Estimating Causal Individual Treatment Effects for Personalized Medicine Using Causal-Inspired Machine Learning

Abstract

Personalized medicine relies on identifying which patients will benefit most from a given treatment. Traditional average treatment effect (ATE) estimates often fail to capture the underlying heterogeneity in treatment response. In this study, we apply causal-inspired machine learning methods to estimate individual treatment effects (ITEs) using observational data from the Infant Health and Development Program (IHDP) dataset. We apply DoWhy, a Python library for causal inference, to estimate ATEs using multiple models (linear regression, propensity score matching, weighting, and stratification) and also, we extend the analysis to ITEs using meta-learners (T-Learner). The results from this study reflect a significant variation in treatment effects across individuals, reinforcing the need for personalized treatment policies. We conclude with implications for clinical decision-making and future research directions in causal machine learning for medicine.