Flexible evaluation of surrogate markers with Bayesian model averaging

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Epidemiology

Background:

• Surrogate markers are crucial for estimating treatment effects when primary endpoints require long-term follow-up in randomized clinical trials.
• Existing methods for evaluating surrogate marker effectiveness include model-based and nonparametric approaches, each with limitations such as potential model mis-specification or poor performance with small sample sizes.

Purpose of the Study:

• To propose and evaluate a novel Bayesian model averaging (BMA) approach for estimating the proportion of treatment effect explained by a surrogate marker.
• To offer a method that balances the flexibility of nonparametric approaches with the inferential strengths of parametric models.

Main Methods:

• Developed a Bayesian model averaging framework to estimate the proportion of treatment effect explained by a surrogate marker.
• Compared the proposed BMA method against traditional model-based and nonparametric approaches through simulation studies.
• Applied the BMA method to real-world data from the Diabetes Prevention Program study, using hemoglobin A1c as a surrogate for fasting glucose.

Main Results:

• Simulation studies indicated that the BMA approach outperforms existing methods, particularly when surrogate marker support is inconsistent and sample sizes are small.
• The application to the Diabetes Prevention Program data demonstrated the practical utility of the BMA method in a relevant clinical context.

Conclusions:

• The proposed Bayesian model averaging approach provides a robust and flexible method for evaluating surrogate markers in clinical trials.
• This method is advantageous in scenarios with limited data or complex surrogate marker relationships, offering improved estimation of treatment effects.