winratio

Win ratio is a statistical measure used in the analysis of randomized clinical trials, particularly when endpoints are composite and can be ordered by clinical importance. The method compares every treated patient with every control patient and tallies results as wins, losses, or ties according to a pre-specified hierarchical ordering of outcomes. For each treated-control pair, the patient with the more favorable outcome is a win for the treatment; the patient with the less favorable outcome is a loss; equal outcomes are ties. The win ratio is the total number of wins divided by the total number of losses.

Interpretation and computation follow the hierarchical outcome framework. A win ratio greater than 1 indicates that, across all cross-pairs, treated patients are more likely to experience a better outcome than control patients. A ratio less than 1 favors control. Confidence intervals and hypothesis tests are typically obtained by permutation or bootstrap methods; time-to-event data and censoring can be incorporated by designing the outcome hierarchy to reflect survival time or event status. The approach yields a single summary statistic while accommodating multiple or ordinal outcomes.

Applications and scope. Win ratio is commonly applied in cardiovascular trials with composite endpoints but is suitable for other diseases where outcomes can be ordered by importance. It provides a nonparametric alternative to hazard ratios and does not require proportional hazards. The method emphasizes clinical relevance by using a predefined hierarchy of endpoints, allowing investigators to compare overall treatment effect across a spectrum of outcomes.

Advantages and limitations. Advantages include intuitive interpretation, use of all randomized participants, and compatibility with complex or ordinal endpoints. Limitations include the need to prespecify a clinically meaningful hierarchy, potential reduced power with many similar outcomes, and less familiarity among clinicians compared with traditional effect measures.