Abstract
We propose a general pick-the-winner design framework for two-arm randomized clinical trials with binary and continuous endpoints that meets prespecified type I error rate and power requirements for winner selection. Traditional two-group comparison tests often require sample sizes that are impractical in early phase trials. Existing pick-the-winner designs offer an efficient alternative but generally emphasize the probability of correct selection rather than explicit control of type I and II error rates for winner selection. Our framework comprises four components: pick-the-winner-based hypotheses, winner selection strategies, definitions of type I error rate and power for winner selection, and design calibration. It accommodates flexible winner selection rules, including frequentist, Bayesian, and hybrid rules, and allows existing pick-the-winner principles to be incorporated as specific winner selection strategies. We illustrate the framework using two single-stage winner-selection strategies. In the settings considered, winner selection strategy I (WSI) reduced per-arm sample sizes by approximately 46–51% for binary endpoints and 49–50% for continuous endpoints relative to traditional two-group tests, supporting winner selection when the sample size budget is limited. When larger sample sizes are available, winner selection strategy II (WSII) requires sample sizes comparable to those of traditional two-group comparison tests while providing data evidence that the selected treatment is superior to both the comparator and the standard of care (SOC). The framework can be easily extended to multi-stage or multi-arm trials or trials with joint efficacy and toxicity endpoints.
Disciplines
Biostatistics
Suggested Citation
Lei, Wanni; Chinchilli, Vernon M.; and Zhou, Xi Kathy, "A Pick-the-Winner Design Framework for Two-Arm Randomized Clinical Trials" (October 2026). COBRA Preprint Series. Working Paper 127.
https://biostats.bepress.com/cobra/art127