UC Berkeley

The optimal dynamic treatment rule (ODTR) framework oers an approach for understand-ing which kinds of patients respond best to specic treatments { in other words, treatment ef- fect heterogeneity. Further, given an (optimal) dynamic treatment rule, it may be of interest to evaluate that rule { that is, to ask the causal question: what is the expected outcome had every subject received treatment according to that rule? Following the \causal roadmap," in this dissertation, we causally and statistically dene the ODTR and its value. Building on work by Luedtke and van der Laan, we provide an introduction to and show nite-sample performance for (1) estimating the ODTR using the ODTR SuperLearner (Chapter 1); and (2) estimating the value of an (optimal) dynamic treatment rule using dierent estimators, such as cross-validated targeted maximum likelihood (CV-TMLE; Chapter 2). We addition- ally augment the ODTR SuperLearner by considering stochastic treatment rules and risk criteria that consider the variability of the value of the rule (Chapter 3). We apply these estimators of the ODTR and its value to the \Interventions" Study, an ongoing random- ized controlled trial, to identify whether assigning cognitive behavioral therapy (CBT) to criminal justice-involved adults with mental illness using an ODTR signicantly reduces the probability of recidivism, compared to assigning CBT in a non-individualized way. We hope this work contributes to understanding the toolbox of methods that can be used to advance the elds of precision medicine, public policy, and health.