A Bayesian Basket Trial Design Using Local Power Prior

Abstract: In recent years, basket trials, which enable the evaluation of an experimental therapy across multiple tumor types within a single protocol, have gained prominence in early-phase oncology development. Unlike traditional trials, where each tumor type is evaluated separately with limited sample size, basket trials offer the advantage of borrowing information across various tumor types. However, a key challenge in designing basket trials lies in dynamically determining the extent of information borrowing across tumor types to enhance statistical power while maintaining an acceptable type I error rate. In this paper, we propose a local power prior framework that includes a 3-component borrowing mechanism with explicit model interpretation. Unlike many existing Bayesian methods that require Markov Chain Monte Carlo (MCMC) sampling, the proposed framework offers a closed-form solution, eliminating the time-consuming nature of MCMC in large-scale simulations for evaluating operating characteristics. Extensive simulations have been conducted and demonstrated a good performance of the proposal method comparable to the other complex methods. The significantly shortened computation time further underscores the practical utility in the context of basket trials.