Age-specific differences in gait domains and global cognitive function in older women: gait characteristics based on gait speed modification

Participants

Participants for this study were recruited as part of a community-wide survey in Busan metropolitan city in 2018–2019. We contacted to 519 women aged 65 years and over living in the community, with 296 participants responding (response rate: 57.0%). Participants unable to walk without any support and with a history of severe orthopedic problems or neurosurgical and neurophysiological problems in the preceding six months were excluded. Ultimately, 164 community-dwelling cognitively normal older women aged 65 years and over participated in the overground walking test (Fig. 1; Table 1). All the participants read and signed an informed consent form that was approved by the Institutional Review Board of Dong-A University (IRB number: 2-104709-AB-N-01-201808-HR-023-02).

Instrumentation

We used a shoe-type gait analysis system based on an IMU sensor (DynaStab™, JEIOS, Busan, South Korea) composed of shoe-type data loggers (Smart Balance SB-1; JEIOS, South Korea) with a data acquisition system (Lee et al., 2018). The shoe-type data logger included IMU sensors on both the left and right outsoles (IMU-3000TM; InvenSense, San Jose, CA, USA); these were used to measure triaxial acceleration (up to ±6 g) and triaxial angular velocities (up to ±500° s⁻¹) along three orthogonal axes (Kim et al., 2016; Lee et al., 2018). The data were transmitted wirelessly to a data acquisition system via Bluetooth, and multiple shoe sizes were available to fit all participants. The local coordinate system of the IMU sensors was established in the anteroposterior, mediolateral and vertical directions (Fig. 2).

Test procedures

Prior to the overground walking test, biometric data, including the body height, weight, and body fat percentage, of all the participants were recorded. All participants completed a questionnaire to assess their physical activity (PA) levels. Habitual PA levels were assessed by a self-reported questionnaire (international PA questionnaire-short form, IPAQ-SF). All participants performed a warm-up protocol comprising a stretching program. They performed the overground walking test wearing the shoe-type embedded IMU sensors.

Global cognitive function assessments

The MMSE was used to assess global cognitive function (Folstein, Folstein & McHugh, 1975); this is the most common screening tool for dementia and for assessing global cognitive function. All participants completed this questionnaire and we assessed their responses in terms of their global cognition. It is a 30-point questionnaire that includes tests pertaining to orientation, attention, memory, language and visual-spatial skills.

Gait performance measures

All participants completed three trials of the overground walking test along a straight 20 m walkway (approximately 40 steps) at slower (80% of usual), preferred (usual walking), and faster (120% of usual) speeds. The preferred speed refers to the comfortable and stable walking speed of the participant without any support during overground walking. The walking speeds were defined using a metronome (beats/min). Each speed was calculated relative to the preferred speed, that is, 20% slower or faster than the preferred speed (Chung & Wang, 2010). The participant was asked to walk at the preferred speed to measure cadence using a metronome before each trial. An experimental operator mentioned the walking speeds to the participants before each trial, and the participants were required to perform the overground walking test at the speed that was as close as possible to the target walking speed paced by a metronome. The intraclass correlation coefficient (ICC) was calculated to distinguish between the real and calculated speed consistency at slower and faster speeds. They were also given verbal or visual instructions to perform overground walking and practiced walking under all speed conditions with the metronome for approximately 10 min during a familiarization session.

Data analysis

The overground walking data were collected at 100 Hz. The shoe-type IMU system data were filtered using a second-order Butterworth low-pass filter with a cutoff frequency of 10 Hz (Kim et al., 2016; Lee et al., 2018). The two initial acceleration steps and final deceleration steps at the end of the test were excluded as we intended to analyze only the consecutive steps in the steady-state condition (Fig. 2). Gait events were identified as heel strikes that occurred when the linear acceleration along the anteroposterior axis reached its maximum value. Subsequently, toe-offs occurred when the linear acceleration along the vertical axis reached its maximum value during a gait cycle (Kim et al., 2016; Lee et al., 2018).

The spatiotemporal parameters, such as the walking speed, cadence, stride length, step length, stride time, step time, single support phase, double support phase and stance phase, were calculated. The walking speed was calculated the average speed as time for the 20 m walk from an IMU sensor (time/distance); This was converted to meter per second. The normalized spatiotemporal parameters were divided by the height of each participant. In addition, the coefficient of variance (CV (standard deviation/mean) × 100) values for all spatiotemporal parameters were calculated. The gait asymmetry (GA) was measured using the differences between the left and right movements during walking (Plotnik, Giladi & Hausdorff, 2007). These parameters can be categorized into the following gait domains: pace (walking speed, stride length, step length and normalized variables), rhythm (cadence, stride and step time), phase (single support, double support and stance phase), variability (CV for spatiotemporal parameters) and asymmetry (GA) (Lord et al., 2013).

