Abstract
High-density lipoprotein (HDL)-related inflammatory ratios (monocyte-to-HDL cholesterol ratio [MHR], lymphocyte-to-HDL cholesterol ratio, neutrophil-to-HDL cholesterol ratio [NHR], platelet-to-HDL cholesterol ratio) represent composite biomarkers integrating lipid metabolism and inflammatory pathways. We developed machine learning models to evaluate their utility in coronary heart disease (CHD) classification using a large population-based dataset. We analyzed data from the National Health and Nutrition Examination Survey 2009 to 2020, including 14,745 US adults aged ≥20 years (mean age 51.8 ± 17.6 years). Self-reported CHD diagnosis was the outcome variable. Machine learning models (eXtreme gradient boosting, random forest, logistic regression) were developed to evaluate cross-sectional associations between HDL-related inflammatory ratios and CHD prevalence. Self-reported CHD prevalence was 5.7% (n = 840). All HDL-related inflammatory ratios were significantly elevated in CHD patients: MHR (0.54 ± 0.35 vs 0.42 ± 0.23, P < .001), lymphocyte-to-HDL cholesterol ratio (2.05 ± 3.12 vs 1.55 ± 1.02, P < .001), and NHR (4.06 ± 2.89 vs 3.11 ± 1.77, P < .001). eXtreme gradient boosting demonstrated optimal performance with an area under the receiver operating characteristic curve of 0.8892, accuracy of 96.55%, and precision of 86.00%. SHapley Additive exPlanations analysis identified age as the most important predictor, with MHR and NHR ranking among the top 5 features. Machine learning models incorporating HDL-related inflammatory biomarkers achieved high discrimination (area under the receiver operating characteristic curve = 0.8892) for identifying cross-sectional associations with CHD prevalence. These findings reveal significant cross-sectional associations between HDL-related inflammatory ratios and CHD prevalence, rather than predictive relationships for incident events. These readily available biomarkers from routine blood tests provide substantial value for cardiovascular risk stratification. Prospective validation is warranted to establish their utility for predicting incident CHD events.