|
Abstract:
|
Clinical treatments of many chronic disorders are time-varying and adaptive to an individual patient's changing health status. In this context, a dynamic treatment regime is a list of decision rules that tells how the level of treatment should be tailored through time according to an individual's response to ongoing therapy. In this paper, we consider the problems of estimating dynamic treatment regimes and conducting inference on them from longitudinal data on patients. For estimation, we propose a method called Q-Learning. But this method being non-smooth, neither asymptotic normality nor bootstrap can be used to get valid tests of hypothesis. We illustrate this via simulations. As a remedy, we propose a regularized method and prove that the estimators obtained from this method are asymptotically normal. Performance of this method is assessed through simulations.
|
