Assessing the hierarchical beta-binomial model as a basic information sharing tool in basket trials

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Current basket trial statistical methods predominantly use Bayesian hierarchical models with logit-transformed response rates.
• This logit-transformation complicates model interpretation and application in clinical settings.

Purpose of the Study:

• To investigate the hierarchical beta-binomial model as a practicable alternative to logit-transformed models for information sharing in basket trials.
• To systematically compare the two modeling approaches for their suitability in clinical trial design.

Main Methods:

• Systematic comparison of distributional assumptions and Bayesian behavior of beta-binomial and logit-transformed models.
• Development of a calibration procedure for setting comparable priors across models.
• Derivation of an evaluation measure for assessing the sharing property in simulation studies.

Main Results:

• The hierarchical beta-binomial model demonstrates a feasible approach for information sharing in basket trials.
• The beta-binomial model offers improved interpretability compared to logit-transformed models.
• A novel evaluation measure was developed to assess the sharing property.

Conclusions:

• The hierarchical beta-binomial model is a viable and less complex alternative for statistical information sharing in basket trials.
• This approach enhances model understanding for stakeholders in clinical research.
• Further adoption of the beta-binomial model could streamline basket trial methodologies.