A rank-based approach to estimating monotone individualized two treatment regimes

Abstract

Developing effective individualized treatment rules (ITRs) for diseases is an important goal of clinical research. Much effort has been devoted to estimating individualized treatment effects in the recent literature. However, there have not been systematic studies on the robust inference for individualized treatment effects when there exist potential outliers. We propose a monotone ITR in the framework of a semiparametric generalized regression with two treatments and estimate the treatment effects via a smoothed maximum rank correlation procedure. We provide sufficient conditions under which the proposed estimator has an asymptotically normal distribution whose variance can be consistently estimated based on a resampling procedure. We evaluate the finite-sample properties of our proposed approach via simulation studies. We also illustrate the proposed method by applying it to a data set from an AIDS clinical trials study.