Association of Gait With Global and Domain-specic Cognitive Function Among Community-dwelling Older Adults: A Cross-sectional Study

Background: Gait is associated with cognitive function and is a trait marker of dementia; however, research on gait and cognitive function usually concentrates on several individual gait parameters. This study used wearable sensors to measure gait parameters in different aspects and comprehensively explored the association of gait with global cognitive function and domain-specic cognitive function. Methods: The data of this cross-sectional study were obtained from 236 community-dwelling Japanese older adults (125 men and 111 women) aged 70–81 years. Gait was measured by asking participants to walk a 6-meter course and back using the Physilog® sensors (GaiUp®, Switzerland). Global cognitive function and cognitive domains were evaluated by face-to-face interviews using the Japanese version of the Montreal Cognitive Assessment. Twenty gait parameters were summarized as independent gait factors using factor analysis. A generalized linear model and linear regression model were used to explore the relationship of gait with global cognitive function and domain-specic cognitive function adjusted for several confounding factors. Results: Factor analysis yielded four gait factors: general cycle, initial contact, propulsion, and mid-swing. Among them, general cycle factor was signicantly associated with global cognitive function (β=-0.565, [-0.967, -0.163]), executive function (P=0.012), and memory (P=0.045); initial contact was associated with executive function (P=0.019). Conclusion: Better gait was related to better cognitive function, especially the general cycle, which was correlated with both global and domain-specic cognitive function. The predictive value should be examined in future cohort studies.

Several individual gait parameters are signi cantly correlated with global cognitive function in older adults, including preferred gait speed [7,8], maximum gait speed [9], and stride length [10]. Moreover, some studies have demonstrated that gait is related to domain-speci c cognitive function. For instance, impaired gait velocity and reduced cadence are associated with worse executive function and immediate memory [11].
The mean walking speed and maximum walking speed are signi cantly associated with the attention domain [12].
Nevertheless, gait is multidimensional and cannot be evaluated by a single characteristic. A single parameter applied before could not show a comprehensive association between gait and cognitive function.
Recently, a wearable sensor has been invented that can measure gait from general, temporal, visual, and clearance aspects, while simultaneously allowing people to walk in a natural environment. Moreover, based on the knowledge that quantitative gait parameters are highly correlated with each other and their individual relationships with cognitive function may be di cult to observe while adjusting for other gait variables [13], conceptual models using principal component analysis or factor analysis to provide a simpli ed framework for selecting grouped gait factors were suggested in gait analysis [14]. Under these circumstances, research that includes gait parameters from different aspects and uses a conceptual model for systemically analyzing gait and cognitive function is scarce.
Therefore, by adopting a more advanced sensor, we aimed to examine comprehensively the associations between gait and cognitive function to make gait a more plausible marker of dementia.

Participants
The data for this research were obtained from the Cognition and Activity in Rural Environment of HokkaiDO Senior (CARE-DO) study 2018, which was conducted after the Japan Gerontological Evaluation Study 2016.
The Japan Gerontological Evaluation Study is a large panel study directed at understanding the health, social, and behavioral issues among the older population in Japan [15]. In the Japan Gerontological Evaluation Study 2016 wave, more than 200,000 citizens who did not receive long-term care insurance from 39 cities were included. Among them, 5013 community-dwelling Hokkaido residents aged between 69 and 80 years were invited to attend the CARE-DO Study in winter 2017, and 569 participants who responded to the winter survey in 2017 were subsequently invited to the CARE-DO summer study in September and October 2018. A total of 309 participants who attended were regarded as the participants of this study.
The CARE-DO Study 2018 included four main parts: physical examinations, gait and cognitive assessments, and a self-administered questionnaire. Height was measured using the corresponding scale during the physical examination. After the participants completed all assessments, a self-administered questionnaire that contained information on age, sex, and history of diseases, such as diabetes (yes or no) and hypertension (yes or no), were allocated to each of them and expected to be sent back after 2 weeks. After excluding 21 participants with missing data on cognitive and gait assessments; one with depression; three with Parkinson's disease or Alzheimer's disease; ve with stroke; 35 with musculoskeletal pain; and eight who once fell down or had a bone fracture; 236 participants were regarded as valid participants and were included in the data analysis process (Fig. 1).
This study was approved by the ethics committee of the Graduate School of Medicine, Hokkaido University (no.18-025), all the experiment protocol for involving human data was in accordance with the relevant guidelines. A written informed consent was obtained from all the participants.

