Abstract
BACKGROUND: This study aimed to evaluate the early diagnostic value of Serum-Amyloid-A to High-Density-Lipoprotein-Cholesterol Ratio (SHR) for coronary artery disease (CAD) with clinically relevant stenoses and develop a machine learning diagnostic model based on eXtreme Gradient Boosting (XGBoost). METHODS AND RESULTS: Data from 1,108 CAD patients (with coronary luminal diameter stenosis ≥50% or evidence of functional myocardial ischemia) and 962 controls were retrospectively analyzed. Receiver operating characteristic (ROC) analysis showed SHR (area under the curve (AUC) = 0.769) outperformed C-reactive protein (CRP) (p = 0.006) and Serum amyloid A (SAA) (p < 0.001). Four XGBoost models were constructed, and the best model (CRP + SAA + SHR + 13 other variables) achieved an AUC of 0.876. SHR correlated nonlinearly with age (p < 0.001), and its diagnostic efficacy was higher in younger patients (40 years old, OR = 16.29) than in older adults (80 years old, OR = 4.37). Machine learning models can address the decline in diagnostic capability of SHR in the elderly population. CONCLUSION: SHR is a superior composite biomarker for early diagnosis of CAD with clinically relevant stenoses, outperforming CRP and SAA. Machine learning model integrating multiple indicators shows excellent diagnostic performance. Elevated SHR indicates higher CAD risk in younger individuals, providing a new strategy for early screening of CAD with clinically relevant stenoses.