文献类型：期刊文献

英文题名：Predicting Hearing Loss in the Chinese Middle-aged and Elderly: A Machine Learning Approach Using Clinical and Demographic Data

作者：Yang, Jinyuan[1]; Tan, Xin[2]; Dong, Guojie[1]; Fan, Sijing[3]; Zheng, Hongyun[3]; Cui, Yicong[4]; Fang, Wen[5]; Kong, Chui[6]

机构：[1] Chinese PLA General Hospital, Senior Department of Otolaryngology Head and Neck Surgery, Beijing, China; [2] Hunan Normal University, College of Engineering and Design, Institute of Interdisciplinary Studies, Changsha, China; [3] East China University of Science and Technology, School of Sports. Science and Engineering, Shanghai, China; [4] Tsinghua University, Division of Sports Science and Physical Education, Peking, China; [5] East China University of Science and Technology, College of Sports Science and Engineering, Shanghai, China; [6] Tsinghua University, Department of Automation, Beijing, China

年份：2025

起止页码：251

外文期刊名：2025 19th International Conference on Complex Medical Engineering, CME 2025

收录：EI(收录号：20260720056754)

语种：英文

外文关键词：Accelerated aging - Adaptive boosting - Diagnosis - Geriatrics - Health risks - Logistic regression - Neural networks - Population dynamics - Population statistics - Predictive analytics - Public health - Risk assessment - Support vector regression

摘要：Hearing impairment (HI) is an increasingly prevalent public health concern among aging populations, yet scalable and effective screening tools remain limited. Within East Asia, China bears a disproportionately high HI burden-despite comprising 87% of the regional population aged 45 and above, it accounts for 97% of reported cases. To develop a predictive framework, we analyzed data from 7,901 individuals aged 45+ in the China Health and Retirement Longitudinal Study (CHARLS). Using LASSO regression for variable selection, 22 key demographic, clinical, and behavioral features were identified. A suite of seven machine learning models-including logistic regression, support vector machine (SVM), gradient boosting machine (GBM), neural network, XGBoost, AdaBoost, and LightGBM-was trained and evaluated using a 70/30 train-test split. Model interpretability was enhanced through SHAP analysis applied to the best-performing algorithm. Notably, our results revealed significant geographic variation in HI prevalence, ranging from 2.7% in Zhejiang to 10.2% in Inner Mongolia-a 3.8-fold disparity. Among the models, GBM yielded the highest predictive accuracy (AUROC: 0.887 in training; 0.678 in testing), and SHAP analysis highlighted age, cognitive status, and depressive symptoms as leading contributors to HI risk. Furthermore, the model demonstrated clinical applicability through individualized risk assessment, as illustrated by profiling a high-risk 52 -year-old participant. In summary, our findings underscore substantial spatial and regional differences in HI distribution across China and validate the potential of a GBM-SHAP-based framework as a practical and interpretable tool for early HI risk detection. This approach supports precision public health efforts by enabling targeted screening strategies in aging and resource-constrained populations. ? 2025 IEEE.