Decision List for Optimal Treatment Regime

• Activity Number: 69
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Yichi Zhang*+ and Eric B. Laber and Anastasios Tsiatis and Marie Davidian
• Companies: North Carolina State University and North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: Treatment regime ; Decision list ; NSABP clinical trial

Abstract

A treatment regime is a rule that recommends a treatment for each patient based on his/her diagnostic information. Due to patient heterogeneity, using a treatment regime could lead to better clinical outcome than giving an identical treatment to the entire population of patients. To obtain an interpretable regime that can be potentially accepted by physicians, we suggest to employ the concept of decision list, a finite sequence of "if-then" statements. Furthermore, we require the condition in each if-clause be some simple inequality involving at most two covariates. Consequently, the most relevant covariates are identified, which may provide insight into therapy. We propose a novel algorithm to learn the optimal treatment regime from a randomized or observational study, by maximizing the augmented inverse probability weighted estimator for the mean clinical outcome over the class of decision lists. The proposed method performs favorably to competing methods in simulation studies, and is illustrated by the application to the NSABP clinical trial on breast cancer.