Hearing and Cognitive Function in Aged Adults: A Machine Learning and Mediation Analysis of NHANES

ABSTRACT Objective Cognitive impairment is a major public health challenge. Early identification is crucial, especially in the often‐overlooked 60–69 age group. This study developed an efficient risk identification tool. Methods Using data from the US National Health and Nutrition Examination Survey (NHANES), three machine learning algorithms screened for key predictive features. Eight models were combined with a nomogram. Mediation, restricted cubic spline, and dose–response analyses explored the hearing–cognition relationship among older adults in the United States. Results Five key features including race, education, alcohol consumption, pure tone audiometry (PTA), and systemic inflammation response index (SIRI) were identified. The random forest model performed best (area under the curve, AUC = 0.865). Education, race, and PTA were the top three contributors. A significant positive linear relationship existed between PTA and cognitive risk, with each 1 dB HL increase raising the risk by 2%. The nomogram highlighted the highest risk for those with less than ninth‐grade education, Hispanic ethnicity, and high PTA. Conclusion A exploratory risk stratification tool is developed. Education, race, and hearing level are key contributing variable. The linear PTA‐cognition relationship suggests hearing screening may aid in early risk stratification for cognitive impairment.