The relationship between heavy metals and metabolic syndrome using machine learning

Area of Science:

• Environmental Health
• Toxicology
• Computational Biology

Background:

• Heavy metal exposure is a known risk factor for metabolic syndrome (MetS).
• Understanding the specific associations between heavy metals and MetS is crucial for public health.
• Machine learning (ML) offers novel approaches to analyze complex health data.

Purpose of the Study:

• To assess the associations between heavy metal exposure levels and metabolic syndrome (MetS) using machine learning.
• To identify key heavy metals contributing to the risk of developing MetS.
• To develop and validate a predictive model for MetS based on heavy metal exposure.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (2003-2018).
• Employed Lasso regression for variable selection and 9 ML models for association analysis with 5-fold cross-validation.
• Applied SHapley Additive exPlanations (SHAP) to interpret the AdaBoost model predictions.

Main Results:

• The AdaBoost model demonstrated superior performance with an AUC of 0.807, accuracy of 0.720, and sensitivity of 0.792.
• Higher levels of cadmium, cesium, and increased BMI, female sex, and age were associated with increased MetS probability.
• Lower levels of cobalt and molybdenum were linked to a decreased probability of MetS.

Conclusions:

• The AdaBoost ML model effectively identified correlations between heavy metal exposure and MetS.
• Interpretable methods highlighted cadmium, molybdenum, cobalt, cesium, uranium, and barium as significant factors in MetS prediction.
• This study underscores the utility of ML in understanding environmental risk factors for metabolic diseases.