Statistical analysis

All statistical analyses were performed using SPSS for Windows (version 25.0; IBM Corp., Armonk, NY, USA). The Shapiro–Wilk test was used to determine whether the data had a normal distribution. Before additional analysis, Z-normalization (value–mean/standard deviation) of all variables was performed. Especially, Z-normalization was performed to assess the global cognitive function because the MMSE score is not used as a continuous variable. Subsequently, one-way analysis of variance with the Bonferroni post-hoc test was used to compare the age-specific differences in the MMSE scores (MMSE cutoff < 24) and spatiotemporal parameters under different three walking speeds with the statistical significance levels set at 0.0167 (0.05/3). Multinomial logistic regression was conducted to determine the age classifiers for all participants aged 65–69 years under each walking speed that contained all confounders. Furthermore, stepwise multivariable linear regression analysis was performed to identify the independent variables and explain the significance of the dependent variables (MMSE scores). The covariates were the age, education, % body fat and PA. Previous studies reported that excessive body fat in adulthood is prospectively associated with increased risks of cognitive dysfunction and dementia in later life (Anstey et al., 2011; Loef & Walach, 2013). The level of statistical significance was set at 0.05.

Comparisons among the age-specific groups in terms of the MMSE scores and spatiotemporal parameters

Table 2 lists only statistically significant results for the MMSE scores and gait-related variables at three different speeds for the age-specific groups that comprised cognitively normal participants. The MMSE score shows that a significant part of the subjects score <24, which defines a degree of dementia. The MMSE scores were significantly lower for women aged >75 years compared to those aged 65–69 years (p = 0.008). Most variables indicated significant differences between the age-specific groups, namely 65–69 years and >75 years regarding the pace (stride length, step length and normalized variables at all speeds as well as the walking speed and normalized walking speed at the preferred and faster speeds), phases (double support phase and stance phase at faster walking speed), and variability (CV of stance phase at faster walking speeds) (all p < 0.05). In addition, the pace (stride length, step length and normalized variables at all speeds as well as the walking speed and normalized walking speed at the preferred speed) showed significant differences between the groups aged 70–74 years and >75 years (all p < 0.05). The double support phase and stance phase at faster walking speeds showed significant differences between groups aged 65–69 years and 70–74 years (all p < 0.05).

Multinomial logistic regression results

Table 3 summarizes only statistically significant results from the multinomial logistic regression for all participants. In the multinomial logistic regression models adjusted for confounders, gait-related variables at each walking speed, namely the walking speed, cadence, stride length, step length, stride time, step time, single support phase, double support phase, stance phase, CVs of the variables (stride length, step length, single support phase, double support phase and stance phase) and GA were considered.

When variables associated with slower walking speeds were added to the adjusted model, GA in women aged >75 years was associated with age based on the results obtained for women aged 65–69 years (OR = 1.51 and p < 0.05). As for the variables associated with the preferred walking speeds, the double support phase (OR = 1.53 and p < 0.05) and stance phase (OR = 1.58 and p < 0.05) in women aged >75 years were associated with age based on the results obtained for women aged 65–69 years. At faster walking speeds, the stride time (OR = 1.57 and p < 0.05), double support phase (OR = 1.69 and p < 0.05), and stance phase (OR = 1.70 and p < 0.05) for women aged 70–74 years were associated with age based on the results for women aged 65–69 years, which were added to the adjusted model. In addition, the step time (OR = 1.64 and p < 0.05), double support phase (OR = 1.66 and p < 0.05), stance phase (OR = 1.74 and p < 0.05), and the CVs of the step length (OR = 1.55 and p < 0.05), single support phase (OR = 1.55 and p < 0.05), and stance phase (OR = 1.63 and p < 0.05) in women aged >75 years were associated with age based on the results for women aged 65–69 years when variables associated with faster walking speeds were added to the adjusted model.

Results of stepwise multivariable linear regression analysis

Table 4 lists only statistically significant results for the associations between gait parameters based on the MMSE scores for older women at three different speeds. After adjustments to the confounders, the MMSE scores for all participants were significantly associated with the double support phase (slower, β = −0.173 and p = 0.024) and stance phase (preferred, β = −0.242 and p = 0.002; faster, β = −0.245 and p = 0.002). For women aged 70–74 years, cadence (β = −0.280 and p = 0.018) was significantly associated with the MMSE score at faster walking speeds. In addition, the CVs of the stride length (slower, β = −0.568 and p = 0.006) and stance phase (preferred, β = −0.471 and p = 0.005; faster, β = −0.494 and p = 0.005) were significantly associated with the MMSE score in women aged >75 years.

Age-specific differences in gait domains according to gait speed modification in older women

At the preferred speed, our findings of step length and cadence yielded similar results compared with a previous study in women aged 60–69 years (Senden et al., 2009) (present vs. reference, step length: 0.64 ± 0.06 m vs. 0.66 ± 0.07 m; cadence: 117.39 ± 9.19 st/min vs. 116.53 ± 5.60 st/min). These findings imply that our findings were as good as those found in the other study (Senden et al., 2009). These results help validate our findings, which is necessary for clinical use.

