Predicting Individualized Treatment Effects via Random Forests of Interaction Trees
Abstract
The structure for approving medical treatments in the U.S. consists of assessing the overall effect and is consequently designed only for the “average patient”. The inevitability that treatments will be very successful for some patients and not for others has influenced a new founded and controversial strategy for approving treatments. Precision medicine is an approach that takes individual differences into account and has therefore become a growing interest in many biomedical applications. Its aim is to deliver tailored or personalized treatment based on specific characteristics an individual has. It is not intended to create treatments unique to patients but rather to classify individuals into subpopulations that differ in their response to a specific treatment. To advance precision medicine, it is vital to understand the differential effects of a treatment.Tree-based methods are dominant among many approaches in this effort to understand differential treatment effects. In general, they excel in dealing with complex interactions. The treatment-by-covariates interactions involved in differential treatment effects may be of nonlinear and high order forms. By recursively grouping data on the basis of a two-sample test statistic, trees facilitate a powerful comprehensive modeling. Interaction trees (IT), proposed by Su et.al. in 2009, is an extension of trees that, in its construction, explicitly allows covariate-treatment interactions to be assessed. Ultimately, subpopulations in which treatment effects are heterogeneous are uncovered.
Subject Area
Statistics
Recommended Citation
Franco, Annette Pena, "Predicting Individualized Treatment Effects via Random Forests of Interaction Trees" (2017). ETD Collection for University of Texas, El Paso. AAI10689119.
https://scholarworks.utep.edu/dissertations/AAI10689119