Abebe S, Poli I , Jones RD and Slanzi D. Mach. In: Learn. Knowl. Extr. 2024, July 25:1, 1–20.
Abstract
In medicine, dynamic treatment regimes (DTRs) have emerged to guide personalized 1 treatment decisions for patients, accounting for their unique characteristics. However, existing 2 methods for determining optimal DTRs face limitations, often due to reliance on linear models 3 unsuitable for complex disease analysis and a focus on outcome prediction over treatment effect 4 estimation. To overcome these challenges, decision tree-based reinforcement learning approaches 5 have been proposed. Our study aims to evaluate the performance and feasibility of such algorithms: 6 Tree-based Reinforcement Learning (T-RL), DTR causal trees (DTR-CT), DTR causal forest (DTR-CF), 7 Stochastic tree-based reinforcement learning (SL-RL), and Q-learning with Random Forest. Using 8 real-world clinical data, we conducted experiments to compare algorithm performances. Evaluation 9 metrics included the proportion of correctly assigned patients to recommended treatments and the 10 empirical mean with standard deviation of expected counterfactual outcomes based on estimated 11 optimal treatment strategies. This research not only highlights the potential of decision tree-based 12 reinforcement learning for dynamic treatment regimes but also contributes to advancing personalized 13 medicine by offering nuanced and effective treatment recommendations.
