Predictive Modeling of Heart Failure Risk Based on Dietary Antioxidants: A Machine Learning Approach

This study investigates the relationship between dietary antioxidants and heart failure (HF) risk using nationally representative National Health and Nutrition Examination Survey data (2005-2018). It aims to identify key dietary antioxidants and develop a machine-learning-based predictive model for HF. Among 9279 participants (434 HF cases), 44 dietary antioxidant variables were extracted from two 24-h dietary recalls. Variance inflation factor filtering, SMOTE balancing, and Boruta selection were applied. Six machine learning models-random forest, LightGBM, K-nearest neighbors (KNN), Naive Bayes, support vector machine (SVM), and Xtreme Gradient Boosting (XGBoost)-were trained with and without demographic/lifestyle covariates. WQS and Qg-comp regressions quantified joint antioxidant effects. XGBoost achieved the highest accuracy (0.965 adjusted, 0.963 unadjusted), F1 score (0.971), and PR-AUC (0.993). WQS showed a 20% lower HF odds per quartile increase in overall antioxidant intake (odds ratio [OR] = 0.80, 95% CI: 0.68-0.93). SHAP analysis of the XGBoost model ranked theobromine, lycopene, caffeine, calcium, vitamin B12, vitamin C, and vitamin B1 as top contributors. Higher intake of specific dietary antioxidants-particularly vitamin B6, folate, and lycopene-is significantly associated with reduced HF risk. The XGBoost model provides a robust, interpretable tool for individual-level HF prediction based on dietary antioxidant profiles, supporting evidence-based dietary strategies for HF prevention.