Predictive value of obesity indices for carotid atherosclerosis: An interpretable machine learning study

Objective: To investigate the associations between commonly used obesity indices and the risk of carotid atherosclerosis (CAS), and to provide evidence for the early identification of individuals at high risk.

Methods:  A cross-sectional survey was conducted between August 2021 and August 2022, covering residents aged 18 and above across 25 counties in 14 cities of Henan Province. A total of 36,376 participants were included. All subjects underwent standardized assessments, including questionnaires, physical examinations, blood sample collection, biochemical assays, and carotid ultrasonography. Six obesity indices—Body Mass Index (BMI), Waist Circumference (WC), Waist-to-Hip Ratio (WHR), Waist-to-Height Ratio (WHtR), A Body Shape Index (ABSI), and Body Roundness Index (BRI)—were selected as predictors. The Boruta algorithm was employed for optimal feature selection. The dataset was divided into a training set (n=25,463) and a testing set (n=10,913) using stratified sampling, with parameter optimization performed via ten-fold cross-validation. Restricted Cubic Spline (RCS) analysis was used to model the non-linear relationship between obesity indices and CAS. Four machine learning algorithms—Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)—were applied to evaluate the predictive value of each index. Model performance was assessed using the Area Under the Receiver Operating Characteristic Curve (AUC). The SHapley Additive exPlanations (SHAP) method was utilized to interpret feature contributions.

Results: The prevalence of CAS in the study population was 49.6%. All obesity indices were significantly associated with the risk of CAS (P<0.001). Among the models tested, XGBoost demonstrated the superior performance in identifying CAS risk. Of the six obesity indices, the combination of WHtR and the optimal feature subset yielded the best performance, achieving an AUC of 0.8824 in the XGBoost model. SHAP analysis indicated that age, gender, history of hypertension, obesity indices, and physical activity levels were the most significant contributors to the prediction.

Conclusions: All six obesity indices (BMI, WC, WHR, WHtR, ABSI, and BRI) demonstrated varying capacities to identify CAS risk, with WHtR showing the strongest predictive performance. WHtR may serve as a simple, low-cost, and efficient screening tool for identifying individuals at high risk of carotid atherosclerosis in routine clinical practice and population-based screening.

Zhang is a master’s student in Social Medicine and Health Service Management at the School of Public Health, Zhengzhou University. Her research interests focus on the primary prevention of cardiovascular diseases and the social determinants of health. During my graduate studies, She has participated in the preparation of academic monographs and have authored or co-authored several research articles in related fields. In addition, She has actively attended leading academic conferences in cardiovascular and public health, which has strengthened her research skills and contributed to her solid academic record.