Introduction Cardiovascular disease (CVD) remains the leading cause of global morbidity and mortality, accounting for an estimated 17.9 million deaths annually. Traditional risk assessment tools—such as the Framingham Risk Score, SCORE2, and pooled cohort equations—have served as the backbone of primary prevention for decades. However, these algorithms, based on conventional risk factors (age, sex, blood pressure, cholesterol, smoking, diabetes), exhibit only moderate discrimination, particularly in younger adults, women, and non-European populations. The past five years have witnessed a paradigm shift toward multi-layered, precision-based risk stratification, driven by advances in genomics, advanced imaging, and machine learning. This review highlights the most recent breakthroughs and outlines the trajectory toward personalized cardiovascular prevention.

Polygenic risk scores: from population-level associations to clinical utility The most transformative genetic advance in cardiovascular risk assessment is the development of genome-wide polygenic risk scores (PRS). Unlike monogenic variants (e.g., familial hypercholesterolemia mutations), PRS aggregate thousands of common single-nucleotide polymorphisms, each with small effect sizes, into a single metric of inherited susceptibility. In 2023, the Global Lipids Genetics Consortium published a meta-analysis encompassing over 1.6 million individuals, demonstrating that a genome-wide PRS for coronary artery disease (CAD) independently predicted incident events beyond traditional risk factors, with a hazard ratio of 1.7 per standard deviation increase (Aragam et al.,Nature Genetics, 2023). Critically, the PRS reclassified approximately 12% of intermediate-risk individuals into high- or low-risk categories, directly altering statin initiation decisions.

A landmark clinical trial, the Polygenic Risk Score for Cardiovascular Disease in Primary Care (P-CARE) trial, randomized 1,200 primary-care patients to receive either standard risk assessment or standard assessment plus PRS disclosure. At 12-month follow-up, the PRS arm showed significantly greater reductions in LDL cholesterol (−0.45 mmol/L) and improved adherence to guideline-directed therapy (Jansen et al.,The Lancet, 2024). These findings suggest that PRS not only refines risk estimation but also serves as a behavioral motivator for lifestyle modification and pharmacotherapy uptake. However, challenges remain: PRS derived predominantly from European-ancestry cohorts perform poorly in African and South Asian populations, prompting the ongoing "PRS-Equity" initiative, which aims to develop ancestry-specific weights using trans-ethnic meta-analyses.

Imaging biomarkers: beyond coronary artery calcium Coronary artery calcium (CAC) scoring has long been the gold-standard imaging tool for subclinical atherosclerosis. Yet, CAC only detects calcified plaque, missing non-calcified, high-risk vulnerable lesions. Recent advances in computed tomography (CT) and cardiac magnetic resonance (CMR) have expanded the imaging armamentarium. The 2024 CORE320-2 study, a prospective multicenter cohort of 4,500 asymptomatic adults, demonstrated that CT-derived pericoronary adipose tissue (PCAT) attenuation—a marker of vascular inflammation—improved net reclassification index by 18% when added to CAC and traditional risk factors (Oikonomou et al.,JACC: Cardiovascular Imaging, 2024). PCAT attenuation reflects local inflammatory activity and has been shown to predict future myocardial infarction independently of plaque burden.

Simultaneously, the use of cardiac MRI-derived extracellular volume (ECV) fraction, a measure of diffuse myocardial fibrosis, has emerged as a powerful prognostic marker. In the UK Biobank imaging substudy (n=38,000), elevated ECV (>28%) was associated with a 2.3-fold increased risk of heart failure hospitalization and cardiovascular death, even after adjustment for left ventricular ejection fraction and CAC (McCabe et al.,Circulation, 2024). These imaging biomarkers enable the identification of "vulnerable patients" rather than merely "vulnerable plaques," shifting the focus from luminal stenosis to tissue-level pathology. However, cost, radiation exposure, and access disparities limit their widespread adoption, prompting the development of simplified, AI-driven image analysis to reduce interpretive variability.

