Artificial intelligence–derived epicardial fat predicts major adverse cardiovascular events in suspected CAD

Background and Purpose Epicardial adipose tissue (EAT) is a metabolically active fat depot associated with coronary artery disease (CAD) and adverse cardiovascular outcomes, suggesting its role as a potential risk biomarker. However, clinical implementation has been limited by the challenges in standardized and reproducible quantification. We therefore developed and validated an artificial intelligence (AI)–enabled approach for automated EAT measurement from coronary CT and evaluated its prognostic value in patients with suspected CAD. Methods This multicenter study was conducted within the APOLLO cohort, comprising 5000 Asian Admixture patients undergoing coronary CT across three hospitals in Singapore. For this analysis, 947 patients were included. EAT volume was quantified using a validated deep learning algorithm and indexed to body surface area to derive the EAT index (mL/m2). The primary endpoint was major adverse cardiovascular events (MACE), defined as cardiovascular death, myocardial infarction, stroke, acute coronary syndrome, or coronary revascularization. Associations between EAT index and outcomes were assessed using Cox proportional hazards models. Binary logistic regression estimated hazard ratios for MACE. Event-free survival stratified by EAT index was evaluated using Kaplan–Meier analysis. Results Over a median follow-up of 111 months, 113 patients (11.9%) experienced MACE. Patients with events had a higher EAT index than those without (73±26 vs. 65±25 mL/m2, P<0.01). In multivariable Cox analysis, EAT index was independently associated with MACE (HR 1.011 per 1 mL/m2 increase; 95% CI 1.004–1.018; P=0.002). Binary logistic regression showed an approximately twofold higher risk of MACE with elevated EAT index (HR 1.86; 95% CI 1.25–2.76; P=0.002). Kaplan–Meier analysis further demonstrated significantly higher incidence of MACE with elevated EAT index (log-rank P=0.0013). Conclusion AI-derived EAT index independently predicts long-term MACE in patients with suspected CAD. Automated EAT quantification from coronary CT provides incremental prognostic value and may improve clinical risk stratification.