Development and temporal validation of a prediction model for cognitive impairment in older adults with hearing loss...

Older adults with hearing loss are at increased risk of cognitive impairment, yet tailored risk stratification tools remain limited.

In this cross-sectional study, 524 adults aged ≥60 years with hearing loss were enrolled between June 2023 and September 2024. Participants were divided by enrollment period into a development cohort (June 2023 to May 2024, n = 367) and a temporally independent validation cohort (June 2024 to September 2024, n = 157). Cognitive function was assessed using the Montreal Cognitive Assessment, with a score <26 indicating cognitive impairment after correction for education. Candidate predictors were selected based on the Integrated Framework for Population Health Risk Management. Feature selection was performed using LASSO regression, and six machine-learning models were developed and compared. Model performance was evaluated using AUC, F1 score, sensitivity, specificity, Youden index, PPV, NPV, calibration curves, and decision curve analysis. The final candidate model was optimized using grid search with five-fold cross-validation, and SHAP was applied for interpretability.

The prevalence of cognitive impairment was 40.8%. LASSO identified five key predictors: age, pure-tone average, depression, hearing-aid use, and social activities. Multivariable analysis showed that depression, older age, and higher pure-tone average were associated with a higher likelihood of cognitive impairment, whereas hearing-aid use and participation in social activities were protective factors. Among the six models, Random Forest showed the best overall performance in the validation cohort. After hyperparameter tuning, the optimized Random Forest model achieved an AUC of 0.952 in the training cohort and 0.871 in the validation cohort. In the validation cohort, the F1 score, sensitivity, Youden index, and NPV increased to 0.737, 0.779, 0.583, and 0.863, respectively. SHAP analysis indicated that pure-tone average, age, and social activities were the most influential predictors in the optimized model.

The optimized five-variable Random Forest model demonstrated good predictive performance for cognitive impairment in older adults with hearing loss. This low-cost and interpretable tool may support early screening, risk stratification, and targeted intervention in clinical and community settings.