Regression-Based Estimation of Optimal Adaptive Treatment Strategies: Key Methods

Adaptive treatment strategies are routinely employed in diverse settings, from clinical care to educational and psychosocial supports, often in the form of a stepped care model. This form of treatment is guided by a set of decision rules that dictate how treatments or interventions should be allocated based on the evolving condition of the individual under care. We provide a review of regression-based methods of estimating optimal adaptive treatment strategies, focusing on the case of a continuous outcome. We provide a detailed overview of Q-learning, G-estimation, and dynamic weighted ordinary least squares (dWOLS). We also consider extensions to the case of censored outcomes. We demonstrate the developments using an example studying the impact of treatment decisions using data from the Quebec Human Immunodeficiency Virus Cohort Study.