Comparison of Borrowing Methods for Incorporating Historical Data in Single-Arm Phase II Clinical Trials

Area of Science:

• Clinical trial methodology
• Statistical inference in medicine

Background:

• Leveraging historical data in clinical trials enhances efficiency, reduces study size, and shortens duration.
• Careful selection of external data sources is critical to prevent bias when incorporating historical information.
• Borrowing methods offer a solution for effectively utilizing historical data in clinical trial design.

Purpose of the Study:

• To illustrate and compare the latest methods for borrowing historical data in single-arm phase II clinical trials.
• To examine the impact of various borrowing methods on statistical power and type I error rates.
• To guide researchers in selecting appropriate borrowing strategies for clinical trial design.

Main Methods:

• Implementation of static and dynamic versions of the power prior method.
• Incorporation of overlapping coefficient, loss functions, and meta-analytic predictive priors.
• Comparison with standard approaches (no historical data) and pooling approaches (all historical data).

Main Results:

• Dynamic borrowing methods demonstrated lower type I error inflation compared to pooling methods.
• The power prior approach with an overlapping coefficient effectively measured subject similarity, considering confounders and outcomes.
• A discounting function for the power parameter ensured the similarity between historical and current trial data.

Conclusions:

• A comprehensive overview of borrowing methods, including frequentist, Bayesian, static, and dynamic techniques, was provided.
• The study guides researchers in selecting the most suitable borrowing strategy for their specific clinical trial needs.
• Dynamic borrowing methods, particularly the power prior approach, offer advantages in controlling statistical errors and ensuring data similarity.