Traditional Chinese medicine (TCM) emphasizes syndrome differentiation and individualized interventions. However, in real-world clinical studies, complex confounding and the unobservability of counterfactual outcomes pose major challenges for evaluating individualized treatment effects. Causal machine learning, grounded in the potential outcomes framework, provides a flexible, nonparametric approach to model treatment assignment and outcome distributions. By adjusting for baseline and time-varying confounding, it enables the generation of patient-specific counterfactual outcomes and more accurate estimation of causal effects at the individual level. This methodological advance addresses the limitations of conventional approaches that focus mainly on average treatment effects. Taking advanced colorectal cancer as a case example, we demonstrate how causal machine learning can identify "which patients benefit most from which treatment?" thereby supporting personalized clinical decision-making and optimization of therapeutic strategies in TCM. Overall, causal machine learning holds both methodological significance and direct clinical value for individualized efficacy research in TCM.
