Single-Task or Dual-Task? Gait Assessment as a Potential Diagnostic Tool for Alzheimer's Dementia.

BACKGROUND: A person's gait performance requires the integration of sensorimotor and cognitive systems. Therefore, a person's gait may be influenced by concurrent cognitive load such as simultaneous talking. Although it has been known that gait performance of people with Alzheimer's dementia (AD) is compromised when they attempt a dual-task walking task, it is unclear if using a dual-task gait performance during an AD assessment yields higher diagnostic accuracy.
OBJECTIVE: This study was designed to compare the predictive power for AD of dual-task gait performance in an AD assessment to that of single-task gait performance.
METHODS: Participants (14 with AD and 15 healthy controls) walked across the GAITRite© Portable Walkway mat under three different cognitive load conditions: no simultaneous cognitive load, walking while counting numbers by ones, and walking while completing category naming.
RESULTS: Multiple logistic regression revealed that the gait performance under a dual-task condition (i.e., concurrent counting or category naming) increased the proportion of variance explained by the FAP, SL, and DST, of the incidence of AD.
CONCLUSION: Dual-task walking and talking may be a more effective diagnostic feature than single-task walking in a comprehensive AD diagnostic assessment.

INTRODUCTION

It has long been believed that the motor system is separate from the cognitive system both functionally and anatomically. Based on this perception, walking has been considered an over-learned and automatic activity. However, the motor and cognitive systems are interwoven at the cerebrum level [1], thus, gait coordination in walking cannot be completely spontaneous automatic. Rather, it involves continuouscontrol of the body position [2] as well as higher-mental processes such as attention, working memory, decision-making, and problem-solving [3]. In other words, successful gait performance requires appropriate integration of the sensorimotor and cognitive systems [4]. When a person fails to combine the two systems successfully, the risk of falls may be elevated. Previous studies have shown that a concurrent cognitive load can affect postural control such as gait and result in injurious falls [1, 5–8].

As gait performance requires the coordination of sensorimotor and cognitive systems, gait instability is commonly observed in people with Alzheimer’s dementia (AD) [9-14]. The high incidence of gaitinstability among people with dementia has resulted in recommendations for providing fall-prevention training [15, 16]. In addition, gait compromise in people with AD may be an early sign of the disease [9, 13, 17–19]. To date, the majority of studies investigating gait performance in people with AD have employed a single-task paradigm, which involves walking only. However, dual-task gait assessment such as walking and talking testing can provide more practical insight given that people commonly combine the two activities in their daily lives [14]. This combination creates a commonly occurring dual-task activity that requires both motor control and cognitive performance [20]. The manipulation of the concurrent cognitive load (i.e., talking) affects gait [21] and thus, the gait assessment under different levels of cognitive load (i.e., different speech tasks) may be useful to show the impact of cognitive load on gait. Concurrent walking and talking is more taxing for people with AD because of their cognitive impairments [14, 20, 21].

Given the relationship between cognitive and walking performances, authors of recent studies have started to employ dual-task walking assessment and have suggested adopting the dual-task paradigm as a screening tool for cognitive impairment. For example, Rosso and colleagues proposed that dual-task walking assessment can be used as a risk assessment tool for mild cognitive impairment (MCI) [22]. Similarly, Mancioppi and colleagues stated that a dual-task assessment incorporating cognitive and motor tasks is effective for MCI diagnosis [23]. Finally, de Oliveira Silva and colleagues suggested that poor dual-task gait performance should be considered as a functional screening tool for dementia [24].

Despite the usefulness of a dual-task walking assessment to screen cognitive impairment, it is still unclear whether the dual-task assessment significantly improves the accuracy in detecting AD, compared to the single-task paradigm. In order to suggest the inclusion of dual-task gait performance in AD screening/diagnosis tools, it should be clear that the dual-task assessment is more desirable than the single-task assessment. Therefore, this study was designed to investigate the predictive power of gait performance both as a single-task and when gait is paired with a concurrent cognitive load in a dual-task. It is hypothesized that 1) gait performance of people with AD will deteriorate when completing a dual-task of walking and talking in comparison to the single-task of walking without a cognitive load, showing greater decrements when the dual-task demand is increased, and 2) the addition of a concurrent cognitive load will yield a significant improvement in AD detection.

