Estimating Optimal Dynamic Treatment Regimes with Shared Decision Rules Across Stages

• Activity Number: 21
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Bibhas Chakraborty*+ and Erica Moodie and Palash Ghosh
• Companies: Duke-NUS Graduate Medical School and McGill University and Duke-NUS Graduate Medical School
• Keywords: dynamic treatment regime ; Q-learning ; shared decision rule

Abstract

A dynamic treatment regime is a set of decision rules that dictate how to individualize treatment to patients based on available treatment and covariate history. In many scientific domains, these rules tend to be shared across stages of intervention. In this paper, for estimating shared treatment rules in a three-stage setting, we propose two novel simultaneous estimation procedures, one based on standard Q-learning, and the other based on a bias-corrected version of Q-learning. Through extensive simulations, we show how these newly proposed methods compare with simple and weighted averages of stage-specific estimates from standard and bias-corrected Q-learning, in terms of bias, variance and mean squared error of the parameters indexing the estimated dynamic regime, percentage of optimal treatment allocation using the estimated regime, and the Value function of the estimated regime. Analysis of a data set from a depression clinical trial is provided as an illustration.