Enhancing interpretability for Bayesian basket trial designs by effective sample size.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Bayesian Inference

Background:

• Growing interest in Bayesian methods for information sharing across tumor types in basket trials.
• Existing Bayesian hierarchical models (BHM) lack a clear measure for quantifying information borrowing.
• Difficulty for non-statisticians to understand complex Bayesian designs due to lack of interpretability.

Purpose of the Study:

• To introduce and validate the effective sample size (ESS) as a tool to quantify information borrowing in Bayesian basket trials.
• To demonstrate the utility of ESS in both the design and analysis phases of clinical trials.
• To provide a more interpretable metric for complex Bayesian statistical models.

Main Methods:

• Leveraging the effective sample size (ESS) concept for Bayesian basket trials.
• Proposing ESS-based strategies for information borrowing.
• Utilizing mean squared error (MSE) to derive ESS, balancing bias and variance.
• Reanalyzing the RAGNAR study and conducting simulation studies to demonstrate ESS interpretability.

Main Results:

• ESS effectively quantifies the impact of information borrowing on mean squared error (MSE).
• ESS intuitively characterizes the degree of information borrowing across tumor types.
• ESS demonstrates alignment with type I error rates and statistical power, serving as a valuable analytical complement.

Conclusions:

• Quantifying information borrowing with ESS aids trialists in designing Bayesian basket trials.
• ESS facilitates reasonable evaluation and interpretation of Bayesian analysis results.
• ESS supports sensitivity analyses and helps determine the appropriate amount of information borrowing in trials.