Depth Importance in Precision Medicine (DIPM): A Tree and Forest Based Method

Abstract

We propose the novel implementation of a depth variable importance score in a classification tree method designed for the precision medicine setting. The goal is to identify clinically meaningful subgroups to better inform personalized treatment decisions. In the proposed Depth Importance in Precision Medicine (DIPM) method, a random forest of trees is first constructed at each node. Then, a depth variable importance score is used to select the best split variable. This score makes use of the observation that more important variables tend to be selected closer to root nodes of trees. In particular, we aim to outperform an existing method designed for the analysis of high-dimensional data with continuous outcome variables. The existing method uses an importance score based on weighted misclassification of out-of-bag samples upon permutation. Overall, our method is favorable because of its comparable and sometimes superior performance, simpler importance score, and broader pool of candidate splits. We use simulations to demonstrate the accuracy of our method and apply the method to a clinical dataset.