METHODS

Participants

Fourteen individuals with AD and fifteen healthy older adults participated in this study, approved by the institutional review board at Ohio University. The individuals with AD were those who 1) were diagnosed with probable AD per the NINCDS-ADRDA Work Group procedures [25] but with no additional neurological diagnosis, 2) obtained a score representing mild to moderate dementia (a score between 80 and 129) on the Dementia Rating Scale 2 (DRS-2) [26], and 3) had no history of injurious falls within 12 months of participation. The healthy older adults were eligible to participate in this study if they obtained scores in the normal range (≥26) using the Montreal Cognitive Assessment (MoCA) [27]. A MoCA score of 26 and higher is considered not pathological [27]. In addition, the healthy participants were required to have no history of injurious falls in the past 12 months.

All participants exhibited sufficient vision and hearing to respond appropriately to the orally delivered study instructions. In addition, they were able to walk 580 cm with or without assistance. The average age of the people with AD was 78.03 years (standard deviation [SD] = 12.06) with an average of 15.81 years (SD = 2.48) of education. This group obtained a mean score of 88.14 (SD = 7.07) on the DRS-2. For healthy older adults, the mean age was 72.71 years (SD = 11.86) with 15.92 years (SD = 2.08) of education. On average, the healthy older adults scored a 27.73 (SD = 1.29) on the MoCA, which was above the cutoff score (26) for possible cognitive impairment. The two groups did not differ significantly in age (t(25) = 1.08, p = 0.23) or education (t(25) = 0.61, p = 0.53). The participants’ demographic information can be found in Table 1.

Table 1

Demographic information of participants

	People with AD	Neurologically Healthy Controls
Number of Participants	14 (9 Women, 5 Men)	15 (10 Women, 5 Men)
Walking Assistance	2 walker, 1 cane, 1 human assistance	1 cane
Age (SD), y	78.03 (12.06)	72.71 (11.86)
Education (SD), y	15.81 (2.48)	15.92 (2.08)
Duration of disease, y	5.11 (2.08)	Not Applicable
Cognitive Test Sore	DRS-2, 88.14 (7.07)	MoCA, 27.73 (1.29)

SD, standard deviation; DRS-2, Dementia Rating Scale-2; MoCA, Montreal Cognitive Assessment.

Instrumentation

The GAITRite© Portable Walkway System from CIR Systems, Inc., was employed to obtain quantified gait performance from each participant. The walkway system consists of an electronic walkway mat and the gait analysis software. The walkway mat is 580 cm-long and 88 cm-wide and encapsulates 18,432 sensors in an active area 488 cm-long and 61 cm-widewith 1.27 cm between the sensors. The sensors record 24 temporal and 14 spatial gait parameters from each footfall at a 120 samples per second and a temporal resolution of 18.75 ms. The GAITRite© system is allows walkers to be tested using walking aids of their choice. The gait analysis software isolates traces of walking aids that are separate from footfalls and erases them. When signs of walking aids are not identified correctly, they can be manually removed. The GAITRite© Portable Walkway System exhibits strong validity and reliability. McDonough and colleagues found excellent intraclass correlation coefficient (ICC) between paper-and-pencil and GAITRite© -measured spatial parameters (ICC >  95) and between video-based and GAITRite© -measured temporal parameters (ICC >  93) [28]. Bilney and colleagues reported that GAITRite© system had strong validity and test-retest reliability for specific gait parameters when compared to the Clinical Stride Analyzer, a valid and reliable tool for gait assessment (e.g., ICC for velocity, stride length, and cadence between the two systems = 0.99) [29]. Parallel results were found by Webster and colleagues who stated that the GAITRite© system had excellent validity and reliability for specific gait parameters (e.g., velocity, cadence, step time variables, etc.) in comparison to the Vicon-512, a 3-demensional motion analysis system. (e.g., ICC >  92 and repeatability coefficients between 1.0% and 5.9% of mean values for velocity) [30].

