Active Learning Clinical Trials for Personalized Medicine

• Activity Number: 12
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Yingqi Zhao*+ and Stanislav Minsker and Guang Cheng
• Companies: University of Wisconsin-Madison and Duke University and Purdue University
• Keywords: Active learning ; Adaptive clinical trials ; Cost effectiveness ; Individualized treatment rules ; Personalized medicine

Abstract

Individualized treatment rules (ITRs) has become increasingly popular due to its adaptively to individual patient characteristics. The data collected from randomized clinical trials are often utilized for estimating the optimal ITRs. However, it is known that these trials are expensive and time consuming and, in particular, are not designed for the purpose of estimating ITRs. We propose a cost-effective Active Learning method for estimating the optimal ITRs from an ongoing clinical trial. In particular, our method recruits only the "most informative" patients (in terms of learning the optimal ITRs). We illustrate the theoretical advantages of our active learning approach by deriving the corresponding risk bounds. Simulation studies and data analyses demonstrate the favorable empirical performance over competing methods.