Spatiotemporal gait parameter fluctuations in older adults affected by mild cognitive impairment: comparisons among three cognitive dual-task tests

This study explored the gait performance in normal aging people and subjects with cognitive decline in finishing single or three different dual task paradigms. In a significant confounders balanced population, important variables associating with MCI were filtered using random forest and LASSO models in each task. The commonly differed variables between NC and MCI groups were temporal gait characteristics including swing time, mid stance, and terminal swing as well as spatial characteristic DTC in the dual task modes as further evaluated by logistic regression analysis. Compared with single task, all subjects walked slower when executing an additional cognitive task as reflected by the substantially decreased velocity, and this decrease was particularly obvious in finishing the words recall task requiring the memory capability. Therefore, disturbed temporal gait parameters under dual tasks may provide objective evidence for the clinical screening of cognitive decline in aging population.

In our study, by comparing the demographic data of the 260 enrolled participants categorized as NC and MCI, we found that participants in MCI group had significantly higher body weight as well as BMI, and lower educational levels, indicating that obesity and educational background may associate with cognitive decline. Several epidemiological studies reported that people with higher BMI had a greater risk for developing MCI and AD [29‐31]. Obesity was associated with lower brain volumes in cognitively normal elderly subjects and higher BMI was associated with brain volume deficits in both AD and MCI [32]. Educational background is a well acknowledged contributor associating with cognitive status. Multiple lines of evidence revealed that higher education level served as a protective factor to reduce the risk of MCI and AD [33, 34]. Rolstad et al. reported that stable MCI patients with higher education had lower concentrations of t-tau as compared to those with lower education, and higher education may offer protection against tauopathy [35]. Therefore, propensity score matching method was used to balance these significant confounding variables between the two groups. Additionally, although MCI and NC groups present comparable frequency of sleep disorder in this study, it still exerts an adverse influence on individual’s cognitive function. A longitudinal population-based cohort study revealed that sleep disturbance was associated with worse future cognitive performance for the 60-year-olds [36]. A significant V-shaped association is shown between sleep duration and MCI/dementia risk in women with either short (≤ 6 h/night) or long (≥ 8 h/night) sleep duration involving higher risk of cognitive impairment [37]. Sleep abnormalities can accelerate AD pathophysiology, promoting the accumulation of amyloid-β and phosphorylated tau [38]. In our study, about half of the participants in NC or MCI groups reported sleep disorder with regarding to abnormal sleep duration or sleep latency, indicating that sleep patterns underwent significant modifications in micro and macrostructure along with aging. Therefore, sleep problems should be noticed and individualized interventions targeting sleep disturbances in elderly people should be recommended to prevent or decelerate conversion to dementia. Besides, further studies are needed to expand our understanding on the contribution of sleep disorder to cognitive decline and the associated behavior, such as walking and gait characteristics.

Accumulating studies supported that walking is a complicated activity involving both motor and cognitive functions and their interplay or coordination rather than an automatic process [15]. Based on this rational, gait characteristics of patients with cognitive decline have been discussed in several previous studies [16, 39, 40]. A meta-analysis summarizing the effect of MCI on gait involving 11 studies concluded that MCI affected specific gait parameters, and these changes were particularly pronounced when subjects were challenged cognitively [18]. In these retrieved studies, four criteria were used for the diagnosis of MCI and three modalities of instruments were used to measure the gait function, including electronic walkways, force plates and body-worn sensors. Items of gait parameter varied substantially across the included studies even when the same instrument was used, and only routine spatial gait parameters were reported. Therefore, more studies are needed due to insufficient evidence of these heterogeneous studies. In the past decades, wearable sensors were prevalent in gait analysis and proved to be useful as they permitted a simple, objective assessment of human gait [41]. In our study, a wearable sensor was used and about 30 items of gait characteristics were captured. This portable device could collect spatial-temporal and kinetic gait features. These measured features were pooled into algorithms and spatial-temporal features were selected for their association with MCI, indicating their sensitivity in detecting cognitive decline.