Procedures

This study followed the study protocol used in a previous study [14]. Participants were directed to complete walks on the 980 cm-long GAITRite© mat including 200 cm off either end for acceleration and deceleration under three different cognitive load conditions: the baseline (single-task), low cognitive load (dual-task), and high cognitive load (dual-task)conditions. In the baseline condition, participants were asked to walk along the GAITRite© walkway mat normally without talking. The low cognitive load condition was defined as walking while counting numbers by ones. For this task, a two-digit number was randomly selected as a starting number and given to each individual. Lastly, the high cognitive load condition consisted of walking while generating as many words in a given category (e.g., animal) as possible. This task was drawn from the category fluency test by Benton [31].

Prior to completing the three walking conditions, each participant had an opportunity(es) to practice the dual-task conditions while seated and walking to ensure that s/he understood the tasks. Participants were instructed to use any kind of walking assistance (e.g., human assistance or a cane) as needed to represent their daily ambulation. All participants were able to complete the dual-task walking. On average, individuals with AD counted 10 numbers (SD = 1.6) and generated 4 words (SD = 2.0) while healthy controls counted 12 numbers (SD = 2.1) and generated 8 words (SD = 1.3) under the cognitive load conditions. Each condition was repeated two times, and the three walking conditions were distributed randomly to account for potential order effects. Breaks were offered to the participants as many times and at any time during the experiment session.

Analyses

For statistical analyses, four gait parameters were selected as dependent variables: functional ambulation profile (FAP), stride length (SL), velocity, and double support time (DST). The FAP is a composite gait score that ranges from 0 to 100; a score between 95 and 100 is typical for healthy adults. First described by Nelson [32], the FAP is widely used to evaluate the stability of gait. Although FAP is anaggregate score considering various gait parameters, other important parameters such as SL and DST are not factored into the FAP calculation. In addition, for the FAP computation, a person’s gait velocity is normalized after taking the person’s leg length into account. Therefore, SL, DST, and the direct measurement of gait velocity were selected for separate analyses.

In order to determine the predictive power of gait performance with three different task settings (baseline, low cognitive load, and high cognitive load) to the incidence of AD, multiple logistic regression analyses were conducted with the incidence of AD as the dependent variable (Responses: Yes/No), and one of the four selected gait parameters combined with single- or dual-task conditions as the independent variables (e.g., FAP under the baseline condition, FAP under the low cognitive load condition, and FAP under the high cognitive load condition, etc.). Furthermore, to examine whether the gait performance under the dual-task conditions can account for greater variance in the incidence of AD, the performance at the baseline, low and high cognitive load conditions was entered sequentially for each gait parameter, resulting in several different models. This was followed by a model comparison based on residual deviation with the model with only single-task performance as the reference. These analyses were conducted for each of the four gait parameters separately. The analyses were conducted using R version 3.6.3.

RESULTS

Functional ambulation profile

All models of the Omnibus test for multiple logistic regression indicated that FAP both under the single- and dual-task conditions accounted for a significant amount of variability between the two groups of participants (Table 2). Furthermore, Table 2 shows a 15% proportional improvement in pseudo R2 from model 1 (reference model) to model 4 (full model). Thus, the predictive power increased significantly when the FAP score under the high cognitive load condition was added to the reference model with or without the score under the low cognitive load condition. Similarly, Table 3 shows a model comparison analysis based on residual deviance. Specifically, model 3 utilizing the FAP scores under the baseline and high cognitive load conditions (p = 0.001) and the model 4 using the FAP scores under all of the three conditions (p = 0.004) were significantly better than the reference model, in which only FAP under the baseline condition was included, or the model 2 where the FAP scores under the baseline and low cognitive load conditions were taken into account. The results are also illustrated in Fig. 1.

