Adaptively leveraging external data with robust meta-analytical-predictive prior using empirical Bayes.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• Robust meta-analytical-predictive (rMAP) priors effectively utilize external data but require pre-specifying mixture coefficients, posing challenges during study design.
• The selection of an appropriate mixture coefficient is critical for balancing prior-data conflict, yet it is difficult to determine prospectively.

Purpose of the Study:

• To propose a novel empirical Bayes robust MAP (EB-rMAP) prior for adaptively leveraging external or historical data in statistical analyses.
• To address the practical limitations of pre-specifying mixture coefficients in rMAP priors at the study design stage.
• To develop a flexible and computationally efficient framework applicable to various endpoint types.

Main Methods:

• The EB-rMAP prior framework is developed, building upon Box's prior predictive p-value.
• A tuning parameter is incorporated to balance model parsimony and flexibility.
• The framework is designed for application to binomial, normal, and time-to-event data endpoints.
• Computational efficiency is a key feature of the EB-rMAP prior implementation.

Main Results:

• Simulation studies confirm that the EB-rMAP prior demonstrates robustness when encountering prior-data conflict.
• The proposed method effectively preserves statistical power in the presence of conflicting data.
• The EB-rMAP prior was successfully applied to a dataset encompassing 10 oncology clinical trials, including a prospective study.

Conclusions:

• The EB-rMAP prior offers a robust and adaptive approach to incorporating external data in statistical modeling.
• This novel framework simplifies study design by eliminating the need for pre-specified mixture coefficients.
• The EB-rMAP prior provides a flexible, efficient, and powerful tool for biostatistical applications in clinical research.