AI prediction models with omics data utilization for atherosclerosis: A systematic scoping review and meta-analysis

Abstract

Background and aims Cardiovascular disease (CVD) remains a leading cause of global mortality, necessitating advanced methodologies to elucidate its complex pathophysiology. The application of artificial intelligence (AI) to interpret high-dimensional omics data offers a significant opportunity for precision medicine. This study aims to systematically review the current landscape of AI technologies in cardiovascular omics research and compare predictive performance of omics-trained AI prediction models (APMs) against conventional risk prediction models (CRMs) in atherosclerosis. Methods We employed a two-phase systematic review framework. Study 1 (scoping review) mapped the broad landscape of AI applications in cardiovascular omics research by reviewing 218 eligible studies. Study 2 (meta-analysis) comprised a systematic meta-analysis of 38 distinct, atherosclerosis-specific studies to quantify the incremental performance of APMs over CRMs, assessed via the difference in area under the curve (Delta AUC). Results Study 1 (scoping review) demonstrated substantial growth in AI modeling, multi-omics, and advanced omics methodologies from 2024 onwards. In Study 2 (meta-analysis), APMs significantly outperformed CRMs (pooled Delta AUC = 0.0586; 95% CI: 0.0335-0.0836; p < 0.0001) with a moderate level of between-study heterogeneity (I-2 = 40.02%, Cochran&apos;s Q test p = 0.0182). Conclusions Subsequent subgroup analyses revealed no significant moderator effects across differing experimental designs or validation strategies, indicating that the performance advantage of APMs remained robust across diverse analytical conditions.