OCTAおよびAIによる網膜微小血管バイオマーカーを用いた心血管リスク層別化：洞察

Cardiovascular disease (CVD) remains a major global burden, and retinal microvascular imaging offers a noninvasive means to capture systemic microvascular status. Evidence from prospective cohorts shows that arteriolar narrowing and venular widening predict incident coronary heart disease, with pooled adjusted hazard ratios approximately 1.20 (95% CI 1.13-1.27). Network-level metrics, including reduced fractal dimension and increased tortuosity, further reflect microvascular remodeling and have been associated with higher risks of CVD and heart failure. In diabetes, both the presence and severity of diabetic retinopathy, as well as an AI-derived retinal age gap, correlate with elevated rates of cardiovascular events, kidney disease progression, and all-cause mortality. OCTA studies demonstrate that vessel-density loss and foveal avascular zone alterations occur in hypertension, coronary disease, heart failure, and stroke; longitudinal data indicate that sustained reductions in vessel density of several percentage points are associated with increased major adverse cardiovascular events. AI models trained on retinal images, particularly when integrated with clinical variables, achieve discrimination comparable to or exceeding traditional risk scores, with reported AUC improvements of up to ∼0.07. Despite these advances, most studies remain cross-sectional, OCTA metrics vary significantly across devices, and AI models often lack external validation and explicit assessment of incremental predictive value. Priority areas include prospective multi-ethnic cohorts with standardized imaging protocols, harmonized OCTA acquisition and analysis, and interpretable, externally validated AI systems capable of demonstrating measurable clinical benefit before retinal biomarkers can be incorporated into routine cardiovascular risk stratification.