Presentation (CTI)

OBJECTIVES: In epidemiology and public health, causal approaches are essential to elucidate disease etiology and intervention effects beyond associations. This review synthesized the conceptual foundations of Causal AI and its relationship to traditional causal inference methodologies, and developed an integrated framework for applying Causal AI to healthcare research by combining causal structure learning and counterfactual reasoning with machine learning.
METHODS: We conducted a narrative literature review to synthesize theoretical and methodological foundations for applying Causal AI in healthcare. A targeted search on Google Scholar identified relevant literature using keywords including machine learning, causal AI, healthcare, and treatment effect. Studies were selected based on conceptual relevance and methodological representativeness rather than exhaustive coverage, and were thematically synthesized to identify foundational approaches, emerging methodologies, and their implications.
RESULTS: We selected 32 core peer-reviewed articles. Based on a foundational framework identified in the literature, we structured the review around three methodological pillars of Causal AI: causal representation learning, causal discovery, and causal reasoning. Synthesizing established causal inference roadmaps, we outlined the analytical workflow—from causal model specification through result interpretation—into seven sequential stages and mapped applicable AI tools to each stage, systematically organizing opportunities for AI integration throughout the causal inference process. Finally, we categorized Causal AI methodologies addressing critical characteristics of healthcare data, including unobserved confounding, heterogeneous treatment effects, multi-environment settings, and longitudinal and survival data, providing a reference framework for selecting appropriate strategies based on research questions and data structures.
CONCLUSIONS: Causal AI addressed limitations of predictive machine learning by explicitly modeling causal structures and estimating intervention effects and counterfactual outcomes. This integrated framework provided a conceptual foundation for systematic application of Causal AI in healthcare research. It served as a methodological resource for study design and analytical decision-making in real-world evidence generation.