Multivariate Bayesian Dynamic Borrowing for Repeated Measures Data With Application to External Control Arms in...

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Pharmacometrics

Background:

• Borrowing analyses are crucial for enhancing the efficiency and validity of clinical trial interpretations.
• Accurate long-term treatment effect estimation is vital for informed clinical decision-making, especially in studies with continuous endpoints.
• Existing methods may not fully leverage external data or account for complex scenarios like intercurrent events.

Purpose of the Study:

• To develop a robust Bayesian dynamic borrowing method for multivariate data in clinical trials.
• To enable causally valid, long-term treatment effect estimation from open-label extension studies by incorporating external control arm data.
• To provide a generally applicable framework for Bayesian dynamic borrowing analyses using multivariate normal likelihoods.

Main Methods:

• Utilized robust mixture priors within a multivariate dynamic borrowing framework.
• Developed a Bayesian approach for estimating multivariate summary metrics.
• The method accommodates various parameter models and addresses missing data due to intercurrent events through hypothetical estimand strategies.

Main Results:

• The proposed method allows for dynamic incorporation of prior beliefs from external control arms.
• It facilitates the estimation of long-term treatment effects for continuous endpoints.
• Demonstrated applicability to multivariate summary metrics and complex data scenarios.

Conclusions:

• The developed Bayesian dynamic borrowing method offers a robust approach for clinical trial analysis.
• This methodology enhances the ability to derive reliable long-term treatment effect estimates.
• The framework is broadly applicable, particularly for open-label extension studies and handling intercurrent events.