Bayesian basket trial design with exchangeability monitoring

Area of Science:

• Oncology
• Biostatistics
• Clinical Trial Design

Background:

• Precision medicine aims to personalize treatments, challenging traditional approaches relying on population averages.
• Basket trials evaluate targeted therapies across various cancer subtypes defined by biomarkers, not histology.
• Existing basket trial designs face limitations in statistical power and demonstrating consistent effectiveness across subtypes.

Purpose of the Study:

• To present a novel sequential basket trial design using Bayesian monitoring rules.
• To address limitations of existing basket trial analyses, particularly regarding power and evidence of subtype-specific effectiveness.
• To introduce a hierarchical modeling strategy for information sharing among diverse patient subtypes.

Main Methods:

• Development of a sequential basket trial design with Bayesian monitoring.
• Implementation of a hierarchical modeling strategy for pooling data across discrete, potentially non-exchangeable subtypes.
• Validation through analysis, permutation, and simulation studies using a real-world basket trial example (vemurafenib in BRAF V600 mutant melanoma).

Main Results:

• The proposed methodology offers improved statistical power compared to traditional basketwise analyses, especially with imbalanced enrollment.
• The Bayesian approach facilitates interim analyses and provides robust measures of treatment effectiveness across different cancer subtypes.
• Demonstrated feasibility and effectiveness using a case study involving vemurafenib for BRAF V600 mutant non-melanoma.

Conclusions:

• The novel Bayesian sequential design enhances the efficiency and evidential value of basket trials in precision medicine.
• This approach allows for more reliable assessment of targeted therapy effectiveness in subpopulations with specific biomarkers.
• The methodology provides a framework for more informative clinical trial designs in precision oncology.