Residual Weighted Learning for Estimating Individualized Treatment Rules

• Activity Number: 564
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xin Zhou*+ and Michael Kosorok
• Companies: University of North Carolina at Chapel Hill and University of North Carolina at Chapel Hill
• Keywords: Dynamic Treatment Regime ; Individualized Treatment Rule ; Truncated Hinge Loss ; RKHS

Abstract

Personalized medicine has received increasing attention among statisticians, computer scientists, and clinical practitioners, since patients often show significant heterogeneity in response to treatments. A major component of personalized medicine is to estimate individualized treatment rules. Recently, Zhao et al. (2012) proposed outcome weighted learning (OWL) to construct individualized treatment rules that directly optimize the clinical outcome. However, this approach may suffer a severe over-fitting problem when the sample size is small. In this article, we propose a new method, called Residual Weighted Learning (RWL) to alleviate this problem, and hence to improve the finite sample properties. Not like OWL which weights the misclassification errors by the clinical outcomes, the RWL weights the errors by the residuals from a regression fit of outcomes. We utilized the truncated hinge loss function in the RWL, and provide a difference of convex (d.c.) algorithm to solve the corresponding non-convex optimization problem. The performance of our proposed RWL method is illustrated in simulation studies and an analysis of chronic depression data.