Table 2

Omnibus Test of Functional Ambulation Profile (FAP)

Model	IVs	McFadden Adjusted R2	Cox & Snell R2	Nagel R2	p
1	FAP1	0.63	0.64	0.85	< 0.0001
2	FAP1 + FAP2	0.63	0.66	0.88	< 0.0001
3	FAP1 + FAP3	0.85	0.75	1	< 0.0001
4	FAP1 + FAP2 + FAP3	0.80	0.75	1	< 0.0001

Table 3

Model Comparison of Functional Ambulation Profile (ref. = model 1)

Model	Residual Deviance	AIC	p
1	10.71	14.71	–
2	8.93	14.93	0.1815
3	0.00	6	0.001064**
4	0.00	8	0.004716**

**p < 0.01.

Fig. 1

Functional Ambulation Profile (FAP) by cognitive load.

Velocity

Figure 2 shows that both groups walked slower when taxed with a higher cognitive load. On average, people with AD walked at a rate of 48.47 cm/s (SD = 17.35) under the baseline condition, 42.76 cm/s (SD = 15.11) under the low cognitive load condition, and 38.54 cm/s (SD = 13.26) under the high cognitive load condition. Similarly, healthy older adults’ walking velocity was 103.63 cm/s (SD = 19.28) under the baseline condition which decreased slightly to 100.17 (SD = 19.24) under the low cognitive load condition and to 76.31 cm/s (SD = 13.26) under the high cognitive load condition.

Fig. 2

Velocity by cognitive load.

Similar to FAP, both single- and dual-task velocity explained significant amount of variability in predicting AD (for all models, p <  0.001, see Table 4). Although adding the dual-task velocity improved the model fit, the improvement was not statistically significant as detailed in Table 5.

Table 4

Omnibus Test of Velocity (VEL)

Model	IVs	McFadden Adjusted R2	Cox & Snell R2	Nagel R2	p
1	VEL1	0.66	0.65	0.87	< 0.0001
2	VEL1 + VEL2	0.67	0.68	0.90	< 0.0001
3	VEL1 + VEL3	0.66	0.67	0.90	< 0.0001
4	VEL1 + VEL2 + VEL3	0.62	0.68	0.90	< 0.0001

Table 5

Model Comparison of Velocity (ref. = model 1)

Model	Residual Deviance	AIC	p
1	9.67	13.67	–
2	7.40	13.40	0.1319
3	7.62	13.62	0.1525
4	7.31	15.31	0.3078

Stride length

In accordance with FAP and velocity, participants’ SL decreased for cognitive load (see Fig. 3). People with AD walked with a mean SL of 84.26 cm (SD = 24.66) under the baseline condition, which was shortened to a mean of 76.85 cm (SD = 23.65) under the low cognitive load condition, then to 68.53 cm (SD = 14.99) under the high cognitive load condition. The healthy controls’ SL was more persistent across cognitive load conditions than the AD group’s as the baseline condition SL was 119.26 cm (SD = 16.01), which decreased to 118.54 cm (SD = 15.92) and to 112.25 cm (SD = 10.10) under low and high cognitive load conditions respectively.

Fig. 3

Stride length by cognitive load.

The logistic regression analysis showed that the single-task SL alone did not predict the AD status of the participant, but any combination of single- and dual-task SL did (i.e., Models 2 and 4, p <  0.001; Model 3, p = 0.001, as in Table 6). Similarly, when compared to the reference model, the single-task SL combined with the SL under low cognitive load (p <  0.01), with the SL under high cognitive load (p <  0.001), or with the SL under both low and high cognitive load (p <  0.001) conditions significantly improved the model fit (Table 7).

