Borrowing using historical-bias power prior with empirical Bayes

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Incorporating historical data enhances current analysis precision without new observations.
• Traditional power priors struggle with limited historical studies and assume no historical bias.

Purpose of the Study:

• Develop a novel conditional power prior (historical-bias power prior) to address limitations of existing methods.
• Allow for historical bias and control information borrowing based on data criteria.

Main Methods:

• Utilized an empirical Bayes approach to create the historical-bias power prior.
• Embedded a Frequentist test-then-pool strategy within the weight function for adaptive borrowing.
• Investigated the impact of historical bias on borrowing approach operating characteristics via simulation.

Main Results:

• The historical-bias power prior demonstrated accurate estimation and robustly powerful tests for experimental treatment effects.
• Maintained good Type I error control, particularly when historical bias was present.
• The method effectively bridges Frequentist test-then-pool and Bayesian power prior approaches.

Conclusions:

• The historical-bias power prior offers a flexible and robust method for incorporating historical data, especially when bias is a concern.
• This approach improves statistical inference by adaptively leveraging relevant historical information.