A fall risk prediction model based on the CHARLS database for older individuals in China

Falls represent the second leading cause of injury-related mortality among older adults globally. The occurrence of falls is the consequence of the interaction of numerous complex risk factors. The objective of this study was to develop a validated fall risk prediction model for the Chinese older individuals.

The study used data from the China Health and Retirement Longitudinal Study (CHARLS), a dataset representative of the Chinese population. Thirty-eight indicators including biological factors, behavioral factors and health status were analyzed in this study. The study cohort was randomly divided into the training set (70%) and the validation set (30%). Variables were screened using LASSO regression analysis, the best predictive model based on 10-fold cross-validation, logistic regression model was applied to explore the correlates of fall risk in the older individuals, a nomogram was constructed to develop the predictive model, calibration curves were applied to assess the accuracy of the nomogram model, and predictive performance was assessed by area under the receiver operating characteristic curve and decision curve analysis.

A total of 4,913 cases from the 2015 CHARLS database for people aged 60 years and older were ultimately included, and a total of 1,082 (22.02%) of the older individuals had experienced a fall within two years. Multivariate logistic regression analysis showed that Sleeping time, Hearing, Grip strength, ADL score, Cognition, Depression, Health, KD, and Pain DRUG were predictors of fall risk in the older individuals. These factors were used to construct nomogram models that showed good agreement and accuracy. The AUC value for the predictive model was 0.644 (95% CI = 0.621-0.666), with a specificity of 0.695 and a sensitivity of 0.522. For the internal validation set, the AUC value was 0.644 (95% CI = 0.611-0.678), with a specificity of 0.629 and a sensitivity of 0.577. The Hosmer-Lemeshow test value of the model for the training set is p = 0.9368 and for the validation set is p = 0.8545 (both > 0.05). The calibration curves show a more significant agreement between the nomogram model and the actual observations. The ROC and DCA indicate a better predictive performance of the nomogram.

The comprehensive nomogram constructed in this study is a promising and convenient tool for assessing the risk of falls in the Chinese older individuals and to help older adults understand the risk level of falls, avoid and eliminate modifiable risk factors, and reduce the incidence of falls.

Not applicable.