Using Reinforcement Learning Strategies to Discover the Optimal Treatment for Advanced Colorectal Cancer Patients

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Zheng Ren*+ and Michael R. Kosorok
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Personalized Medicine ; Biomarkers ; Colorectal cancer ; Clinical trials ; Dynamic treatment regime ; Reinforcement learning

Abstract

Reinforcement learning methods have been developed to identify individual dynamic treatment regimens for cancer patients with the ability to identify the optimal treatment strategy from a complex clinical setting, including selecting from several first line treatment options, several second line treatment options and the time of initiating second line treatment. In this study, we use reinforcement learning to analyze data from a colorectal cancer trial. Biomarkers are believed to be useful for improving treatment and prognosis, so the optimal dynamic treatment rule was determined for individuals based on their clinical factors and biomarkers using reinforcement learning. The best treatment plan obtained from reinforcement learning is then compared with results from other methods in a simulation study, demonstrating that reinforcement learning is useful for discovering personalized medicine using a good clinical trial design with a reasonable sample size.