We evaluated the gait performance of normal aging people and patients with MCI in single task and three different dual tasks challenging cognitive capabilities. It was found that participants with MCI exhibited significantly higher variability of stride length in single task than the normal controls, suggesting the disturbed gait regularity in these subjects. This gait disturbance also can be reflected under dual task condition when performing naming animals test. The greater variability of stride length in MCI were also reported in other studies [42, 43]. Moreover, our study identified that in dual task mode, temporal gait parameters swing time, in paralleled with the percent of terminal swing phase in a gait cycle greatly reduced in cognitively impaired individuals compared to normal aging people. A gait cycle is defined as the period from the initial contact of one foot to the following occurrence of the same event with the same foot. Currently, the gait cycle could be partitioned at different granularity based on the measuring methods of events and temporal phases. With the advancement of measuring techniques, wearable sensors emerge as the most promising device for extraction and analysis of larger number of features of gait, which enable the gait segment into more sub-phases (ranging from 2 to 8 sub-phases) [10]. Nowadays, the widely used wearable sensors for gait phase recognition include linear accelerometer, gyroscope, force-based measurements, electromyographic sensors, inertial measurement units, and joint angular sensors. It is concluded that analysis of the acceleration allowed researchers to recognize a greater granularity of gait cycles, such as the sub-phases of the swing phase [44]. In our study, IDEEA system carrying five accelerometers were used to monitor body and limb motions constantly. These motion signals are first preprocessed by signal conditioners and then output as electric signals representing motion and speed. Afterwards, the electric signals fed to the microprocessor data acquisition unit at high rate through a cable. Thus, the motion signals were transformed to time-serial waveform curves showing amplitude of relative position and acceleration information of the subject’s gait. Currently, numerous valuable methods are used for gait phase partitioning based on the waveform curves. Basically, threshold method, time-frequency analysis, and peak heuristic algorithms are used for event and phase detection. While, machine learning based approaches, containing various algorithms such as Hidden Markov models, Deep Learning Neural Network, and so on, are becoming mainstay techniques. Different computation methodologies provide different performances regarding the parameters such as the number of detectable phases, events, and detection delay [5]. Taborri et al. summarized that both threshold-based methods and machine-learning approaches could obtain satisfactory performance in gait phase detection and permit the sub-partitioning of the swing phase [44]. In IDEEA system, wavelet-based algorithm and Bayesian analysis are used to analyze the trajectory and recognize the phases of gait. The different combinations of signals from those five sensors represent different physical activities. Thus, the sensors and algorithms applied in this study allow to segment the gait cycle into eight sub-phases, particularly for the swing phase partitioning [44]. Swing time refers to the duration between the Toe-Off and the Heel-Strike of one leg inside a gait cycle, which takes 0.36 ~ 0.40 s and accounts for approximate 40% of this cycle. During this phase, the leg first pushes backwards and then swings forwards, transforming the potential energy into kinetic energy, and resulting in the highest values in the acceleration and angular velocity signal to propel the forward motion of the whole body [10]. The actual swing is divided into three phases: initial, mid and terminal swing phase at approximately 60–75%, 75–85% and 85–100% of the gait cycle, respectively. The terminal swing phase, the ending of swing phase and the entire gait cycle, is responsible for decelerating forward motion of the lower limbs and preparing for foot landing for the next gait cycle. The decreased swing time and terminal swing in MCI may indicate an impaired capacity for moving forwards when conducting a cognitive task. Meanwhile, we found that MCI patients showed elevated mid stance than the control. The mid stance is the only phase of a single support for whole gravity, which functions to maintain the stability of the knee joint and control the forward inertial motion of the limb. Therefore, the participants with MCI may experience insufficient walking stability and posture control when performing a dual task that needed being compensated by prolonged mid stance. Conventionally, gait velocity, cadence and stride length were frequently focused in MCI patients in previous studies, while the detailed temporal characteristics were rarely portrayed [45]. Thus, a comprehensive understanding of specific gait pattern in MCI population is needed since gait is a complex integrated activity and slow speed is a nonspecific variable linking with many subjective and objective factors [46]. We did not observe substantial difference in gait speed between MCI and control under single and dual tasks in spite adjusting the covariates of age, sex and height or weight. The previous study tended to believe that patients with MCI exhibited slower speed under single or dual task conditions although controversy existed in other studies showing no difference between them [47‐49]. It can be explained that our study recruited subjects aging over 50 years with 85% of them ranging from 50 to 70, which may differ from other studies in age distribution. Meanwhile, no priority was permitted to perform motor or cognitive tasks, thus, the attention capacity and preference in allocating these resources between the two tasks may also affect the performance in gait speed.

In our study, three different dual cognitive tasks compassing subsequent 100-7, naming animals and words recall were loaded onto the gait assessment, and their impacts on gait performance of the participants were evaluated. Overall, all subjects exhibited significantly slower speed when conducting an additional cognitive task comparing with performing a single task, which was well acknowledged in other studies [45]. Meantime, we noticed that subjects had poorer gait performance in words recall test, as reflected by the more obvious decrease in velocity and higher velocity cost comparing with the other two tests. The arithmetic tasks (subsequent subtraction by 7s from 100) and verbal fluency tasks (enumerating animal names) seemed to have comparable effects on gait function. The three dual tasks challenged different cognitive resources. Serial 100-7 is a mental tracing task engaging numerical processing skills; verbal fluency tasks involving semantic knowledge and retrieval processes, while words recall activities engaging episodic memory encoding and retrieval processes. Activation of neural circuit in bilateral prefrontal cortices are implicated in all these tasks [50]. Related studies showed that verbal fluency tasks resulted in similar magnitude of interference as mental tracking tasks when walking [51]. In our study, words recall trial caused higher burden and disturbance on gait, indicating that this task is more complex and high-demanding, which is consistent with the high frequency of memory impairment in MCI. As depicted by Schwenk et al., more complex cognitive tasks seem to be required to elicit the gait speed differences between healthy from cognitively declined subjects [46].