Interpreting the Win Ratio in Hierarchical Composite Endpoints: Challenges, Limitations, and Perspectives with Examples from Chronic Kidney Disease Trials
Henrik F. Thomsen, Samvel B. Gasparyan, Julie F. Furberg, Christoph Tasto, Nicole Rethemeier, Patrick Schloemer, Tuo Wang, Niels Jongs, Yu Du, Tom Greene
[stat.ME]
Win statistics based methods have gained traction as a method for analyzing Hierarchical Composite Endpoints (HCEs) in randomized clinical trials, particularly in cardiovascular and kidney disease research. HCEs offer several key advantages, including increased statistical power, the mitigation of competing risks, and hierarchical weighting of clinical importance for different outcome components. While, as summary measures, the win ratio (WR) along with the Net Benefit (NB) and the Win Odds (WO) provide a structured approach to analyzing HCEs, several concerns regarding their interpretability remain. In this paper, we explore critical aspects of WR interpretation that have received limited attention. Specifically, we discuss the challenge of defining an appropriate estimand in the context of HCEs using the WR, the difficulties in formulating a relevant causal question underlying the WR, and the dependency of the WR on the variance of its components, which complicates its role as an effect measure. Additionally, we highlight the non-collapsibility of the WR, akin to hazard and odds ratios, further complicating its interpretation. While the WR remains a valuable tool in clinical trials, its inherent limitations must be acknowledged to ensure its appropriate application and interpretation in regulatory and clinical decision-making.