Bayesian adaptive design for covariate-adaptive historical control information borrowing.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Rising clinical trial costs necessitate innovative approaches, including the use of historical control data.
• Existing methods for borrowing historical data may lead to covariate imbalance, potentially compromising trial validity.
• Covariate imbalance can introduce confounding factors, affecting study endpoints and the internal validity of current trials.

Purpose of the Study:

• To propose a novel Bayesian adaptive design for clinical trials that incorporates historical control data.
• To address the issue of covariate imbalance in trials utilizing historical controls.
• To enhance the reliability and validity of novel treatment assessments by improving the allocation ratio to the intervention arm.

Main Methods:

• A Bayesian design is proposed that adaptively borrows and updates treatment allocation ratios.
• The design utilizes covariate-dependent similarity assessments between current and historical control data.
• Covariate-dependent discrepancy parameters and regularized local regression are employed for parameter estimation.

Main Results:

• The proposed design allows for varying degrees of similarity between current and historical controls based on subject characteristics.
• The method ensures covariate balance across treatment arms and between current and historical control data.
• Extensive evaluation was conducted using data from two placebo-controlled trials on vertebral fracture risk in post-menopausal women.

Conclusions:

• The novel Bayesian adaptive design effectively addresses covariate imbalance when borrowing historical control data.
• This approach enhances the internal validity and reliability of clinical trial results.
• The design offers a more robust method for assessing novel treatments in the presence of rising trial costs.