A Bayesian decision-theoretic approach to multiple testing in basket trials.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Bayesian Statistics

Background:

• Basket trials evaluate single treatments across diverse patient groups, presenting significant multiple testing challenges.
• Existing methods for analyzing basket trials can be computationally intensive and complex.
• There is a need for efficient and adaptive statistical approaches in basket trial analysis.

Purpose of the Study:

• To propose a novel Bayesian decision-theoretic framework for analyzing basket trials.
• To develop a method that adaptively penalizes false positives and false negatives.
• To offer a computationally efficient and generalizable approach for identifying promising treatment effects in basket trials.

Main Methods:

• A Bayesian decision-theoretic approach using a family of adaptive loss functions.
• Independent estimation of treatment effects across baskets to reduce computational burden.
• Derivation of an optimal Bayes decision rule by minimizing posterior expected loss.

Main Results:

• The proposed method adaptively adjusts for multiple testing, enhancing the detection of promising baskets.
• Treatment effect estimation is independent across baskets, allowing for flexible trial designs and various endpoints.
• Tuning parameters enable control over the degree of information borrowing and decision conservativeness.
• Simulation studies show competitive performance against existing methods.

Conclusions:

• The novel Bayesian approach provides an efficient and flexible solution for multiple testing problems in basket trials.
• This method facilitates the identification of promising treatments while managing statistical complexities.
• The approach is practical and demonstrates utility in real-world applications, such as the vemurafenib basket trial.