Interpretable machine learning approaches for predicting prostate cancer by using multiple heavy metal exposures...

Area of Science:

• Environmental Health
• Oncology
• Computational Biology

Background:

• Environmental pollution is a significant factor in prostate cancer (PCA) development.
• Limited research exists on the association between multiple heavy metal exposures and PCA risk using machine learning.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting prostate cancer risk based on heavy metal exposure.
• To identify specific heavy metals associated with prostate cancer risk.

Main Methods:

• Utilized data from 8022 samples (2003-2018 NHANES database) including 18 blood and urinary heavy metals/minerals and 14 covariates.
• Evaluated eight machine learning models, with Random Forest (RF) showing superior performance.
• Integrated four interpretable methods to identify key risk factors.

Main Results:

• The RF model achieved 72.8% accuracy and an AUC of 0.869.
• Positive associations with PCA risk were found for blood lead (Pb), urinary cesium (Cs), and urinary antimony (Sb).
• Blood cadmium (Cd) showed a negative association; urinary Cs and Sb were identified as novel risk factors.

Conclusions:

• Machine learning models, particularly RF, can effectively predict PCA risk based on heavy metal exposure.
• Blood Pb, urinary Sb, and urinary Cs are significant contributors to PCA risk, offering potential targets for prevention and control strategies.