Identifying optimally cost-effective dynamic treatment regimes with a Q-learning approach

Abstract Health policy decisions regarding patient treatment strategies require consideration of both effectiveness and cost. We propose a two-step approach for identifying an optimally cost-effective interpretable dynamic regime. First, we develop combined Q-learning policy-search to estimate optimal list-based regimes under constraint on expected costs. Second, iterative procedure select regime from set candidate corresponding different cost constraints. Our can in the presence time-varying confounding, censoring, correlated outcomes. Through simulation studies, examine operating characteristics our flexible modelling approaches. also apply methodology identify assigning adjuvant therapies endometrial cancer patients.