Table 6

Omnibus Test of Stride Length (SL)

Model	IVs	McFadden Adjusted R2	Cox & Snell R2	Nagel R2	p
1	SL1	0.28	0.41	0.54	0.123
2	SL1 + SL2	0.41	0.54	0.72	< 0.0001
3	SL1 + SL3	0.85	0.75	1	0.001
4	SL1 + SL2 + SL3	0.80	0.75	1	< 0.0001

Table 7

Model Comparison of Stride Length (ref. = model 1)

Model	Residual Deviance	AIC	p
1	37.79	29.13	–
2	17.59	23.59	0.006046**
3	< 0.000	6	< 0.0001***
4	< 0.000	8	< 0.0001***

**p < 0.01; ***p < 0.001.

Double support time

Given the nature of this parameter, participants’ DST was in an inverse pattern to the other three parameters. Namely, DST of both people with AD and healthy older adults increased with higher cognitive load. People with AD needed a mean DST of 0.950 s (SD = 0.784) to complete walking along the 580 cm walkway mat under the baseline condition. The mean DST increased to 1.143 s (SD = 0.767) under the low cognitive load condition and to 1.409 s (SD = 0.738) under the high cognitive load condition. The healthy older adults walked with an average DST of 0.357 s (SD = 0.081) under the baseline condition and 0.377 s (SD = 0.084) under the low cognitive load condition. The DST for the healthy older adults increased to 0.497 s (SD = 0.106) under the high cognitive load condition (see Fig. 4). Similar to FAP and velocity, significant improvements in identifying people with AD occurred for both single- and dual-task DSTs (all p <  0.001, see Table 8). The model comparison revealed that adding one or both of the dual-task DST to the reference model significantly improved the model fit (Table 9). More specifically, adding DST under low cognitive load condition to the baseline model yielded a significant improvement (p <  0.01), so did adding DST at under the high cognitive load condition (p <  0.001), and adding both (p <  0.001).

Fig. 4

Double support time by cognitive load.

Table 8

Omnibus Test of Double Support Time (DST)

Model	IVs	McFadden Adjusted R2	Cox & Snell R2	Nagel R2	p
1	DST1	0.55	0.59	0.79	< 0.0001
2	DST1 + DST2	0.70	0.69	0.92	< 0.0001
3	DST1 + DST3	0.85	0.75	1	< 0.0001
4	DST1 + DST2 + DST3	0.80	0.75	1	< 0.0001

Table 9

Model Comparison of Double Support Time (ref. = model 1)

Model	Residual Deviance	AIC	p
1	14.10	18.18	–
2	6.10	12.10	0.004486**
3	< 0.000	6	0.0001663***
4	< 0.000	8	0.0008343***

**p < 0.01; ***p < 0.001.

DISCUSSION

The present investigation was designed to demonstrate the usefulness of adopting the dual-task gait assessment as a component for AD diagnosis. Two hypotheses were developed in alignment with the purpose: First, gait performance of people with AD will be compromised when they are cognitively taxed, and second, the dual-task gait assessment will improve the accuracy for screening AD than the single-task gait assessment. The findings of the current study fully confirmed the first hypothesis and partially the second hypothesis.

It was evident that concurrent cognitive load affected gait in people with AD: all of the four gait parameters (i.e., FAP, velocity, SL, and DST) were compromised as cognitive demands increased. As Kahneman proposed, the simultaneous activity of walking and talking requires cognitive resource allocation [33], specifically executive function and attention divided between gait and cognition [34-39]. The ability to share the cognitive resource is particularly impaired in people with AD [40-42]. Neuroimaging studies have shown that brain networks are shared by cognitive and motor control in the frontal and temporal areas, which results in poorer gait performance such as slow gait (velocity) [21, 43–45]. Due to the increased load on cortical activity for complicated situations when gait stability is affected, people with dementia have difficulty allocating sufficient cognitive resources available in the frontal or temporal lobes [44]. These situations may result in a higher risk and prevalence of injurious falls among people with AD [3]. This claim reinforces studies showing poorer gait performance associated with damage/atrophy in the prefrontal and hippocampal regions [45-48]. More specifically, these authors found that reduced walking speeds share neural substrate such as smaller hippocampus [48-51]; accumulation of amyloid-β [51]; hyperintensities in the subcortical regions [52]; and stride length variability with Apolipoprotein E4 [53, 54]. In sum, the dual-task gait assessment has shown its effectiveness as a potential dementia diagnostic tool [14, 55–57].

