Data were obtained from the China Health and Retirement Longitudinal Study (CHARLS), a nationwide population-based prospective cohort survey of adults aged 45 years or older. Further details on the study design and sampling strategies of the CHARLS have been described previously [33]. Briefly, in the baseline survey, a multistage stratified probability-proportionate-to-size sampling was adopted, and 17,708 adults from 450 villages or urban communities in 150 counties across 28 provinces in China were recruited. The baseline survey was implemented during the 2011-2012 period (wave 1), with subsequent waves completed approximately every 2 years, and data from wave 3 were collected from 2015 to 2016. This study included 7290 participants aged 60 years or older at baseline. Among these, participants were excluded due to a self-reported diagnosis of dementia or Parkinson disease or both at wave 1 (n=256), loss or death before wave 3 (n=646), or not having finished the gait speed test (n=1805 at wave 1 and 606 at wave 3) or cognitive tests (n=784 at wave 1 and 184 at wave 3). Ultimately, 3009 individuals were included in this study (Figure 1).

Two well-established composite measures were used to assess global cognitive function: mental intactness and episodic memory, which have been validated in Chinese people [34]. These metrics are similar to those used in the American Health and Retirement Study and in previous studies [35,36]. Mental intactness was based on components from the Telephone Interview of Cognitive Status, which described time orientation (naming the month, day, year, week, and season), numerical ability (serial subtraction of 7 from 100, 5 times), and visual and spatial abilities (redrawing task of 2 overlapping pentagons) [36,37]. Mental intactness scores were equal to the sum of the correct answers, with numbers ranging from 0 to 11. Episodic memory was based on immediate and delayed word recall tests. After the interviewer randomly read a list of 10 Chinese nouns, the respondents were asked to memorize as many words as they could immediately (immediate recall) and several minutes later (delayed recall). Episodic memory scores were the sum of correctly repeated words, ranging from 0 to 20. Global cognition scores were the aggregate of mental intactness and episodic memory scores ranging from 0 to 31, with higher scores indicating better cognitive performance.

Gait speed was computed over 2 straight 2.5-m path walking courses. Those who had difficulty stopping walking (eg, recent surgery, injury, or health problems) were excluded from this test. Eligible older participants were instructed to walk 2.5 m along a noncarpeted walking course (there and back) at their usual speed. Timing began when one of the participants’ feet crossed the start line and landed completely, and ended when the participant crossed the finish line completely. The average velocity (m/s; accurate to 2 decimal places) was calculated, with higher scores indicating faster speeds.

The following baseline measurements were included as potential covariates: age, gender, residence region, marital status, educational level, self-reported health, current smoking and drinking status, self-reported visual and hearing impairments, depressive symptoms, restrictions on activities of daily living (ADL), and BMI. Information on chronic diseases included hypertension, diabetes mellitus, dyslipidemia, heart disease, stroke, cancer, lung disease, arthritis, kidney disease, digestive disease, and asthma. The number of diseases was used as a covariate. Restriction was considered as 1 or more dependency in ADL, including dressing, bathing, eating, getting into/out of bed, and using the toilet [38]. Depressive symptoms were measured using 10 items from the Center for Epidemiologic Studies Depression Scale, and a score of 10 or more was defined as positive [39].

Continuous variables are presented as means and SDs and categorical variables are shown as numbers (proportions). Cross-lagged path models were first performed to examine the possible longitudinal reciprocal association between cognition and gait speed measured at 2 time points 4 years apart [32,40]. Figure 2A demonstrates this study’s general modeling strategy, including 5 paths and their corresponding coefficients. Cross-lagged paths concurrently include 2 cross-lagged coefficients, β_CL-1 and β_CL-2. β_CL-1 represents the effect of gait speed at time 1 on cognitive scores at time 2, whereas β_CL-2 implies the regression of cognitive scores at time 1 on gait speed at time 2. The temporal sequence was determined by comparing the estimated standardized cross-lagged coefficients [40]. Cross-sectional paths between gait speed and cognitive scores were also modeled, and the coefficient β_CS-Baseline at time 1 represented the baseline correlation between gait speed and cognitive scores. Finally, 2 autoregressive coefficients, β_AR-Gait and β_AR-Cognition, were obtained from the paths between times 1 and 2 for gait speed and cognitive scores, respectively, accounting for with-person stability in each measure from the baseline to the subsequent time.

Before the cross-lagged path analysis, the baseline and subsequent values of gait speed and cognitive scores were adjusted for a series of covariates by regression residual analyses, and then standardized with Z-transformation (mean 0, SD 1). Standardized coefficients were reported from the 4 models. A crude model and 3 multivariable-adjusted models were built: model 1 was built without any adjustments; model 2 was adjusted for age and sex; age, sex, residence, marital status, educational level, and smoking and drinking status were further controlled in model 3; and finally, model 4 was comprehensively adjusted, including the covariates in model 3 plus self-reporting of health scores, number of diseases, depressive symptoms, visual and hearing impairments, restriction on ADL, and BMI. A comparative fit index (CFI) >0.90 and standardized root-mean-square residual (SRMR) <0.05 were used to suggest the good fitness of the models [41]. Linear mixed-effects models were further implemented to visualize the longitudinal associations between cognition and gait speed with aging.

