BACKGROUND
Disability significantly impacts the quality of life in older adults and poses substantial challenges to healthcare systems and resource allocation in China's aging society. This phenomenon necessitates accurate predictive models of disability for early intervention and management.
OBJECTIVE
To built an accurate predictive models of disability for early intervention and management.
METHODS
Data from 2,450 elderly individuals in the 2015-2020 China Health and Retirement Longitudinal Study (CHARLS) were analyzed. The dataset was randomly split into 70% training and 30% testing sets. LASSO regression with 10-fold cross-validation identified predictive variables, which were used to develop an XGBoost model. Model performance was assessed using ROC, calibration, and clinical decision and impact curves, with SHAP values interpreting variable importance.
RESULTS
The key predictors identified were age, hand grip strength, balance, CS-5, pain, depression, cognition, respiratory function, and comorbidities. The XGBoost model achieved an AUC of 0.846 (95%CI:0.825-0.866) in training and 0.698 (95%CI:0.654-0.743) in testing. Decision and impact curves demonstrated significant clinical utility, with SHAP analysis highlighting pain, respiratory function, and age as top predictors. The SHAP summary plot illustrated the positive or negative impact of these features on disability risk. A web-based tool was developed for personalized risk assessment.
CONCLUSIONS
We developed a reliable model for predicting five-year disability risk in the Chinese elderly, integrating physical, cognitive, and psychological dimensions. This model effectively identifies high-risk individuals and helps allocate limited medical resources rationally. Future work will update the model with new CHARLS data and validate it with external datasets for broader applicability.
