An explainable machine learning method for identifying key atherogenic lipid biomarkers in abdominal obesity among
Xi Meng1,2, Shuaijing Zhang1,2, Yunkai Lin1,2
1NMPA Key Laboratory for Safety Evaluation of Cosmetics, Guangdong Provincial Key Laboratory of Tropical Disease Research, School of Public Health, Southern Medical University, Guangzhou, 510515, China.
Background:
Abdominal obesity (AO) significantly contributes to cardiometabolic diseases and poses an increasing public health challenge. Composite lipid-derived indices have been proposed as simple indicators of atherogenic dyslipidemia, but their relative importance for identifying AO at the population level remains incompletely understood. This study aimed to evaluate the link between various atherogenic lipid indices and AO, identifying key lipid-related predictors through interpretable machine learning methods.
Methods:
This cross-sectional study analyzed 5,612 adults from a population-based survey in Guangdong Province, China. Atherogenic lipid indices encompass the atherogenic index of plasma (AIP), non-high-density lipoprotein cholesterol (non-HDL-C), atherogenic coefficient (AC), Castelli risk indices I and II (CRI-I and CRI-II), lipid composite index (LCI), and the triglyceride-to-high-density lipoprotein cholesterol ratio (TG/HDL-C). Feature selection was conducted using Boruta and Least Absolute Shrinkage and Selection Operator (LASSO) methods prior to machine learning modeling. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, and F1 score. SHapley Additive exPlanations (SHAP) analysis was applied to quantify feature importance.
Results:
Individuals with AO were generally older and more likely to exhibit lower education levels, reduced physical activity, cardiometabolic comorbidities, and unfavorable metabolic profiles. Feature selection identified 15 key predictors. Logistic regression demonstrated the most stable performance (training AUC: 0.767; testing AUC: 0.768; accuracy: 0.712; F1 score: 0.667) with good calibration and clinical utility. SHAP analysis consistently identified AIP, sex, CRI-II, uric acid, and diastolic blood pressure as the most influential predictors of AO.
Conclusion:
Composite lipid indices, particularly AIP and CRI-II, are strongly associated with AO and may serve as practical indicators for identifying individuals at elevated metabolic risk. Because these indices are derived from routinely measured lipid parameters, they may support scalable approaches for metabolic risk screening and monitoring in both clinical and community health settings.