AUTHOR=Nabavi Ali , Kashkooli Mohammad , Nabavizadeh Sara Sadat , Safari Farimah 

TITLE=Heavy metal biomarkers and their impact on hearing loss risk: a machine learning framework analysis

JOURNAL=Frontiers in Public Health

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2025.1471490

DOI=10.3389/fpubh.2025.1471490

ISSN=2296-2565

ABSTRACT=BackgroundExposure to heavy metals has been implicated in adverse auditory health outcomes, yet the precise relationships between heavy metal biomarkers and hearing status remain underexplored. This study leverages a machine learning framework to investigate these associations, offering a novel approach to understanding the interplay between environmental exposures and hearing loss.MethodsWe conducted a retrospective cross-sectional analysis using data from the 2012–2018 National Health and Nutrition Examination Survey (NHANES), encompassing 2,772 participants after applying exclusion criteria. Demographic, clinical, and heavy metal biomarker data (e.g., blood lead and cadmium levels) were analyzed as features, with hearing loss status—defined as a pure-tone average threshold exceeding 25 dB HL across 500, 1,000, 2000, and 4,000 Hz in the better ear—serving as the binary outcome. Multiple machine learning algorithms, including Random Forest, XGBoost, Gradient Boosting, Logistic Regression, CatBoost, and MLP, were optimized and evaluated. Model performance was assessed using accuracy, area under the curve (AUC), sensitivity, and specificity, while SHAP (SHapley Additive exPlanations) elucidated feature contributions.ResultsThe CatBoost model demonstrated the strongest performance, achieving an accuracy of 74.9% and an AUC of 0.792 on test data. Age, education level, gender, and blood levels of lead and cadmium emerged as the most significant features associated with hearing loss, as determined by SHAP analysis. These findings highlight key correlates of hearing impairment within the study population.ConclusionThis study underscores the utility of a machine learning framework in identifying associations between heavy metal biomarkers and hearing loss in a nationally representative sample. While not designed to forecast hearing loss over time, our findings suggest potential clinical relevance for identifying individuals with elevated heavy metal exposure who may warrant further audiometric evaluation. This work lays a foundation for future longitudinal studies to explore these relationships more comprehensively.