👤 Paolo Alberto Del Sole

Conventional biomarkers such as low-density lipoprotein (LDL) and high-density lipoprotein may fail to identify patients' risk for significant coronary artery disease (CAD). This study evaluates the associations between multiple biomarkers and different CAD phenotypes, exploring a machine-learning biomarker-based patient clustering. We included 787 patients on primary prevention from the prospective ACTION registry (January 2024 to June 2025). All patients underwent coronary CTA and simultaneous biomarker testing, including LDL, high-density lipoprotein, triglyceride, apolipoprotein A-1, apolipoprotein B, lipoprotein(a) [Lp(a)], glycated hemoglobin (HbA1c), and high-sensitivity C-reactive protein. Of 382 patients (48.5%) with coronary artery calcium = 0, 42 (11%) had coronary plaque. These patients showed higher Lp(a) vs those without plaque (16.5 vs 11.5, P = .030), despite comparable SCORE2 risk (3.5% vs 3.0%, P = .284). Three biomarker-defined groups were identified after a machine learning unsupervised clustering: Cluster 1 had a favorable lipid profile with the lowest prevalence of CAD-Reporting and Data System (RADS) ≥ 3 (9.9%). Cluster 2 and 3, despite their significant intercluster differences in terms of Lp(a), LDL, and HbA1c levels, both showed a significantly higher prevalence of CAD-RADS ≥ 3 compared to cluster 1 (respectively 21.8% and 17.9%; vs cluster 1, P = .001). High-risk biomarker signatures were significantly associated to the prevalence of CAD-RADS ≥ 3, independently from the baseline SCORE2 (adjusted odds ratio 2.25; 95% confidence interval 1.32-3.82). Distinct biomarker signatures associate with distinct CAD prevalence and severity that conventional lipid markers fail to distinguish. Lp(a) appears relevant for early plaque detection in coronary artery calcium = 0 patients. A comprehensive biomarker evaluation may help identifying high-risk subgroups overlooked by a conventional assessment. Show less