DBMS: dynamic borrowing method for frequentist hybrid control designs based on historical-current data similarity

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Clinical trials, especially for rare diseases, often face challenges with small sample sizes, impacting statistical power.
• Existing Bayesian methods for information borrowing are established, but frequentist approaches are also emerging.
• Current frequentist methods like test-then-pool determine data incorporation based on hypothesis testing outcomes.

Purpose of the Study:

• To introduce a novel dynamic borrowing method for leveraging historical data in clinical trials.
• To develop a flexible approach that adjusts the degree of information borrowing based on data similarity.
• To enhance statistical power in clinical trials, particularly those with limited sample sizes.

Main Methods:

• A dynamic borrowing method is proposed, adjusting information borrowing from 0% to 100%.
• Two similarity measures are presented: one based on the t-distribution's density function and another using a logistic function.
• Method performance is evaluated using Monte Carlo simulations.

Main Results:

• The proposed dynamic borrowing method allows for flexible and adaptive use of historical data.
• Simulations demonstrate the effectiveness of the dynamic borrowing approach in various scenarios.
• The reanalysis of an actual clinical trial data confirms the practical utility of the method.

Conclusions:

• Dynamic information borrowing offers a promising strategy to optimize the use of historical data in clinical trials.
• The proposed methods provide a data-driven approach to balance the incorporation of external information.
• This technique can significantly improve statistical power and efficiency in rare and pediatric disease studies.