Risk stratification of major adverse cardiovascular events by AI-guided quantitative coronary CT according to diabetes status: the CONFIRM2 registry

Background Diabetes mellitus (DM) accelerates coronary atherosclerosis and is associated with adverse cardiovascular outcomes. Artificial intelligence–guided quantitative CT (AI-QCT) enables comprehensive plaque characterization and quantification, offering potential for refined risk stratification. We evaluated the association between AI-QCT–derived plaque burden and composition and major adverse cardiovascular events (MACE) in patients with and without DM. Purpose To assess AI-QCT–derived plaque burden for MACE risk stratification according to diabetes status. Methods CONFIRM2 is a global, multicenter registry of symptomatic patients with suspected coronary artery disease referred for coronary computed tomography angiography. In this analysis, 4,977 patients were included (750 with DM and 4,227 without DM). AI-QCT was used to quantify plaque burden and characterize plaque composition. Plaque characteristics and clinical outcomes were compared according to DM status. The primary endpoint was MACE, defined as all-cause death, myocardial infarction, late revascularization, cerebrovascular events, unstable angina, or heart failure. Results Patients with diabetes were older and exhibited a greater atherosclerotic burden compared with non-DM patients. Total plaque volume (TPV) was higher in DM (median 106.2 [IQR 36.4–288.1] vs. 50.8 [15.3–166.9] mm³), with higher non-calcified plaque (74.0 [26.2–170.7] vs. 38.8 [12.5–105.6] mm³) and calcified plaque volumes (22.0 [0.8–96.6] vs. 5.3 [0–44.4] mm³) (all p<0.001). Percentage atheroma volume was greater (3.8% [1.5–10.2] vs. 1.9% [0.6–5.8], p<0.001), and obstructive and multivessel disease were more prevalent (p<0.001). During follow-up, 234 MACE occurred, with higher events in DM compared with non-DM patients (8.9% vs. 4.0%, p<0.001). Cox regression demonstrated a stepwise increase in risk across TPV stages, with hazard ratios of 1.9, 7.5, and 10.4 in non-DM patients and 2.3, 9.5, and 9.9 in DM patients (p<0.05 for moderate and high TPV). In multivariable models, atheroma burden conferred a higher risk in patients with DM (Pinteraction=0.013). Kaplan–Meier analyses showed low event rates in minimal plaque burden and progressively higher MACE with increasing TPV; within each stage, patients with DM had higher events, most pronounced in moderate plaque burden. Addition of TPV to RF-CL categories improved discrimination for MACE (AUC 0.61 vs. 0.74, p<0.001). Conclusions Atheroma quantification by AI-QCT stratifies risk patients with and without DM. MACE increases with plaque burden, and patients with DM have higher event rates at similar TPV levels, particularly in the moderate stage. These findings support quantitative plaque assessment for improved risk stratification in DM patients.