Optimal real-time dynamic treatment regimes with application to oxytocin use in preventing postpartum hemorrhage

Abstract: Real-time dynamic treatment regimes (real-time DTRs) refers to decision rules that personalize patient treatment in real-time based on treatment and covariate histories. These rules are crucial for real-time clinical decision support systems and automated drug delivery systems for chronic diseases. Although considerable statistical and machine learning DTR methods have been developed, they are designed for a small number of fixed decision points, and thus can not adapt to real-time cases. This paper proposes a new semiparametric Bayesian method for estimating an optimal treatment regime in real-time, which allows for the existence of latent individual level variables. Specifically, random real-time DTRs are defined through interventional parameters, the optimal values of which are estimated by maximizing the posterior predictive utility. The proposed approach is compared with alternative methods using simulated datasets, and applied to estimate the optimal real-time oxytocin administration regime for preventing postpartum hemorrhage.