Bayesian approach to non-inferiority trials for normal means

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Inference

Background:

• Regulatory guidelines require novel statistical methods in clinical trials to align with frequentist principles, focusing on type-I error control and similar conclusions.
• Non-inferiority trials are crucial for evaluating new treatments against established active controls.

Purpose of the Study:

• To develop and evaluate a Bayesian approach for analyzing non-inferiority trials with normal response data.
• To ensure the proposed Bayesian method protects type-I error and yields comparable conclusions to frequentist strategies.

Main Methods:

• A Bayesian framework was constructed using non-informative priors for the experimental treatment's mean and variance.
• Priors for the active control's parameters were derived from historical trial data.
• A Bayesian decision criterion was developed to assess non-inferiority, subsequently compared to frequentist methods via simulations.

Main Results:

• Both Bayesian and frequentist approaches demonstrated similar performance in protecting type-I error and reaching conclusions.
• The Bayesian approach exhibited higher statistical power, particularly when variances were unknown.
• Both methods consistently identified non-inferiority when applied to two real-world datasets.

Conclusions:

• The proposed Bayesian approach is a valid and robust alternative for non-inferiority trials, aligning with regulatory expectations.
• This Bayesian method offers enhanced power and provides valuable posterior probabilities for various effect sizes, aiding in treatment evaluation.