Gait measurement
All participants were asked to walk a round trip of a 6-meter straight path at a self-selected walking speed. Gait was measured using Physilog® sensors (GaiUp®, Switzerland), which consist of two small, lightweight (19 g), inertial sensors for each foot, and elastic straps to attach the sensors to the dorsum of the foot, and Gait Analyzer software version 3.1 (GaitUp) running on a Windows personal computer. Two sensors on both the feet could be synchronized wirelessly, and no calibration procedure was required before and during the measurement. The position of the sensor on the foot did not affect the measurements [16]. The following 21 kinematic gait parameters from four aspects could be obtained through this device: general (cycle generation, cadence, stride time, stride velocity, and turning angle); temporal (stance, swing, loading, foot-at, pushing, and double support); spatial (peak angle velocity, swing speed, strike angle, lift-off angle, swing width, and three-dimensional (3D) path length); and clearance (maximum heel clearance, maximum toe clearance 1, minimum toe clearance, and max toe clearance 2). As gait cycle duration is just a different format of cadence, the parameter of gait cycle duration was excluded before the analysis. The data at the time of accelerating or decelerating and making a turn were excluded by the default function of the Gait Analyzer. The mean values of two feet were used. Detailed descriptions of these parameters cis provided in Additional le1.

Cognition measurement
Global cognitive function was evaluated using the Japanese version of the Montreal Cognitive Assessment (MoCA-J) through face-to-face interviews. The MoCA-J is a one-page 30-point test administered in approximately 15 min by trained investigators, and higher scores represent better cognitive function. To regulate the effect of education, participants with an educational background of less than 12 years were asked to add one score to the total score. The MoCA-J can assess the following six cognitive domains: memory ( ve points): delayed recall; executive function (four points): letter uency, trial making, verbal abstraction; visuospatial (four points): cube copy, clock drawing; languages ( ve points): naming, sentence repetition; attention (six points): digit span forward and backward, letter A tapping, serial-7 subtraction; and orientation (six points): orientation of date and place [17].

Statistical analysis
The chi-square test was used to compare the categorical variables, and a one-way analysis of variance was used to compare the continuous variables.
All 20 continuous gait parameters were standardized to have a mean of 0 and a standard deviation of one because of the different scales before analysis. Subsequently, factor analysis using the principal component method was performed to reduce the larger number of highly correlated variables to a smaller number of uncorrelated independent predictors. The initial factors were subjected to an orthogonal quartimax rotation. Gait parameters with a rotated loading of > 0.5 were considered as the dominant contributors to the main factors. The name was labeled according to the interpretation of the original gait parameters. A generalized linear model was used to identify the relationship between the MoCA-J score and retained gait factors in all the participants and both sexes. Due to the total score of each cognitive domain was small and direct adoption of the linear regression model may cause bias, tertile was made for each gait factor. The mean value of each group was adjusted for age, sex, height, and education as the least square mean value. P for trends were tested by linear regression model, in all the participants and both sexes, test for trend based on variable containing median value for each tertile. The main confounding factors in this study were age, sex, height, educational status, diabetes, and hypertension. Statistical signi cance was set at P < 0.05. All statistical analyses were performed using the SAS software (version 9.4; SAS Institute Inc., Cary, NC, USA).

