A Bayesian basket trial design using a calibrated Bayesian hierarchical model.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Oncology

Background:

• Basket trials assess targeted therapies across cancer types with shared molecular aberrations.
• Bayesian hierarchical modeling aims to boost statistical power by borrowing information across cancer types.
• Existing Bayesian models struggle with appropriate information borrowing and may inflate type I error rates.

Purpose of the Study:

• To propose a novel calibrated Bayesian hierarchical model for basket trial design.
• To improve the control of type I error rates compared to standard Bayesian hierarchical models.
• To offer a more flexible and statistically robust approach for evaluating targeted therapies.

Main Methods:

• A calibrated Bayesian hierarchical model is introduced, defining the shrinkage parameter as a function of treatment effect similarity across subgroups.
• The calibration uses simulation to adaptively control information borrowing based on subgroup treatment effect heterogeneity.
• The model assumes a binary endpoint for evaluating treatment efficacy.

Main Results:

• Simulation studies demonstrate superior control of type I error rates compared to traditional Bayesian hierarchical models.
• The proposed method shows improved performance even in scenarios with moderate deviations from the null hypothesis.
• While type I error rates can be inflated in specific cases, they remain better controlled than existing methods.

Conclusions:

• The calibrated Bayesian hierarchical model offers a practical and flexible design for basket trials.
• This approach provides better control over type I error rates, enhancing reliability.
• Software for implementation is publicly available, facilitating adoption in clinical research.