An Effective Machine Learning for Prioritizing Hearing Loss Screening in Older Adults: A Primary Care Approach.

BACKGROUND/

OBJECTIVES

The high prevalence of hearing impairment in older adults and the scarcity of diagnostic resources in primary health care (PHC) highlight the need for solutions based on artificial intelligence. This study aimed to develop and evaluate a decision tree model to prioritize hearing examinations in this population.

METHODS

An observational study was conducted using 60 clinical profiles. The decision tree model was developed in R, using the Recursive Partitioning and Regression Trees (rpart) package after data preprocessing that included cleaning, dichotomization, and multiple imputations by chained equations (MICE). The algorithm was trained on 70% of the sample and tested on the remaining 30%. Its performance was evaluated using standard metrics.

RESULTS

The model achieved satisfactory performance with good accuracy, sensitivity, and specificity. The area under the curve (AUC) was 0.825, outperforming a classic logistic regression model (AUC 0.700). Based on the model's key predictors, which include the presence of tinnitus, not having a partner, female sex, hypertension and/or diabetes, and age ≥ 65 years, a user-friendly, interactive web-based dashboard was developed for clinical use.

CONCLUSION

The decision tree model is a promising tool that demonstrates technical viability for prioritizing hearing loss screening. While further large-scale external validation is required, it serves as a foundational step for integrating AI-driven prioritization in primary health care.