Estimation of Optimal DTRs by Directly Modeling Regimes

Abstract

In this chapter, we consider several approaches to estimating the optimal dynamic treatment regime by directly modeling the regimes as opposed to modeling the conditional mean outcome: inverse probability of treatment weighting, marginal structural models, and classification-based methods. The fundamental difference between the approaches considered in the current chapter and those considered in previous chapters (e.g. Q-learning and G-estimation) lies in the primary target of estimation (and inference): the methods considered presently target the parameters of the decision rule itself.