Results
The study included 125 men and 111 women with an average age of 75.28 years and 75.53 years, respectively. Approximately 45% of the patients had hypertension and 14% had diabetes. In gait assessments, women had signi cantly shorter stride length, quicker cadence, and shorter stance period than men. In cognitive measurements, women performed signi cantly better on memory (Table 1).  Table 2). The factor with the highest variance had strong loadings on stance, cadence, foot-at, stride velocity, double support, swing, swing speed and peak angle velocity and was termed "general cycle" factor. The second factor re ecting the strike angle, maximum toe clearance stride length loading, and 3D path length was named "initial contact" factor. The third factor loaded on pushing and lift-off angle was termed "propulsion" factor. The nal factor loaded heavily on maximum heel clearance and maximum toe clearance and was called "mid-swing" factor ( Fig. 2). A lower score for the general cycle factor and propulsion factor represented better gait, and a higher score of initial contact and mid-swing factor represented better gait. Gait and global cognitive function  neither in the crude model nor in the adjusted models (Table 3). Gait and domain-speci c cognitive function Of the six cognitive domains, signi cant differences were found in the general cycle factor (P = 0.012), initial contact factor (P = 0.019) with executive domain, and general cycle factor with memory domain (P = 0.045).
Better performance of the general cycle and initial contact was correlated with higher scores of executive function. Better performance of the general cycle factor was correlated with a better memory domain (Table 4). Men and women showed different results. Better performance of initial contact was related to higher scores of executive function in women (Additional Table 2); no signi cant relationship was found between gait and speci c domain in men (Additional Table 3).

