Learning non-monotone optimal individualized treatment regimes

TL;DR

This paper introduces a robust method for identifying optimal individualized treatment regimes that accounts for model misspecification and non-monotonic treatment effects, demonstrated through simulations and real data.

Contribution

It presents a novel estimation approach that is robust to various model misspecifications and incorporates a single index structure for nonparametric treatment difference estimation.

Findings

Method performs well in simulations under misspecification.

Successfully applied to real data on maternal smoking and birth weight.

Theoretical properties of the treatment strategy are established.

Abstract

We propose a new modeling and estimation approach to select the optimal treatment regime from different options through constructing a robust estimating equation. The method is protected against misspecification of the propensity score model, the outcome regression model for the non-treated group, or the potential non-monotonic treatment difference model. Our method also allows residual errors to depend on covariates. A single index structure is incorporated to facilitate the nonparametric estimation of the treatment difference. We then identify the optimal treatment through maximizing the value function. Theoretical properties of the treatment assignment strategy are established. We illustrate the performance and effectiveness of our proposed estimators through extensive simulation studies and a real dataset on the effect of maternal smoking on baby birth weight.