The application of Bayesian statistics in pivotal clinical trials has gained increasing attention, particularly for underpowered confirmatory trials such as those in pediatric and rare disease populations. Bayesian external information borrowing offers a promising approach to enhance statistical efficiency by leveraging historical data. However, regulatory acceptance requires clear evidence on operating characteristics and practical design guidance. This study systematically compares two widely used Bayesian borrowing frameworks—mixture prior (MP) and inflated standard error (ISE)—under two borrowing strategies: treatment response borrowing (TRB) and treatment effect borrowing (TEB). Benchmarking against conventional frequentist methods and Japanese approaches, we evaluate type I error and power across a comprehensive grid of sample sizes, borrowing weights, and decision thresholds. Beyond these quantitative findings, we provide a fully reproducible template that demonstrates how to rigorously explore, visualize, and justify design choices for Bayesian borrowing in regulatory submissions. The paper specifies what simulation outputs, performance metrics, and graphical summaries should be presented to reviewers, and illustrates how to ensure that all reported results align with statistical common sense. In doing so, it serves as both a practical case study and a reusable toolkit to bridge the gap between methodological theory and regulatory practice.
Li et al. (Tue,) studied this question.