Discussion
Our study found that among older adults aged 70-81 years, the general cycle is most associated with global cognitive function; when cognitive function is speci ed in domains, results showed better general cycle and initial contact phase related to better executive function, the general cycle phase was also related to the memory domain.
To the best of our knowledge, this is the rst study to include most of the various gait parameters that almost consist of the whole gait phase when studying the association between gait and cognitive function.
To date, various techniques have been used in gait analysis, including stopwatches, electronic walkways, body-worn sensors, electromyography, and 3D motion analysis [6,14]. The technique used in this study was an easily wearable sensor that measured 3D gait. The 3D motion analysis is considered accurate and is used as the gold standard in gait analysis [18]. Moreover, we used conceptual models to summarize various parameters of gait factors. Previous studies have also used a conceptual model [13,14,[19][20][21]; however, gait assessments in all those articles were performed using a walkway (GAITRite, CIR System Inc.). The nature of walkways determined that it can only assess general (velocity, width, cadence, and stride time) and temporal (time for swing, stance, single and double support, and turning) parameters, while those used in this study could additionally evaluate gait from spatial and clearance aspects, bene t from further factors of initial contact, propulsion, and mid-swing were output, which made the gait analysis more comprehensive.
Hence, compared to most other studies using simple techniques, our results are more precise and show a complete relationship between gait and cognitive function.
After adjusting for several confounders, the results showed that only general cycle factors (predominantly cadence, speed, and stance time) were signi cantly associated with global cognitive function in all the participants. People with lower scores for the general cycle factor were inclined to have higher MoCA-J scores. Internal biomechanism can be referred to as the proved mechanism between gait speed and cognitive function. Some lesions of the brain, such as an increased proportion of the periventricular and subcortical white matter hyperintensities, atrophy of the medial temporal areas, hippocampal atrophy [22], or small gray matter volumes in the bilateral cortical and subcortical regions, based on magnetic resonance imaging [23] can slow the gait speed, while simultaneously impairing cognitive function. In addition, recent studies have started researching the biomechanism between other parameters and cognitive function; for example, higher cerebral amyloid-β deposition was shown to be associated with increased double support time [24]. Therefore, by involving more gait aspects, a more robust mechanism could be clear between gait and cognitive function.
With respect to newly invented factors of initial contact, propulsion, and mid-swing, no signi cance was found for global cognitive function. Although previous studies claimed that the contributor of stride length in the initial contact factor was related to cognitive function [25], their relationship was probably caused by the high correlation with gait speed. In addition, according to a study of brain dynamics while walking, electrocortical activity progressively decreases in the pre-swing phase and acceleration phase [26], which indicates that compared to the preparatory phases for the most important point of walking, the general cycle factor that includes characteristics of posture control (double support and foot-at) is better associated with global cognitive function.
Further exploration of domain-speci c cognitive function supported the strong association between executive function and gait [27,28]. The term executive function refers to the higher-level cognitive skills we use to control and coordinate our other cognitive abilities and behaviors [29] traditionally, it is associated with the frontal lobes and related brain networks, in particular, the dorsolateral prefrontal cortex and cingulate cortex [30]. A previous study found that gait shared similar brain regions in which a great burden of subcortical white matter hyperintensities on magnetic resonance imaging is related to increased dual-task costs while walking [31], which could be the main reason for their high correlation. In addition, past studies have suggested that gait requires memory; shared anatomy can also be used to explain their robust relationship [32]. However, some individual gait parameters have a signi cant relationship with language and attention domain [33]. After being detained by the factor analysis, combined gait characteristics did not show any relationship with other domains except for executive function and memory domain. Therefore, enough evidence for an association between gait and the other two domains, orientation and visuospatial, in all the participants was not found [34].
Moreover, our results varied greatly in terms of sex. According to basic characteristics, men performed worse on both gait and cognitive assessment than women, which is proved by the previous studies that older women usually outperformed on the test of motor speed [35] and memory than men despite the same level of cognitive degeneration [36,37]. Moreover, women showed a stronger relationship between gait and global cognitive function, and no signi cance relationship was observed in men. The cognitive bene t of physical activity may be greater in women than in men [38]; however, the effect and biological mechanism of sex on the relationship between gait and cognitive function remains unclear. These observations illustrate that sex difference is an important confounding factor in the study of gait and cognitive function among older adults, and it should be discussed in future work.
This study has both strengths and limitations. Although the advanced device and conceptual models used in this study makes it more comprehensive and precise, the following aspects still warrant some attention: rst, this was a cross-sectional study, which meant that the causality between gait and cognitive function could not be determined. Second, during the process of measuring gait parameters, the rst and last two cycles were not excluded given the short total distance, while the initiation and termination of gait are sometimes thought to be unstable and inaccurate [39]. As claimed by other studies, older adults need at least four gait cycles to reach steady-state walking speed [40]. Third, only the MoCA-J questionnaire was used in the cognitive assessment procedure, although the MoCA was identi ed as an applicable method for evaluating global and domain-speci c cognition, a previous study claimed that the MoCA is not comparatively sensitive or speci c for identifying attention or language impairments [41,42]. To maintain the veracity of cognitive function, especially in the study of exploring different cognitive domains, extensive neuropsychological batteries should be added in the future research. Finally, the participants in our study were older than 70 years, and 70% of them had mild cognitive impairment (MCI) (MoCA < 26 is de ned as MCI [43]). However, the prevalence of MCI among older adults usually ranges between 3% and 42% [44]. Atypical MCI prevalence may be associated with potential physical functional impairments and disease history of the participants, which we did not exclude before analysis, and this may make it di cult to generalize the results to a wider population.

Conclusion
This cross-sectional study used advanced gait measurement and found that better gait indicated better performance of cognitive function among community-dwelling older adults aged between 70 and 81 years. In particular, the general cycle was signi cantly correlated with global cognitive function, executive function, and memory, and the initial contact factor was related to executive function. In the future, longitudinal cohort research with a larger sample size and extensive neuropsychological batteries is needed to detect the causality between gait and cognitive function. Availability of data and materials

Abbreviations
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests Authors' contributions WH contributed to the conception, analysis and interpretation of data, and preparation of manuscript. WJ contributed to the analysis and interpretation of data, preparation of manuscript, critical review and feedback. SU, KK and KK contributed to the critical review and feedback. TK contributed to the design, acquisition of data, critical review and feedback. AT contributed to the design, acquisition of data, interpretation of data, critical review and feedback. All authors read and approved the nal manuscript.   Additional lesgaitandcognitivefunction.docx