Several sensitivity analyses were performed. First, we controlled for the specific chronic diseases (11 diseases in total) instead of the number of diseases in model 4. Second, in addition to the covariates in model 4, we further adjusted for living environmental factors to control for their potential confounding, including the type of accommodation (multistory or 1 story), water sources (running water or no running water), room temperature (comfortable or uncomfortable), and household air pollution (clean fuels for both heating and cooking, clean fuel for heating or cooking, solid fuels for both heating and cooking) [42]. Finally, we excluded individuals with extremely low gait speed or global cognitive function scores or both at baseline (≤mean – 2 SD) to reduce the potential influence of the reverse temporal sequence. All analyses in this study were conducted in Stata 16.0 (StataCorp) and P≤.05 (2-tailed) was considered statistically significant.

The Biomedical Ethics Review Committee of Peking University approved the ethical application for the collection of human participant data for the CHARLS (IRB00001052-11015). The study data were anonymous. All the respondents provided written informed consent before the survey.

During the 4-year follow-up among Chinese older adults, there was a longitudinal bidirectional association between global cognitive function, mental intactness, episodic memory, and gait speed after accounting for baseline levels and possible covariates. In addition, the comparison of standardized cross-lagged coefficients suggested that the temporal association from baseline poor global cognition to subsequent slow gait speed may be stronger than vice versa. Trajectories derived from linear mixed-effects models added further evidence on the dynamic associations, with poorer baseline cognition and slower gait speed predicting faster decreases in gait speed and declines in cognition, respectively.

Similar to previous cross-sectional studies [8,16,17], we also found that gait speed was reciprocally and positively associated with cognitive function at baseline among older adults. However, longitudinal studies may provide stronger evidence in the temporal direction. Two previous longitudinal studies, both using linear mixed-effects models, have shown different results, with one [28] reporting a bidirectional association between cognition and gait speed among older individuals in the United Kingdom, and the other suggesting that faster initial gait speed was associated with less later cognitive decline among older American adults, but not vice versa [31]. This inconsistency may be attributable to differences in the mean age, sample size, and heterogeneity of walking speed measures. Nevertheless, owing to a unidirectional preset, this statistical method cannot simultaneously examine bidirectional associations of 2 time-varying factors or temporal precedence [29,32]. Therefore, we fitted cross-lagged path models based on structural equation models and visualized temporal association using linear mixed-effects models, as suggested by a previous study in this area [29].

Our findings provided new insights into the nature of age-related changes in gait speed and cognition. To the best of our knowledge, only a few studies have elucidated the longitudinal reciprocal association between cognition and gait speed in older adults. Two such studies, also based on structural equation models, have also found bidirectional relationships between poor cognition and slow gait speed among older American people [9,27]. Specifically, 1 study included 412 participants aged 60 or older, finding that executive function, but not memory, predicted subsequent usual gait speed, and vice versa [9]. However, another study conducted in women aged 70-79 years reported a mutual association between gait speed and global cognitive function and particular memory. In our study, there was a bidirectional relationship between global cognition and gait speed and between specific domains (mental intactness and episodic memory) and gait speed. The mental status in the CHARLS describes time orientation, numerical ability, and visual and spatial abilities, and is similar to executive function [43]. Compared with these 2 studies, the sample size in our study was relatively larger. Our study had a 4-year follow-up duration that was shorter than the 9- and 6-year follow-up periods. Different cognitive tests on specific domains and discrepancies in the gender of the participants may also lead to this inconsistency. Therefore, more similar studies in older Chinese individuals are needed to verify whether the results hold across a longer follow-up interval or in sex-specific older adults. Another notable study among older American adults reported that early gait speed decline may be a precursor to later global cognitive decline and that the reciprocal association was weaker [29]. Although this study also revealed reciprocal relationships between the 2 factors, its temporal sequence contradicts ours. In our analysis, the standardized cross-lagged coefficient of baseline global cognition on follow-up gait speed was nearly 2 times that of baseline gait speed on subsequent cognition, indicating that global cognitive decline might be the cause of decreased gait speed than its result. The reasons for this discrepancy may be the difference in participants’ baseline average age and the length of the follow-up period, given that both cognition and gait speed decline over time in older adults [8]. Our study participants were over 60 years old, while the older adults were 70-79 years in their study. Moreover, their design had a longer follow-up time of 9 years than our 4-year interval. Thus, the current temporal sequence of cognitive decline to gait slowing requires further evidence, especially in older Chinese individuals.

