Bayesian adaptive basket trial design using model averaging.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Oncology

Background:

• Existing oncology basket trial designs often assume uniform drug activity across all cancer types.
• Investigational products may exhibit varying efficacy in different cancer subsets.
• There is a need for adaptive designs that can identify specific subsets of baskets with drug activity.

Purpose of the Study:

• To develop a Bayesian adaptive design methodology for oncology basket trials with binary endpoints.
• To utilize a Bayesian model averaging framework to account for heterogeneous drug activity across baskets.
• To improve the inference of basket-specific response rates.

Main Methods:

• Developed a Bayesian adaptive design using Bayesian model averaging.
• Inferred basket-specific response rates by averaging over a comprehensive model space.
• Evaluated the approach's performance against existing state-of-the-art methods.

Main Results:

• The proposed Bayesian model averaging approach favorably performs compared to existing methods.
• The methodology explicitly accounts for subsets of baskets with similar or dissimilar activity.
• The approach is computationally feasible and maintains stringent false positive rate requirements.

Conclusions:

• The novel Bayesian adaptive design offers a more accurate and flexible approach for oncology basket trials.
• This method enhances the ability to identify specific patient populations benefiting from investigational products.
• The Bayesian model averaging framework provides a robust tool for complex clinical trial designs.