From Nutritional Epidemiology to Clinical Translation: Machine Learning Uncovers Antioxidant-Driven Prevention...

The aim of this study was to construct a machine learning (ML) model to predict the effect of dietary antioxidants on cardiovascular-arthritis comorbidity.

In this study, 44 dietary antioxidants were selected based on the National Health and Nutrition Examination Survey data from 2007 to 2010 and from 2017 to 2018, and demographic covariates such as gender and age were included for analysis. In addition, 10 mainstream ML models were investigated for the evaluation, and a comprehensive evaluation system of the multi-indicator empowerment algorithms was constructed to comprehensively measure the model performance. To further enhance the model interpretability, SHapley Additive exPlanations (SHAP) values and Local Interpretable Model-agnostic Explanation (LIME) methods were introduced to deeply analyze the prediction mechanism.

A total of 8046 participants were included in this study, of whom 380 had cardiovascular disease and arthritis comorbidities. After multiple covariates were eliminated, 34 indicators of nutritional antioxidant intake and 11 demographic baseline characteristics were selected as key predictors. The multicriteria-based assessment system demonstrated excellent performance of the logistic regression machine model. It performed optimally on the validation set with an area under the receiver operating characteristic curve of 0.871. Notably, the study of SHAP and LIME algorithms revealed the opposite biological effects of total and single flavonoid intake as well as the heterogeneity of dietary antioxidants in different age-sex characterized populations.

This study suggests that future strategies should consider antioxidant types and individual traits, promoting diverse natural foods over single supplements to advance precision nutrition. Antioxid. Redox Signal. 44, 251-270.