Estimating optimal individualized treatment rules with multistate processes

TL;DR

This paper introduces a nonparametric method to estimate optimal personalized treatment rules using multistate process data, with theoretical guarantees and practical validation in clinical trial data.

Contribution

It develops a novel outcome weighted learning approach tailored for multistate process data in randomized trials, with rigorous theoretical properties and inference procedures.

Findings

Method performs well with small sample sizes.

Accurate confidence intervals are achievable.

Effective in high censoring scenarios.

Abstract

Multistate process data are common in studies of chronic diseases such as cancer. These data are ideal for precision medicine purposes as they can be leveraged to improve more refined health outcomes, compared to standard survival outcomes, as well as incorporate patient preferences regarding quantity versus quality of life. However, there are currently no methods for the estimation of optimal individualized treatment rules with such data. In this article, we propose a nonparametric outcome weighted learning approach for this problem in randomized clinical trial settings. The theoretical properties of the proposed methods, including Fisher consistency and asymptotic normality of the estimated expected outcome under the estimated optimal individualized treatment rule, are rigorously established. A consistent closed-form variance estimator is provided and methodology for the calculation of simultaneous confidence intervals is proposed. Simulation studies show that the proposed methodology and inference procedures work well even with small sample sizes and high rates of right censoring. The methodology is illustrated using data from a randomized clinical trial on the treatment of metastatic squamous-cell carcinoma of the head and neck.