Bayesian Empirical Likelihood Regression for Semiparametric Estimation of Optimal Dynamic Treatment Regimes

Area of Science:

• Biostatistics
• Machine Learning
• Health Informatics

Background:

• Dynamic treatment regimes (DTRs) are crucial for personalized medicine, adapting treatments over time based on patient responses.
• Existing Bayesian methods for DTRs often rely on strong distributional assumptions, potentially limiting their accuracy and applicability.
• Accurate estimation of optimal DTRs is essential for improving patient outcomes in complex treatment scenarios.

Purpose of the Study:

• To propose a semiparametric Bayesian modeling framework for DTRs that enhances estimation accuracy.
• To develop a method that avoids restrictive distributional assumptions for intermediate and final outcomes.
• To offer a computationally efficient approach using variational Bayes approximation.

Main Methods:

• Developed a semiparametric Bayesian approach utilizing empirical likelihoods within a regression estimation framework.
• Incorporated flexible mean function specifications for stagewise outcomes (linear or non-linear).
• Employed variational Bayes approximation for computational efficiency, mitigating issues with MCMC and empirical likelihood.

Main Results:

• The proposed semiparametric method demonstrated superior accuracy in estimating optimal DTRs compared to Q-learning and parametric Bayesian methods.
• Performance gains were particularly notable when parametric assumptions regarding regression error distributions were violated.
• Validation through simulations and analysis of STAR*D trial data confirmed the method's robustness and accuracy.

Conclusions:

• The semiparametric Bayesian approach offers a robust and accurate alternative for modeling DTRs, especially in data-rich environments with potential distributional misspecification.
• Empirical likelihoods and variational Bayes provide a powerful combination for flexible and computationally feasible DTR estimation.
• This method holds promise for advancing personalized treatment strategies in various clinical applications.