In the present study adoption of a dual task in gait assessment appeared to be more useful for AD screening than a single task of walking without a concurrent cognitive load. In general, the addition of the concurrent cognitive load improved the probability of detecting AD significantly. Three of the four selected gait parameters (i.e., FAP, SL, and DST) explained a larger proportion of the variance and had higher odds ratios under the high cognitive load condition of simultaneous walking and category naming. For DST, factoring in counting numbers by ones also improved the model fit. However, for velocity neither the low nor high cognitive load increased the probability of detecting AD. It should be noted that people with AD in this study slowed their gait when cognitively taxed. This finding is consistent with findings of previous studies showing slow gait speed in people with AD or other dementias when performing a dual-task activity [20, 21, 36, 37, 58]. This current investigation revealed that including the dual-task gait velocity, either under the low or high cognitive load condition, did not significantly improve the model fit over the single-task gait velocity. This non-significant effect of adding the dual-task gait velocity on predicting incidence of AD may be attributed to the small sample size or the experimental tasks (i.e., counting numbers and category naming) that do not represent daily living activity. Alternatively, the short walkway mat might have not been adequate to induce effects of high cognitive load on walking, as people often recall category members well in the beginning but struggle after they name several [59]. As an option, future studies may consider weighing various gait parameters differently for a more accurate AD screening/diagnosis. To date, there has been no study investigating how to weigh different gait parameters as a component of an AD diagnosis. However, when Meilán et al. [60] weighed different speech parameters to discriminate AD, they reported 84.8% accuracy. Therefore, scrutinizing impacts of different gait parameters on AD screening/diagnosis may yield a more accurate result.

Several limitations exist in this study: First, the small sample size (total n = 29) makes this study underpowered for the logistic regression analyses and the specific subpopulation recruited for this study (i.e., people with mild to moderate AD with limited consideration of functional differences) reduces the generalizability of the current results. The insufficient power may lead to a reduced reproducibility of the findings or may even result in yielding unstable and/or null findings. Therefore, future studies including more participants from more diverse populations are warranted. In fact, a number of studies have proposed that poorer gait performance may be associated with non-Alzheimer’s type dementia [5, 61, 62]. Future studies that are sufficiently powered for logistic regression analyses will provide stronger evidence to support the findings of previous studies. In addition, studies that consider functional differences in people with different levels of cognitive functioning will be useful. Second, participants of this study were instructed to use any kind of walking aids that they typically used for their daily ambulation. When a participant needed human assistance, the requested assistance was provided to the degree that does not impede or benefit the person’s walking performance. Although wheeled-walker may increase walking speed, [e.g., 63, 64], stride length [64] and swing time [64], the increase in walking speed is not significant when a person walks with a cane or crutch [63]. The limited data on the effects of walking aids on parameters of gait make it difficult to draw a generalized conclusion. Thus, future dual-task studies should include samples of people who use a variety of walking aids, including human assistance. Third, walking along the short walkway mat while counting numbers or completing a naming task may not be sufficient to represent daily walking performance. Thus, it is recommended that future investigations adopt more real-life dual-task activities such as spontaneous conversation and consider assisted walking and independent walking groups separately. Finally, investigations of gait parameters other than the four selected ones for this investigation may add valuable information for determining the optimum gait measurements to adopt for dual-task assessments for people with dementia.

Despite the limitations, the current investigation adds theoretical and empirical evidence to the literature emphasizing the role of dual-task gait assessments in an AD assessment battery. The current results indicate the challenges experienced by people with AD whose cognitive reserve is less available when walking and completing a simultaneous oral task. These data indicate that the dual-task gait assessment may supplement the current AD screening/diagnostic tools.