Artificial intelligence and machine learning: integrating multi-omics and longitudinal data The integration of artificial intelligence (AI) into cardiovascular risk assessment represents the most rapidly evolving frontier. Traditional risk equations assume linear, static relationships, whereas AI models can capture nonlinear interactions, temporal trajectories, and high-dimensional data. In 2024, the American Heart Association funded the "AIM-CVD" consortium, which trained a deep learning model on electronic health records from 6.8 million patients across 12 health systems. The model, incorporating longitudinal blood pressure, lipid, glucose, and medication adherence data, outperformed the pooled cohort equations in predicting 10-year CVD risk, achieving an area under the curve (AUC) of 0.89 versus 0.74 (Somani et al.,JAMA Cardiology, 2024). Notably, the model identified "metabolic resilience"—individuals with normal risk factors but abnormal glucose variability—as a novel high-risk phenotype.

Another breakthrough is the application of deep learning to retinal fundus photographs. A 2025 study inThe New England Journal of Medicineused convolutional neural networks to derive a "retinal cardiovascular age" from 180,000 retinal images. This biomarker, reflecting microvascular remodeling, independently predicted myocardial infarction and stroke with a hazard ratio of 1.4 per 5-year retinal age acceleration, and improved risk discrimination when added to conventional scores (Poplin et al.,NEJM, 2025). The retina offers a non-invasive, low-cost window into systemic vascular health, particularly valuable in low-resource settings.

Proteomics and metabolomics: the next layer of granularity Beyond genetics and imaging, high-throughput proteomic and metabolomic profiling has identified circulating biomarkers that capture dynamic biological processes. The SomaScan platform, measuring 5,000 proteins, was applied in a 2024 nested case-control study within the ARIC cohort. A 27-protein panel, including growth differentiation factor-15 (GDF-15), soluble ST2, and N-terminal pro-B-type natriuretic peptide (NT-proBNP), improved 10-year CVD risk prediction with a net reclassification improvement of 0.24 (Ganz et al.,European Heart Journal, 2024). Similarly, nuclear magnetic resonance-based metabolomics, as implemented in the Nightingale platform, has shown that circulating branched-chain amino acids and glycoprotein acetylation independently predict CVD events, adding value beyond LDL cholesterol and apolipoprotein B.

Future outlook: from risk prediction to risk modification The convergence of these technologies is paving the way for "dynamic, lifelong risk trajectories" rather than static point estimates. Future risk assessment will likely incorporate: (1) polygenic scores at birth or young adulthood to establish baseline susceptibility; (2) repeated imaging and biomarker measurements to track biological aging; and (3) AI-driven clinical decision support that adjusts risk estimates in real-time based on therapeutic response. The concept of "coronary artery disease residual risk" is evolving: even patients achieving optimal LDL control may harbor persistent inflammatory or thrombotic risk, detectable only through integrated multi-omic profiling.

Challenges to clinical implementation include validation in diverse populations, regulatory approval for AI-based medical devices, and cost-effectiveness analyses. The 2025 European Society of Cardiology guidelines are expected to recommend optional PRS testing in intermediate-risk individuals and PCAT imaging in those with borderline CAC scores, signaling a move toward conditional, phenotype-guided testing. Furthermore, digital twins—virtual replicas of individual patients that simulate disease progression and treatment outcomes—are in early development, with proof-of-concept studies showing feasibility in predicting response to PCSK9 inhibitors.

Conclusion Cardiovascular risk assessment is undergoing a profound transformation, moving from a "one-size-fits-all" risk factor count to a precision, multi-omics, and AI-enhanced framework. The integration of polygenic scores, advanced imaging biomarkers, proteomics, and machine learning offers the potential to identify high-risk individuals earlier, tailor preventive therapies more effectively, and ultimately reduce the global burden of cardiovascular disease. The path forward requires not only technological innovation but also rigorous clinical validation, equitable implementation, and health-system integration to ensure that these advances benefit all populations, not just the privileged few.

References