Application of machine learning for sex-specific risk prediction of coronary artery calcification

Sex-related differences in coronary artery calcium (CAC) burden and risk are well known but often ignored in traditional models. We aimed to determine whether sex-specific machine learning (ML) improves CAC risk prediction using routine clinical data.

In this cross-sectional study, adults (n = 446) referred for coronary computed tomography angiography were included. The data were subdivided into training (127 men, 185 women) and test (55 men, 79 women) sets. Eight ML classifiers were trained on the overall training dataset and separately for men and women to predict CAC presence (0 vs. >0), severity (0, 1-99, ≥100), and distribution (0, 1-vessel, ≥2-vessel). Algorithm performance was assessed on the unseen test set. CAC was present in 44.2 % of participants, with 17.4 % having CAC≥100 and 23.3 % having multivessel involvement (≥2 vessels). Sex-specific ML improved CAC prediction performance. For CAC presence, top-performing models achieved AUC/F1-scores of 0.690/0.618 overall, 0.799/0.776 in men, and 0.679/0.594 in women. For CAC severity, results were 0.692/0.532 overall, 0.731/0.584 in men, and 0.748/0.562 in women. For CAC distribution, AUC/F1-scores were 0.694/0.596 overall, 0.696/0.551 in men, and 0.738/0.615 in women. Threshold analysis in binary CAC classification (0 vs. >0) showed reasonable rule-out performance in both sexes, while rule-in performance was acceptable only in men. Feature importance rankings showed significant differences between sexes, reflecting sex-specific learning patterns.

Sex-specific ML improved CAC risk prediction, highlighting algorithmic sex-related differences in cardiovascular risk assessment. These findings support the development of sex-specific cardiovascular risk equations to enhance personalized care and treatment.