B0696
Title: Efficient learning of optimal individualized treatment rules
Authors: Weibin Mo - Purdue University (United States) [presenting]
Abstract: Recent development in data-driven decision science has seen great advancement in individualized decision-making. Existing methods typically require the initial estimation of some nuisance models. To protect consistency from nuisance model misspecification, the double robustness property has been widely advocated, while the concern of estimation efficiency is rarely studied. Efficiency is critical for stable and reliable predictions, as well as high power to justify treatment benefits. To improve the efficiency of the estimated individualized treatment rule (ITR), we propose an Efficient Learning (E-Learning) framework for finding an optimal ITR in the multi-categorical treatment setting. We establish the optimality of the proposed E-Learning in the presence of regression model misspecification and heteroscedasticity.
