The Win Ratio at the Design Stage of Clinical Trials

TL;DR

This paper evaluates the statistical power of the win ratio in clinical trial design, demonstrating its advantages over traditional methods and providing a new formula for sample size estimation to enhance trial efficiency.

Contribution

It introduces a novel formula for estimating sample size for win ratio analysis and compares its power to traditional methods through case studies and simulations.

Findings

Win ratio outperforms single-endpoint analyses with moderate effects on lower outcomes.

Win ratio can have up to 50% higher power than time-to-first-event analysis.

Effectiveness of win ratio depends on outcome hierarchy and effect sizes.

Abstract

The win ratio offers a flexible approach to incorporate the hierarchy of clinical outcomes into the analysis of a composite endpoint, enabling simultaneous consideration of multiple outcome types, unlike traditional time-to-first-event (TTFE) analysis or focus on a single outcome. We examined the statistical power of the win ratio compared to single-endpoint analyses and TTFE analysis through a case study and simulation studies. Furthermore, we provide a novel formula to estimate the required sample size for win ratio analysis based on the desired width of its confidence interval, facilitating precision-based trial design. Our results indicate that win ratio analysis generally outperforms single-endpoint analyses when treatment effects on lower-ranked outcomes are moderate compared to those on higher-ranked outcomes. The win ratio can provide greater power than TTFE analysis, especially when the effect on the highest-ranked outcome is substantial, reaching increases in power up to 50\%. Further, even for moderate treatment effects on the highest-ranked outcome, win ratio analysis achieved higher power. Future work should expand our simulations to additional data-generating mechanisms and outcome types, particularly ordinal outcomes, where the win ratio provides an alternative to existing non-parametric and parametric methods. Our findings highlight the potential of the win ratio to improve statistical efficiency in pharmaceutical and other clinical trial designs using composite endpoints, particularly when no single component dominates the treatment effect. However, when continuous outcomes occupy the top of the hierarchy, these tend to drive overall analysis, sidelining contributions of lower-ranked outcomes and limiting benefits of hierarchical win ratio analysis.