Ensuring exchangeability in data-based priors for a Bayesian analysis of clinical trials

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacoeconomics

Background:

• Randomized controlled trials (RCTs) are often infeasible in rare diseases and pediatric studies due to size, duration, and cost.
• Leveraging external control data via priors can reduce sample size but risks biased estimates and inflated Type I error if not carefully constructed.
• Assessing both confirmatory and supplementary knowledge is crucial to prevent
• cherry-picking
• advantageous prior information.

Purpose of the Study:

• To develop and evaluate a method for incorporating supplemental control data into clinical trials.
• To minimize selection bias and prior-data conflict when borrowing control arms.
• To improve the precision of average treatment effect (ATE) estimation in single-arm trials using external controls.

Main Methods:

• Propensity score methods are employed to weight supplemental control subjects based on pretreatment characteristic similarity.
• A measure of overlap in propensity score distributions is proposed to operationalize similarity.
• The study considers a single experimental arm with the control arm entirely borrowed from supplemental data.

Main Results:

• Simulation experiments demonstrate that the proposed method effectively reduces prior-data conflict.
• The method leads to improved precision in estimating the average treatment effect.
• Weighting based on propensity scores mitigates bias introduced by borrowing external control data.

Conclusions:

• The proposed propensity score-based approach offers a robust strategy for utilizing supplemental control data in clinical trials.
• This method enhances statistical efficiency and reliability, particularly when RCTs are challenging.
• Careful assessment of prior information and control subject similarity is key to valid and precise treatment effect estimation.