Data-guided Treatment Recommendation with Feature Scores

Abstract

Despite the availability of large amounts of genomics data, medical treatment recommendations have not successfully used them. In this paper, we consider the utility of high dimensional genomic-clinical data and nonparametric methods for making cancer treatment recommendations. This builds upon the framework of the individualized treatment rule [Qian and Murphy 2011] but we aim to overcome their method's limitations, specifically in the instances when the method encounters a large number of covariates and an issue of model misspecification. We tackle this problem using a dimension reduction method, namely Sliced Inverse Regression (SIR, [Li 1991]), with a rich class of models for the treatment response. Notably, SIR defines a feature space for high-dimensional data, offering an advantage similar to those found in the popular neural network models. With the features obtained from SIR, a simple visualization is used to compare different treatment options and present the recommended treatment.Additionally, we derive the consistency and the convergence rate of the proposed rec-1

Estimating heterogeneous treatment effects versus building individualized treatment rules: Connection and disconnection

“…The most commonly used one is to fit regression models for μ1 (x) and μ0 (x), and then τ (x) = μ1 (x)−μ 0 (x). As an addition to the rich literature, [Chen et al 2022] considered very general regression models and applied dimension reduction for high-dimensional covariates. Moreover, supervised learning algorithms are used to estimate μ1 and μ0 by the machine learning community [Hu et al 2021], including Bayesian Additive Regression Trees (BART) [Chipman et al 2010] and Random Forest (RF) [Wager and Athey 2018].…”

“…We can further justify this by providing a relationship between the value function and the estimation error. More specifically, [Chen et al 2022] have showed that for any ITR d, the reduction in value is upper bounded by the estimation error (See Lemma 1 in [Chen et al 2022]):…”

“…Building personalized treatment, also referred to as an individualized treatment rule (ITR) [Qian and Murphy 2011], is usually done through estimating the CATE first and then defining the optimal ITR as the sign function of the CATE (for two treatment options denoted as 1 and −1). [Chen et al 2022] provided a good introduction on building ITRs and called the aforementioned method an indirect approach. Although related, estimating the CATE and building optimal ITRs are separately studied in the literature.…”