Making every component count: Using the Shapley value to improve win ratio analysis

Clinical trials often use composite endpoints to combine several clinically relevant outcomes into one treatment comparison. The win ratio is attractive for such endpoints because it compares patients hierarchically according to clinical priority, but the resulting overall ratio does not show how the individual components contribute to the reported treatment effect. Reporting all partial win ratios is informative but can be difficult to interpret because these partial summaries are not an additive decomposition of the overall win ratio. We propose SEWRA (Shapley-Enhanced Win Ratio Analysis), which applies the Shapley value to the log win ratio computed over subsets of components. On this scale, component contributions sum exactly to the overall log win ratio and can also be reported as multiplicative factors on the original win ratio scale. We position SEWRA as an attribution tool that complements, rather than replaces, the overall win ratio and component-specific clinical summaries. Simulations and a recurrent-event case study illustrate how SEWRA separates hierarchy-induced component contributions that are not apparent from partial win ratios alone.