Machine learning algorithms for identifying ADL limitation in Chinese older adults with chronic conditions: development and internal-external validation

Limitation in activities of daily living (ADL) represents a major public health concern among older adults, particularly those with chronic conditions in China. This study aimed to develop and internally and externally validate interpretable machine learning (ML) models for identifying ADL limitation in this population. Data from 7073 older adults with chronic conditions were obtained from the China Health and Retirement Longitudinal Study (CHARLS) and randomly allocated into a training set and an internal test set. An external validation set comprising 667 individuals from a community in Xi’an was additionally included. Six ML models were developed and evaluated. Model performance was evaluated primarily using the area under the receiver operating characteristic curve (AUC). Calibration was assessed using calibration plots, calibration intercept, calibration slope, and Brier score. The SHapley Additive exPlanations (SHAP) method was applied to interpret the optimal model at both global and individual levels. All models demonstrated good discrimination (AUC > 0.75). The extreme gradient boosting (XGB) model achieved the best performance and showed generally acceptable calibration across the three datasets. SHAP analysis identified the relative contribution of features to model output. The five most important features were depressive symptoms (0.351), walking speed (0.318), cognitive function (0.315), chronic pain (0.252), and grip strength (0.249). Higher depressive-symptom scores and the presence of chronic pain tended to contribute toward classification as ADL limitation, whereas faster walking speed, better cognitive function, and greater grip strength tended to contribute toward classification as no ADL limitation. The XGB model demonstrated strong discrimination for identifying ADL limitation across the study samples, including a geographically and temporally distinct external validation sample. The identified features may help inform screening and guide more comprehensive functional assessment among older adults with chronic conditions.