Abstract:

Objective To predict vision impairment (VI) and identify its risk factors using machine learning algorithms. Methods Data from the 2016-2018 waves of the China Health and Retirement Longitudinal Study (CHARLS) were selected. Eligible participants were screened, and after data cleaning and preprocessing, 19 430 participants were ultimately included. Participants were randomly assigned in an 8:2 ratio to a training set (15 544 cases) and a validation set (3 886 cases), with VI as the outcome variable. Prediction models were constructed using Random Forest (RF), Extreme Gradient Boosting (XGBoost, XGB), CatBoost (CB), Gradient Boosting (GB), and Light Gradient Boosting Machine (LightGBM, LGBM). Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), F1 score, precision, accuracy, and other indicators. SHAP analysis was used to interpret the optimal model. Results Among the 19 430 participants, 5 026 had VI, yielding a prevalence of 25.9%. The AUC values of RF, XGB, CB, GB, and LGBM were 0.748, 0.771, 0.773, 0.778, and 0.779, respectively. Based on a comprehensive comparison of all indicators, the prediction model constructed using the GB algorithm showed the best predictive performance. SHAP analysis showed that key features such as hearing impairment, place of birth, age, education level, cognitive function, living environment, and offspring support had a significant impact on the prediction results. Conclusion A GB-based prediction model for VI in middle-aged and older adults was developed and deployed on a web platform. This model can effectively identify high-risk individuals, thereby facilitating early implementation of targeted intervention and health education strategies.

Key words: vision impairment, machine learning, prediction model, CHARLS, SHAP, older adults