Predicting cognitive function based on physical performance: findings from the China Health and Retirement Longitudinal Study

Background Physical performance tests are simple means of predicting an individual's risk of cognitive decline. Aims This study aimed to assess the predictive value of physical performance tests and develop predictive models for cognitive function. Methods Cognitive function was tested biennially and calculated for mental intactness, episodic memory, and global cognition. Using a generalized estimating equation (GEE), we examined each baseline physical performance test as a predictor of cognitive decline. Using a multivariate linear regression model (MLRM), we developed predictive models for cognitive function. Bland-Altman analysis was performed to analyze the agreement between estimated and measured cognition. We validated the predictive model internally with 1000 bootstrap resamples. Results Better physical performance test results, except for standing balance, were associated with a slower cognitive decline over time and better cognitive function at follow-up. Regarding the predictive models, all physical performance tests were included in men; only five chair stands test was included in women. Bland-Altman analysis showed that measured cognition was equivalent to estimated cognition in men (mean bias, 0; 95% limits of agreement, - 8.56 to 8.56) and women (mean bias, 0; 95% limits of agreement - 8.79 to 8.7). Bootstrap analysis showed that predictors were selected in 78.4-100% for men and 64.5-100% for women. Discussion Bland-Altman and bootstrap analysis demonstrated good agreement and stability of the predictive models. Conclusions Physical performance tests are simple, easily obtainable, and clinically relevant markers for cognitive function with aging; predictive models based on physical performance can be used to predict cognitive function.

Automated summary

This study found that better physical performance test results were associated with slower cognitive decline and better cognitive function. All physical performance tests were included in men's predictive models, while only the five chair stands test was included in women. Bland-Altman and bootstrap analysis demonstrated good agreement and stability of the predictive models.