Early Identification of Mobility Limitations in Community-Dwelling Middle-Aged and Older Adults: Development of a...

With the aging of the global population, preventing the onset of mobility limitations is considered a worldwide public health priority.

This study aimed to develop a predictive model for incident early mobility limitations (EMLs) in late middle-aged and older adults, based on a simple functional test and modifiable lifestyle factors to facilitate a home-based self-assessment of early mobility decline and promote lifestyle intervention strategies.

Our study population was community dwellers aged 45 years and above who participated in the second and fourth waves of the Guangzhou Nutrition and Health Study. The included participants were healthy, nonfrail adults reporting no limitations in activities of daily living at baseline. At the 6-year follow-up, participants with poor physical performance (walking speed <1 m/s or handgrip strength <28 kg for males and <18 kg for females) or reporting some difficulty in walking and/or climbing stairs were classified as experiencing EMLs. Least absolute shrinkage and selection operator (LASSO) was used to identify predictors from various factors, and 6 machine learning models were trained and evaluated for EML prediction, using bootstrap-based techniques to address class imbalance. Predictive ability was quantified using the area under the receiver operating characteristic curve (AUC). Variable importance analysis was used to identify key predictors.

A total of 1344 participants were included in the analysis, of which 206 (15.33%) developed EMLs after a median follow-up of 6.67 (IQR 5.91-7.26) years. Those who developed EMLs were older, had a higher BMI, a lower alternate Mediterranean diet score, and poorer performance in the sit-to-stand (STS) test, as well as lower estimated muscle power from the STS at baseline. The final models included 6 out of 9 predictors: age, sex, BMI, alternate Mediterranean diet score, STS power, and dietary calcium intake. Four machine learning models (logistic regression, LASSO, support vector machine, and neural network) achieved an acceptable AUC value (≥0.70) or slightly below in the testing dataset, with the neural network performing the best (AUC 0.70, 95% CI 0.63-0.77). Curve analysis showed positive net benefits for LASSO and logistic regression. Bootstrapping did not improve classification performance. Advanced age, lower adherence to Mediterranean diet, lower muscle power estimated from STS test, and higher BMI at baseline emerged as the most important predictors of EMLs.

This study shows that EMLs in Chinese adults aged 45 years and older can be predicted using easy-to-obtain physical performance measures, age, sex, BMI, and specific nutritional factors. The combination of prediction models and variable importance analysis provides valuable insights for the early identification and intervention of EMLs. More efforts are needed to validate our findings in external cohorts.