Consistent with prior unidirectional longitudinal studies [18,21,22,44], our results suggest that poor baseline cognition predicted gait speed decline after 4 years, irrespective of the exclusion of those with very low baseline gait speed and global cognition. Moreover, poor specific domains of cognition, including mental intactness and episodic memory, are predictive of decreased gait speed, which supports the results from previous studies [28,45]. A small pilot randomized controlled trial showed that improvements in domains of cognition, including executive function, attention, visuospatial skills, and processing speed, were associated with increases in walking speed [45]. Memory has also been reported to be positively and longitudinally associated with gait speed in older adults, supporting our results [28].

Although we found a stronger directionality of early cognitive change in subsequent gait speed change than the opposite pathway, we do not discount the importance of early gait change. As with prior results [24,26], although weak, we again demonstrated the predictive role of baseline faster gait speed in later cognition. A recent study also revealed that decreased gait speed precedes global cognitive decline during a longer follow-up duration than ours [23]. Evidence from a prospective study in 70-89-year older adults has also shown results similar to ours [31], with baseline faster gait speed associated with better global cognition, executive function, and memory. In addition, both gait speed and cognitive function were important when assessing the risk of dementia in older adults [29,46].

Neuroimaging studies also support this reciprocal relationship. Age-related changes in gait and cognition have been reported to share common underlying structural alterations in the prefrontal and temporal lobes [47]. Furthermore, several mechanisms may explain the stronger effect of cognition on gait speed. First, adverse structural alterations in the brain, such as hippocampal atrophy, white matter, and an increase in white matter damage, are associated with decreasing gait speed in older adults [48]. Second, walking is a complex task that requires energy, balance, and coordination of multiple systems, including the nervous system, and thus disruption or deterioration of any system tends to decrease gait speed [6,49]. Finally, specific cognitive capacities may be vital for the initiation and maintenance of walking [22]. A functional magnetic resonance imaging study has shown that some areas with greater activation in the brain reflect increased cognitive monitoring of movements in older participants compared with that in younger people [50]. Overall, cognitive decline may precede or co-occur with gait speed decrease among older Chinese adults; therefore, early arrangements should be ensured to maintain both cognitive function and gait status to avoid further adverse outcomes. Meanwhile, cognitive decline might be a sensitive early marker of gait deceleration, indicating that cognition might be an essential factor to be considered along with gait speed to identify risks for gait speed decrease or mobility deficits among older adults.

This study has several strengths. Participants aged 60 years or older were from the CHARLS, which is a nationally representative, population-based prospective cohort in which cognitive functioning was assessed with well-validated measures. Relatively comprehensive information is available in the CHARLS, and related covariates were adjusted as much as possible in this study’s final model. In addition, this study expands upon cross-sectional and traditional longitudinal analyses by performing cross-lagged panel models, which allowed for the capture of a temporal sequence by simultaneously evaluating the strengths of temporal bidirectional associations [32,40]. In addition, compared with similar studies [27,29], our study added linear mixed-effects models to visualize the temporal bidirectional relationships of the 2 factors. Finally, to our knowledge, this is the first study to examine the possible temporal sequence of changes in gait speed and cognition in older Chinese people, and our findings deserve further verification.

Several limitations also need to be considered. First, our analysis involved an approximately 4-year follow-up interval due to the limited availability of gait speed information at the later wave mark; therefore, it could not be determined whether this association continued over a longer period. Second, while the reciprocal associations between cognition and gait speed were statistically significant after accounting for a variety of potential confounding factors, there may still be some related but uncontrolled variables such as the apolipoprotein E4 (ApoE4) genotype. Slow walking speed in those with ApoE4 preceded lower memory scores [24], but unfortunately, these data are not available in the CHARLS. Based on a genomic study, the e4 allele has a 6.9% frequency rate in the Chinese population; therefore, this genotype may have slightly affected our results [35]. Third, the older participants were allowed walking aids for this study, which may obscure the true association between gait and cognition to some extent. As one recent study has shown, the cross-sectional correlation between cognition and gait may be attenuated when considering the interaction of walking-aid usage [8]. Eventually, gait speed was measured at a usual pace, with this study solely focused on this walking condition; therefore, we were unable to determine whether the relationships we reported agree with the results obtained under other walking conditions, such as maximum walking speed, and this requires further research to identify.

This study demonstrates a longitudinal reciprocal association of gait speed with global cognitive function, mental intactness, and episodic memory among Chinese older adults. Cross-lagged path coefficients suggest that baseline global cognition is likely to have a stronger association with subsequent gait speed than the reverse pathway. These findings further reinforce the importance of promoting both cognition and gait conditions early, and provide ideas for improving the clarity of the causal direction of cognition and gait speed changes. Maintaining normal cognitive function may be an important interventional strategy for mitigating age-related gait speed decrease.