A unified Bayesian framework for bias adjustment in multiple comparisons from clinical trials

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacovigilance

Background:

• Multiple comparisons in clinical trials can lead to focusing on the most promising results, potentially causing selection bias.
• Selection bias, a deviation from true treatment effects, can lead to failed pivotal studies and misinterpretation of drug efficacy.
• Existing methods may lack flexibility in addressing selection bias in complex clinical trial settings.

Purpose of the Study:

• To propose a general and unified Bayesian framework to address selection bias in clinical trials with multiple comparisons.
• To offer a flexible and generalized solution that does not require a priori specification of a parametric prior distribution.
• To enable more accurate interpretation of clinical trial results by adjusting for selection bias.

Main Methods:

• Developed a unified Bayesian framework for addressing selection bias in clinical trials.
• Employed Bayesian Model Averaging (BMA) with Gaussian Mixture Models (GMMs) as a flexible prior distribution.
• Conducted simulation studies to compare the proposed method with existing estimators, including the normal shrinkage estimator.

Main Results:

• The proposed Bayesian framework effectively adjusts for selection bias in clinical trials.
• Simulation studies demonstrated superior performance of the BMA-GMM approach over the normal shrinkage estimator.
• The method was successfully applied to a real-world clinical trial investigating dulaglutide's cardiovascular effects.

Conclusions:

• The proposed Bayesian framework provides a robust and flexible method for correcting selection bias in clinical trials.
• Bayesian Model Averaging over Gaussian Mixture Models is recommended for its performance and adaptability.
• This approach enhances the reliability and interpretability of clinical trial findings, particularly in drug development.