Interpretable Machine Learning for Predicting Metabolic Syndrome-Kidney Stone Disease Comorbidity: The Role of Dietary Micronutrients.

This study aimed to develop interpretable machine-learning models to predict the risk of metabolic syndrome-kidney stone disease (MetS-KSD) comorbidity based on dietary micronutrient intake. Using data from the National Health and Nutrition Examination Survey (NHANES) from 2007 to 2018, 54 candidate features, including dietary variables and demographic covariates, were incorporated into the analysis. Six mainstream machine-learning models, including Random Forest, XGBoost, LightGBM, k-nearest neighbors, support vector machine, and Naïve Bayes, were evaluated using a comprehensive multi-metric framework incorporating the area under the receiver operating characteristic curve (AUC-ROC), area under the precision-recall curve (AUC-PR), accuracy, F-beta score, sensitivity, and specificity. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were applied to enhance model interpretability. A total of 4936 participants were included, representing approximately 4,597,435 U.S. adults with MetS-KSD comorbidity after application of sampling weights. After variance inflation factor analysis and Boruta feature selection, 33 features were retained, including 25 dietary micronutrients and 8 demographic variables. When demographic and dietary variables were jointly modeled, Random Forest demonstrated the best performance (AUC-ROC = 0.958; AUC-PR = 0.961). When models were constructed using dietary micronutrients alone, XGBoost achieved optimal performance (AUC-ROC = 0.956; AUC-PR = 0.960). SHAP and LIME analyses identified lycopene, added vitamin B12, magnesium, dietary fiber, theobromine, and vitamin K as key contributing features, with importance rankings varying according to demographic context. Supplementary sensitivity analyses further supported model robustness and the effectiveness of SMOTE-based imbalance correction. These findings suggest that interpretable machine learning may provide a useful framework for nutritional risk stratification in MetS-KSD comorbidity, although external validation in independent populations is still required.