Identifying Two-Stage Optimal Dynamic Treatment Regimes: Compare Performances of Different Methods Under Model Misspecification

• Activity Number: 355 - Contributed Poster Presentations: Biopharmaceutical Section
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Sooyeong Lim* and Chen Chen and Rhonda Szczesniak and Gary Lewis McPhail and Bin Huang
• Companies: Miami University and Cincinnati Children's Hospital and Cincinnati Children's Hospital and Cincinnati Children's Hospital and Cincinnati Children's Hospital
• Keywords: Dynamic Treatment Regime (DTR); Personalized Medicine; Model misspecification; R[d^(opt)]; Cystic Fibrosis; Bayesian additive regression tree (BART)

Abstract

Dynamic Treatment Regimes (DTR) are a sequence of decisions made over time, e.g. medical treatment is dynamically adjusted to the patient’s responses. Utilizing existing electronic medical records collected from clinic, we search for optimal DTR that maximize a desirable outcome for as many patients as possible. Q-learning and A-learning are two reinforcement learning algorithms proposed for finding the optimal DTR. Here, we compare a novel application of Bayesian Additive Regression Tree(BART) to Q-and A-learning for K-stage DTR. To assess how different DTR algorithms perform in correctly identifying the optimal DTR, a revised R[d^(opt)] is proposed. The revised R[d^(opt)] improves the original formula (Schulte et al. , 2012) by ensuring its value always falls within the 0-1 range. Different DTR methods for the two-stage setting are compared using R[d^(opt)] under the potential model misspecification setup. Finally, we applied the different methods to real-world data derived from the US Cystic Fibrosis Patient Registry.