Power considerations for the win ratio: A rank-based simulation approach

The win ratio is an innovative statistical method for evaluating efficacy in clinical trials. The underlying distributions of outcomes, along with potential censoring, ties, and correlations, add complexity to specifying a win ratio for study design purposes. As successful study planning hinges on thorough consideration of sample size and statistical power, we developed a flexible approach to support the use of the win ratio in clinical trials.

We develop a simulation-based approach for study design considerations, relying on the relationship between the win ratio and the rank distribution. We demonstrate that, in the absence of censoring and ties, the strategy samples according to the win ratio specified under the alternative hypothesis. We incorporate administrative censoring, ties, and correlations and conduct a simulation study to evaluate the method in terms of type I error and power. We generate data under specific parametric distributional assumptions and summarize statistical power using sample sizes determined by the rank-based simulation.

Results indicate the proposed approach preserves type I error, samples under the assumed win ratio, and provides informative guidance on statistical power for given sample size. The winratiopss R package provides functionality to implement the proposed approach.

The simulation strategy offers a novel and flexible approach to inform trial design involving win ratio analysis.