Exploring the relationship between heavy metals and diabetic retinopathy: a machine learning modeling approach

Area of Science:

• Ophthalmology
• Environmental Health
• Data Science

Background:

• Diabetic retinopathy (DR) is a leading cause of adult blindness.
• Traditional methods show heavy metals as risk factors for DR, but complex interactions remain unclear.
• Machine learning (ML) offers advanced analytical capabilities for understanding these relationships.

Purpose of the Study:

• To develop and validate an ML model for predicting diabetic retinopathy (DR).
• To identify key environmental risk factors, particularly heavy metals, influencing DR development.
• To elucidate the complex interactions between heavy metals, patient characteristics, and DR risk.

Main Methods:

• Utilized data from the NHANES database (2003-2010) including participant characteristics and heavy metal exposure.
• Developed and compared 11 ML models for DR prediction, selecting the best via ROC analysis.
• Employed permutation feature importance (PFI), partial dependence plots (PDP), and SHapley Additive exPlanations (SHAP) for in-depth analysis.

Main Results:

• The k-nearest neighbor (KNN) model demonstrated superior DR prediction performance, achieving near 100% accuracy.
• Urinary antimony (Sb) levels were identified as the most critical heavy metal predictor of DR risk.
• Antimony (Sb) showed a significant effect on DR, with age-Sb interaction being particularly noteworthy.

Conclusions:

• Urinary Sb levels can serve as a potential non-invasive biomarker for early DR screening and intervention.
• ML models effectively explain the complex influence of heavy metal exposure on diabetic retinopathy.
• Antimony (Sb) may contribute to DR pathogenesis by mediating cellular and systemic senescence.