utilized in numerous COVID-19 research projects. One notable trial employing the win ratio is the ATTR-ACT (NCT01994889) trial, a randomized, double-blind study comparing the efficacy and safety of tafamidis in patients with transthyretin cardiomyopathy. In the trial, 264 patients were assigned to the tafamidis group and 177 to the placebo group. The composite endpoint consisted of a hierarchical assessment of all-cause death and the frequency of CV-related hospitalizations. There were 8,595 wins for tafamidis and 5,071 for placebo, resulting in a highly significant win ratio of 1.70, which is interpretated as the tafamidis group is 1.70 times as likely to have a favorable outcome as the placebo group, with a p-value of 0.0006. The win ratio has proven to be a valuable approach in many clinical settings.
PROMETRIKA biostatisticians collaborate closely with sponsors on study design and can make recommendations for statistical approaches that are fit for purpose to the specific challenges of composite endpoints across different indications. The win ratio has been one approach that we have successfully implemented for our sponsors.
