Bayesian sample size determination using robust commensurate priors with interpretable discrepancy weights

Area of Science:

• Clinical Research Methodology
• Biostatistics
• Evidence-Based Medicine

Background:

• Randomized controlled trials (RCTs) are the gold standard for treatment efficacy.
• Bayesian methods offer a framework for incorporating prior knowledge from historical studies into new trial designs.
• Existing methods for borrowing strength from historical data often use discounting factors that lack interpretability.

Purpose of the Study:

• To address the non-interpretability of discounting factors (weights) in Bayesian meta-analysis for clinical trials.
• To develop a method for incorporating historical data from multiple sources that improves the elicitation of expert opinion.
• To derive an analytical sample size formula that is sensitive to interpretable weights.

Main Methods:

• Focusing on methods for incorporating historical data from multiple sources.
• Highlighting issues of nonmonotonicity in sample size calculations related to discounting factors.
• Proposing a linearization technique to ensure uniform changes in sample size with respect to weights.

Main Results:

• The proposed linearization technique results in interpretable weights, expressed as a percentage of information to borrow or discount.
• An analytical sample size formula was derived based on the proposed method.
• The linearization facilitates a more straightforward elicitation of expert opinion on the degree of borrowing or discounting historical information.

Conclusions:

• The developed method enhances the interpretability of weights in Bayesian clinical trial design.
• This approach simplifies the process of incorporating expert knowledge into sample size determination.
• Improved elicitation of expert opinion can lead to more efficient and robust clinical trial designs.