Machine learning-based prediction model for fall risk among individuals with arthritis in China: an analysis using the China Health and Retirement Longitudinal Study (CHARLS) database.

backgroundFalling has become a global public health problem. Individuals with arthritis have a higher risk of falling because of joint pain, poor muscle strength, and decreased proprioception; however, there is no all-inclusive risk assessment model for fall risks in the Chinese population based on multi-health data.

methodsThis study utilized data from the China Health and Retirement Longitudinal Study (CHARLS) to develop a machine learning (ML)-based predictive model for fall risk among adults with arthritis. Multidimensional health data were integrated to identify key risk factors. Nine ML models, including XGBoost, were employed to assess predictive performance. The dataset was randomly divided into a training set (70%) and a test set (30%). Model performance was evaluated using AUC, calibration curves, and decision curve analysis, among other metrics.

resultsAmong the 4,536 participants, 927 fall incidents were recorded, with a fall incidence rate of 20.44%. LASSO regression identified six key risk factors: age, sleep time, diabetes, depression, grip strength, and ADL score. The XGBoost model demonstrated the best performance, with AUC values of 0.746 (95% CI: 0.726-0.763) in the training set and 0.734 (95% CI: 0.702-0.768) in the test set. Calibration curves showed good agreement between predicted and observed probabilities, and decision curve analysis indicated significant clinical benefits.

conclusionThe ML-based predictive model developed in this study effectively identifies fall risk among adults with arthritis, providing a valuable tool for clinical management and public health strategies. Further validation in external datasets is needed to confirm the model's generalizability and clinical utility.