Multicomponent Dynamic Treatment Regimes with Fractional Factorial Design

Abstract

A dynamic treatment regime (DTR) is a sequence of decision rules, one per decision point, that map individual characteristics to a recommended treatment. An optimal DTR yields the largest mean utility if applied to select treatments in the target population. In this thesis, we consider estimation of an optimal DTR when the treatments consist of many smaller components, e.g., these components might be different aspects of a personalized message targeting some positive behavior change. We introduce a novel design for estimating optimal DTRs in randomized study, where each randomization stage includes a fractional factorial design. We introduce methods for estimating an optimal regime in both the single-stage and multi-stage setting. We derive the estimator for the optimal regime based on the potential outcome framework and Q-learning. We conduct simulation studies to evaluate the performance of the methods. A case study with single-stage design is used as an illustrative example.