Numerous studies have obtained findings similar to our study (Van Kan et al., 2009; Cesari et al., 2005; Studenski et al., 2011). Our study indicated that there were age-related decrements in the pace domain at three different walking speeds in older women with lower PA and education levels. Interestingly, phase domain parameters increased until people reached 70 years and then became steady whereas the pace domain parameters decreased steadily with age. This state may occur because of the motor dysfunction associated with age-related reduction in muscle power (Senefeld, Yoon & Hunter, 2017; Thom et al., 2007; Thompson et al., 2013), requiring economical walking with minimal energy expenditure as a compensative strategy. However, this state was observed at faster walking speeds only in our study. Our logistic analysis showed age-specific changes in the phase domain in women aged 70–74 years; moreover, the phase and variability domains in women aged >75 years were highly dependent on aging. Therefore, these two domains may reflect the aging effects better than other gait domains. Thus, analyzing the effects of aging on gait, it is required to identify age-specific changes with more detail age classifications including people aged >85 years.

Age-specific association between gait domains and global cognitive function according to gait speed modification

Previous studies reported that the variability domain parameter is most strongly associated with cognitive function (Hausdorff et al., 2005; Martin et al., 2013; Van Iersel et al., 2008; Verlinden et al., 2014). Likewise, our study also showed that the higher CV values of the stride length was associated with the lower MMSE scores at slower walking speeds in women aged >75 years. It may be related to the executive dysfunction of cortical sensorimotor control (Beauchet et al., 2009, 2013, 2014) in response to a reduced hippocampal volume and impaired function in older adults (Annweiler et al., 2012). The executive function is related to initiating and modulating the gait performance (Senefeld, Yoon & Hunter, 2017). Therefore, high gait variability could be attributed to the stride-to-stride fluctuations during walking to generate force using muscle with the partial summation of overlapping twitches due to executive dysfunction during modulation in slow walking. Furthermore, our ICC results presented relatively lower value at slower speed (ICC, 0.85) than faster speed (ICC, 0.99). Therefore, it is difficult to adjust the speed when walking at slow speeds for older women with declining cognitive function. Thus, the variability domain at slower walking speeds can be a more sensitive predictor for identifying future cognitive decline than other gait domains at relatively fast walking speeds (Beauchet et al., 2013; Verghese et al., 2008).

Our findings are similar to previous studies that analyzed the association between a longer stance phase and lower MMSE score from the preferred to the faster walking speed in women aged >75 years. Very few studies have shown that phase domain parameters showed the strongest relationship with cognitive function (Taniguchi et al., 2019) involving memory (Verlinden et al., 2014). In addition, a longer stance phase may result from the force to length relationship. The optimal propulsion force during walking could be achieved at optimal step lengths; however, shorter step lengths lead to decreased walking speeds owing to a longer stance phase. This gait pattern as a compensative strategy (increased stance phase with slow walking speed) increases dynamic instability and could lead to falls (Aboutorabi et al., 2016). This pattern appears to result from the deterioration of locomotion as cognitive function declines, to which women may be more vulnerable than men. This finding is supported by the fact that overlapped areas of the basal ganglia and cerebellum are major subcortical structures that cooperate to control the cognitive and motor functions as well as walking (Holtzer et al., 2014; Kikkert et al., 2016). Furthermore, Iosa et al. (2013) and Serrao et al. (2017) emphasized that the golden ratio (60–62% of stance phase vs. 40–38% of swing phase) between the duration of the stance and swing phases of a gait cycle, which is the gait harmony. In particular, the loss of gait harmony could be disrupted in response to the damage in the cerebellum or basal ganglia (Serrao et al., 2017). In addition, gait stability is highly dependent on the ratio between the duration of the stance and swing phases because it could alter the gait variability. Thus, the phase domain should also be considered meaningful to the understanding cognitive function decline in old women.

In addition, Darweesh et al. (2019) in their Rotterdam Study showed a tendency for the association between the phase domain and incident dementia. It may be because of the relatively fewer steps considered in their gait assessment using GAITRite™ compared to our study. Thus, a longer walking time and walkway distance are required to identify the relationship between gait and cognitive function decline.

This study had some limitations. First, all participants performed the walking at a self-controlled speed, which may have artificially changed gait parameters. However, we confirmed the ICC results which were calculated to distinguish between the real and calculated speed consistency at slower and faster speeds (r > 0.77). Second, our study used the MMSE scores, which have a relatively lower sensitivity (Trzepacz et al., 2015). However, the MMSE score is a widely used screening tool for dementia and global cognitive function as well as clinical evaluations. Further research is warranted to address these issues. Third, cognitively normal participants were included into this study. Despite assuming this status, we speculate that some participants could have a diagnostic level of dementia. Fourth, this study did not entirely consider the recommendations to improve the validity of research using the MMSE (Monroe & Carter, 2012); the recommendations should be considered to improve the validity of a future study using the MMSE.