Using horseshoe prior for incorporating multiple historical control data in randomized controlled trials.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• Historical controls are crucial for modern clinical trials, often integrated using meta-analytic methods or power priors.
• Existing methods may face challenges with heterogeneity among historical control groups.

Purpose of the Study:

• To propose a new statistical method for incorporating multiple historical controls in randomized controlled trials.
• To evaluate the performance of this method, particularly when historical controls exhibit heterogeneity.

Main Methods:

• A novel approach using a horseshoe prior, a type of global-local shrinkage prior, to incorporate multiple historical controls.
• The method accommodates scenarios where historical controls share the same distribution as the current control, or where a few are heterogeneous and potentially biased.
• Analysis of two clinical trial examples (binary and time-to-event endpoints) and simulation studies.

Main Results:

• The proposed method decreased the posterior standard deviation of the treatment effect by accounting for bias between current and heterogeneous historical controls.
• Statistical power was higher with the proposed method compared to existing methods when current and historical controls followed the same distribution.
• The method demonstrated advantages when few or no heterogeneous historical controls were anticipated.

Conclusions:

• The proposed horseshoe prior method offers an effective way to incorporate multiple historical controls in clinical trial analysis.
• This approach improves statistical efficiency and precision, particularly in situations with limited or heterogeneous historical data.
• The method is advantageous for trial designs anticipating few or no heterogeneous historical controls.