Associations between urinary and blood heavy metal exposure and heart failure in elderly adults: Insights from an interpretable machine learning model based on NHANES (2003–2020)

Background

The relationship between heavy metal exposure and heart failure is complex and poorly understood. This study employs machine learning techniques to model these associations in a population aged 50 years and older from the National Health and Nutrition Examination Survey (NHANES). Our findings emphasize the need for continued investigation into the mechanisms of these associations and highlight the importance of monitoring and regulatory measures to mitigate heavy metal exposure in populations at risk.

Methods

Five machine learning models were evaluated, with Gradient Boosting Decision Trees (GBDT) selected as the optimal model based on accuracy, interpretability, and ability to capture nonlinear relationships. Model performance was assessed through various metrics, and interpretability was enhanced using SHAP (SHapley Additive exPlanations), permuted Feature Importance, Individual Conditional Expectation (ICE), and Partial Dependence Plots (PDP).

Results

The GBDT model achieved an accuracy of 0.78, with a sensitivity of 0.93 and an AUC of 0.92. Our analysis revealed that higher levels of urinary iodine, blood cadmium, urinary cobalt, urinary tungsten, and urinary arsenic acid were significantly associated with heart failure. Synergistic effects involving age and body mass index (BMI) were also observed, further strengthening these associations.