Development and validation of a machine learning model for predicting pediatric metabolic syndrome using anthropometric and bioelectrical impedance parameters.

Abstract

Objective: Metabolic syndrome (MS) is a risk factor for cardiovascular diseases, and its prevalence is increasing among children and adolescents. This study developed a machine learning model to predict MS using anthropometric and bioelectrical impedance analysis (BIA) parameters, highlighting its ability to handle complex, nonlinear variable relationships more effectively than traditional methods such as logistic regression.

Methods: The study included 359 youths from the Korea National Health and Nutrition Examination Survey (KNHANES; 16 MS, 343 normal) and 174 youths from real-world clinical data (66 MS, 108 normal). Model 1 used anthropometric data, Model 2 used BIA parameters, and Model 3 combined both. The eXtreme Gradient Boosting trained the models, and area under the receiver operating characteristic curve (AUC) evaluated performance. Shapley value analysis was applied to assess the contribution of each parameter to the model's prediction.

Results: The AUCs for Models 1, 2, and 3 were 0.75, 0.66, and 0.90, respectively, in the KNHANES dataset, and 0.56, 0.61, and 0.74, respectively, in the real-world dataset. In pairwise comparison, Model 3 outperformed both Model 1 and Model 2 in both the KNHANES dataset (Model 1 vs. Model 3, p = 0.026; Model 2 vs. Model 3, p = 0.033) and the real-world dataset (Model 1 vs. Model 3, p = 0.035; Model 2 vs. Model 3, p = 0.008). Body fat mass was identified as the most significant contributor to Model 3.

Conclusion: The integrated model using both anthropometric and BIA parameters demonstrated strong predictability for pediatric MS, underlining its potential as an effective screening tool for MS in both clinical and general populations.