A Narrative Review of Multimodal Data Fusion Strategies for Precision Risk Prediction in Coronary Artery Disease: Advances, Challenges, and Future Informatics Directions.

Abstract

Traditional coronary artery disease (CAD) risk scores offer limited precision, often failing to capture the complex, multifactorial nature of the disease. The proliferation of multimodal data from imaging, genomics, electronic health records (EHRs), and wearables offers a transformative opportunity for more individualized risk prediction. This narrative review systematically maps and critically evaluates the landscape of multimodal data fusion for CAD risk prediction. Following Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines, we synthesized 39 empirical studies published from 2009 to 2025 to identify key methodological patterns, informatics challenges, and future directions. Our synthesis reveals consistent methodological patterns: (1) integrating imaging biomarkers (e.g. coronary computed tomography angiography, coronary artery calcium scoring) with clinical data robustly enhances risk discrimination and reclassification; (2) adding polygenic risk scores provides incremental value, typically via late-fusion models; and (3) leveraging longitudinal EHR data with machine learning captures dynamic risk trajectories, outperforming static scores. Advanced machine learning architectures, particularly deep and graph neural networks, are pivotal for enabling automated feature extraction and modeling complex cross-modal interactions. Despite these advances, significant informatics hurdles persist, including data heterogeneity, algorithmic bias, the need for robust external validation, and challenges in clinical workflow integration. Multimodal data fusion is a cornerstone of precision cardiology, but realizing its clinical potential requires a concerted focus on developing fair, interpretable, and scalable methodological frameworks to translate complex data into improved patient outcomes.