Identification of high-risk phenotypes through unsupervised clustering of clinical and CT-calcium score imaging features

Accurate risk assessment for atherosclerosis-related Major Adverse Cardiac Events (MACE) is crucial for timely interventions and improved patient outcomes. Traditional time-to-event models struggle with the intricate relationships among interrelated risk factors, and established risk assessment tools, like the Agatston score, suffer from poor discrimination in intermediate risk groups. In this study, we employ unsupervised clustering techniques to derive subphenotypes integrating clinical and CT calcium score imaging biomarkers, specifically coronary artery calcium and epicardial adipose tissue (EAT)-related features. Notably, race significantly impacts subphenotype EAT characteristics, and the features that relate to increasing MACE risk. Extraordinarily, our unsupervised analysis without knowledge of outcomes (p-values: 0.006, 0.000, 0.000, 0.020, 0.000, 0.000) outperforms traditional Agatston Score groups in risk-discrimination (p-values: 0.359, 0.523, 0.170, 0.439, 0.470), resulting in excellent separation for time-to-MACE outcomes. This suggests that our phenotypes are capturing prognostic, complex relationships between the clinical information and imaging biomarkers, which comprehensively capture atherosclerosis risk.