Subgroup Identification in Randomized Clinical Trial Data Using Random Forests and Regression Trees

• Activity Number: 60
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jared Foster*+ and Jeremy Taylor and Bin Nan
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: randomized clinical trials ; subgroups ; random forests ; regression trees

Abstract

We consider the problem of identifying a subgroup of patients who may have an enhanced treatment effect in a randomized clinical trial. In such cases, there are often a moderate to large number of covariates, and it is desirable that any identified subgroup be defined by a limited number of covariates. For this problem, the development of a standard, pre determined strategy may help to avoid the well-known dangers of subgroup analysis. We present one such strategy, which we refer to as "Virtual Twins." This method involves predicting response probabilities for treatment and control "twins" for each subject using random forests. In order to identify a subset of enhanced treatment effect which is defined by a limited number of covariates, the difference in these probabilities is then used as the outcome in a classification or regression tree. We define a measure Q(A) to be the difference between the treatment effect in estimated subgroup A and the marginal treatment effect. We present several methods developed to obtain an estimate of Q(A), including estimation of Q(A) using estimated probabilities in the original data and using estimated probabilities in newly simulated data.