Propensity-score-based priors for Bayesian augmented control design.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacoeconomics

Background:

• Demonstrating drug effectiveness requires adequate and well-controlled (A&WC) studies, typically randomized controlled trials (RCTs).
• RCTs can be infeasible due to size, duration, and cost, necessitating alternative approaches to reduce sample size.

Purpose of the Study:

• To explore using external data to augment control arms in Bayesian clinical trials.
• To address challenges in forming data-driven priors and minimize selection bias when incorporating external data.

Main Methods:

• Propensity score methods were employed to estimate patient probabilities of trial participation and minimize selection bias.
• Two matching schemes using propensity scores, estimated via generalized boosted methods, were applied to a real-world example.
• Bayesian augmented control was performed on a trial with disproportionate allocation using external data.

Main Results:

• The data augmentation process effectively prevented prior and data conflict.
• Improved precision in the estimation of the average treatment effect was observed.
• Propensity score matching successfully paired similar subjects between external and trial data.

Conclusions:

• Bayesian augmented control using propensity score methods offers a viable strategy to incorporate external data in clinical trials.
• This approach can enhance efficiency by reducing sample size requirements and improving the reliability of treatment effect estimates.
• Careful consideration of data consistency and bias mitigation is crucial for successful data-driven prior formation.