A generalized calibrated Bayesian hierarchical modeling approach to basket trials with multiple endpoints

Area of Science:

• Biostatistics
• Clinical Trial Design
• Oncology

Background:

• Basket trials accelerate drug development by evaluating treatments across multiple cancer subtypes simultaneously.
• Traditional basket trials often rely on single efficacy endpoints (e.g., tumor response), which are insufficient for complex agents like targeted therapies or immunotherapies.
• Comprehensive risk-benefit assessment requires evaluating multiple endpoints for targeted therapies.

Purpose of the Study:

• To extend calibrated Bayesian hierarchical modeling for monitoring phase II basket trials with multiple endpoints.
• To develop generalized models accommodating diverse endpoint types and information-sharing dependencies.
• To introduce adaptive shrinkage parameters and a calibration approach for model determination.

Main Methods:

• Proposed two generalized Bayesian hierarchical models: a latent variable approach and a multinomial-normal hierarchical model.
• Incorporated shrinkage parameters as functions of subgroup homogeneity statistics.
• Developed a general calibration approach to define functional forms for shrinkage parameters.

Main Results:

• The generalized hierarchical models accommodate various endpoint types and dependence structures.
• Theoretical properties of the proposed models were investigated.
• Simulation studies confirmed that the generalized monitoring approach yields desirable operating characteristics.

Conclusions:

• The proposed Bayesian hierarchical modeling approach effectively monitors phase II basket trials with multiple endpoints.
• These methods provide a more comprehensive assessment of targeted therapies' risk-benefit profiles.
• The generalized approach enhances the efficiency and reliability of basket trial monitoring in oncology drug development.