PS-SAM: propensity-score-integrated self-adapting mixture prior to dynamically and efficiently borrow information...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Health Data Science

Background:

• Historical data can enhance randomized controlled trials (RCTs) efficiency and reduce sample size requirements.
• Patient characteristic differences between historical and current trial data pose a challenge.
• Propensity score methods (matching, inverse probability weighting) adjust for baseline heterogeneity but are vulnerable to unmeasured confounders.

Purpose of the Study:

• To develop a robust statistical method for incorporating historical data into RCTs, specifically addressing bias introduced by unmeasured confounders.
• To enhance the accuracy and reliability of causal inference from RCTs by leveraging historical data more effectively.
• To introduce the propensity score-integrated self-adapting mixture (PS-SAM) prior as a solution for adaptive information borrowing.

Main Methods:

• Integration of a self-adapting mixture (SAM) prior with propensity score matching and inverse probability weighting.
• Development of propensity score-integrated SAM (PS-SAM) priors to mitigate bias from unmeasured confounders.
• Utilizing simulation studies to evaluate the operating characteristics of the PS-SAM prior.

Main Results:

• The PS-SAM priors demonstrate robustness, yielding unbiased causal estimates when no unmeasured confounders exist.
• In the presence of unmeasured confounders, PS-SAM priors provide significantly less biased treatment effect estimates and improved type I error control.
• Simulation results confirm the desirable operating characteristics of the PS-SAM prior for adaptive information borrowing.

Conclusions:

• The PS-SAM prior methodology offers a robust approach for integrating historical data into RCTs, effectively handling unmeasured confounding.
• This method improves causal inference by enabling adaptive information borrowing, leading to more reliable treatment effect estimates.
• The proposed methodology is accessible through the R package "SAMprior".