A0237
Title: An effective framework for estimating individualized treatment rules
Authors: Guanhua Chen - University of Wisconsin-Madison (United States) [presenting]
Jared Huling - University of Minnesota (United States)
Joowon Lee - University of Wisconsin, Madison (Korea, South)
Abstract: Estimating individualized treatment rules (ITRs) is fundamental in causal inference, particularly for precision medicine applications. Traditional ITR estimation methods rely on inverse probability weighting (IPW) to address confounding factors and L1-penalization for simplicity and interpretability. However, IPW can introduce statistical bias without precise propensity score modeling, while L1-penalization makes the objective non-smooth, leading to computational bias and requiring subgradient methods. A unified ITR estimation framework is proposed and formulated as a constrained, weighted, and smooth convex optimization problem. The optimal ITR can be robustly and effectively computed by projected gradient descent. The comprehensive theoretical analysis reveals that weights that balance the spectrum of a `weighted design matrix' improve both the optimization and likelihood landscapes, yielding improved computational and statistical estimation guarantees. In particular, this is achieved by distributional covariate balancing weights, which are model-free alternatives to IPW. Extensive simulations and applications demonstrate that our framework achieves significant gains in both robustness and effectiveness for ITR learning against existing methods.
