Semiparametric Bayesian inference for optimal dynamic treatment regimes via dynamic marginal structural models

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Dynamic Treatment Regimes (DTRs) are predominantly analyzed using frequentist methods.
• Bayesian approaches offer superior uncertainty representation, especially at the individual patient level.

Purpose of the Study:

• To extend Bayesian methods for Marginal Structural Models to DTR inference.
• To enable causal inference and personalized decision-making within DTRs.
• To develop a robust Bayesian framework for DTR analysis.

Main Methods:

• Linking observational data to a randomized DTR world for causal inference.
• Maximizing posterior predictive utility derived from nonparametric Bayesian semiparametric models.
• Utilizing posterior predictive inference and nonparametric Bayesian bootstrap for double robust inference.

Main Results:

• The proposed methods facilitate uncertainty quantification at the individual level.
• Demonstrated utility in a simulation study and an HIV therapy adaptation example.
• Enabled personalized treatment recommendations based on individual patient data.

Conclusions:

• Bayesian semiparametric inference provides a powerful framework for DTR analysis.
• The developed methods enhance causal inference and personalized decision-making in DTRs.
• This approach holds significant potential for optimizing treatment strategies in complex health scenarios.