Digital Biomarkers of Cognitive-Frailty – The value of detailed gait assessment beyond gait-speed

Abstract

Background:

Cognitive-frailty (CF), defined as simultaneous presence of cognitive-impairment and physical-frailty, is a clinical symptom in early-stage dementia with promise in assessing the risk of dementia. The purpose of this study was to use wearables to determine the most sensitive digital gait biomarkers to identify CF.

Methods:

Of 121 older adults (age=78.9±8.2years, body mass index=26.6±5.5kg/m²) who were evaluated with a comprehensive neurological exam and the Fried Frailty Criteria, 41 participants (34%) were identified with CF and 80 participants (66%) were identified without CF. Gait performance of participants was assessed under single-task (walking without cognitive distraction) and dual-task (walking while counting backward from a random number) using a validated wearable platform. Participants walked at habitual speed over a distance of 10 meters. A validated algorithm was used to determine steady-state walking. Gait parameters of interest including steady-state gait speed, stride length, gait cycle time, double-support, and gait unsteadiness. In addition, speed and stride length were normalized by height.

Results:

Our results suggest that compared to the group without CF, the CF group had deteriorated gait performances in both single-task and dual-task walking (Cohen’s effect size d=0.42–0.97, p<0.050). The largest effect size was observed in normalized dual-task gait speed (d=0.97, p<0.001). The use of dual-task gait speed improved the area-under-curve (AUC) to distinguish CF cases to 0.76 from 0.73 observed for the single-task gait speed. Adding both single-task and dual-task gait speeds did not noticeably change AUC. However, when additional gait parameters such as gait unsteadiness, stride length, and double-support were included in the model, AUC was improved to 0.87.

Conclusions:

This study suggests that gait performances measured by wearable sensors are potential digital biomarkers of CF among older adults. Dual-task gait and other detailed gait metrics provide value for identifying CF above gait speed alone. Future studies need to examine the potential benefits of gait performances for early diagnosis of CF and/or tracking its severity over time.

Introduction

Dementia causes great stress to our society, health care system, and family caregivers [1–3]. Today 47.5 million people worldwide have dementia, and this number will increase to 75.6 million by 2030 and 135.5 million by 2050 [1]. This “dementia epidemic” [4] creates an urgent need for a robust and rapidly-administered cognitive assessment tool, which is capable of identifying individuals in the earliest stage of cognitive decline and measuring subtle changes in cognitive-motor performance over time [5]. A recent study [6] demonstrated that measures of physical frailty identify individuals with Alzheimer disease (AD) who are at greater risk for cognitive decline and loss of independence. Despite the evidence, cognitive impairment and frailty are rarely assessed, and when assessed, are independently assessed, mainly because conventional cognitive performance assessment tools do not account for motor capacity (an indicator of physical frailty). Conversely, conventional physical frailty assessment tools do not account for cognitive performance.

Identifying persons who experience both motor and cognitive decline, known as cognitive-frailty (CF), a recently recognized symptom in early-stage dementia [4, 7, 6, 8], may have a greater prognostic value in assessing dementia risk. This is because the combination could identify a group in whom gait speed decline is, at least in part, caused by neurodegenerative pathologic conditions of the central nervous system rather than local musculoskeletal problems, such as sarcopenia or osteoarthritis [9–11]. A growing line of research confirmed that CF syndrome is associated with high risks of cognitive decline over time and is a strong predictor of transition to dementia [12–16, 11, 17, 18]. Despite this evidence, motor function and cognitive performance, if assessed, are assessed independently.

AD neuropathology usually starts in the hippocampus and has a profound impact on memory function. However, AD also affects other areas of the brain, including prefrontal areas and its associated cognitive functions, including divided attention, set-shifting, response inhibition, planning, and organizing [19]. The dual-task paradigm (e.g., dual-task walking) is a method for assessing executive function and divided attention performance. It can be used for quantifying both cognitive function and motor capacity, and is sensitive to identify both frailty and mild cognitive impairment [20, 2]. In older adults without the overt disease, greater dual-task “cost” (e.g., the decrement in gait speed induced by performing a cognitive task in parallel with walking) has been linked to worse executive function [21–25], increased incidence of falls [26, 27], more rapid cognitive decline [2, 28, 29], and amyloid deposition [30]. Dual-task performance often declines before memory impairment, highlighting its value as a tool to identify pre-clinical AD diagnosis [23–25].

The spread of wearable digital technologies in healthcare have provided new opportunities to routinely assess gait performance outside of a gait laboratory, including detailed gait metrics rather than simple gait speed calculated by walking distance and stopwatch-measured walking time [31–33]. In this study, we used a validated wearable platform to investigate potential digital biomarkers measured by gait assessment for capturing CF in older adults. We measured detailed gait metrics, including gait speed (normalized by height), stride length (normalized by height), gait cycle time, double support, dual-task cost, and gait unsteadiness in both single-task and dual-task walking tests. We hypothesized that 1) older adults with CF can be identified using gait performance measured by wearables, and specifically dual-task walking will yield a larger effect at discriminating CF than single-task walking and 2) detailed gait metrics measured by the wearable platform can yield better results than using normalized gait speed alone to determine CF.

Materials and Methods

Study Population

We recruited older adults (age 65 years or older) with and without cognitive impairment. Older adults with cognitive impairment were recruited from the Memory Disorders Clinic at the Banner Sun Health Research Institute, the Senior Care Clinic at Baylor College of Medicine, and Department of Neurology, Neuropsychology Section at Baylor College of Medicine. Participants with cognitive impairment were clinically diagnosed with either amnestic mild cognitive impairment or mild dementia by board-certified neurologists using comprehensive neurological exams and cognitive screens or comprehensive neuropsychological evaluations. We also recruited cognitively intact older adults within the age range of ± 5 years compared to the cognitively impaired older adults.

Participants were excluded from the study if they were non-ambulatory or had a severe gait impairment (e.g., inability to walk 10 meters independently with or without an assistive device); had major foot problems (e.g., major amputation, severe neuropathy, foot deformity, foot arthritis, foot pain, etc.); had other neurological conditions associated with cognitive impairment (stroke, Parkinson’s disease, Huntington’s disease, etc.); had any clinically significant medical or psychiatric condition or laboratory abnormality; had severe visual and/or hearing impairment; had changes in psychotropic or sleep medications in the last 6 weeks; or were unwilling to participate.

All participants signed an approved consent form before participation in this study. This study was approved by the local institutional review boards including Baylor College of Medicine (H- 2521) and at the Banner Sun Health Research Institute (1146563).

Demographics and Clinical Information

Demographics and relevant clinical information for all participants were collected using chart-review and self-report, including age, sex, height, weight, use of walking assistance, pain level, activity level, and fall history. Body mass index was calculated based on height and weight. All participants underwent clinical assessments, including Mini-Mental State Examination (MMSE) [34], Fall Efficacy Scale - International (FES-I) [35], Center for Epidemiologic Studies - Depression (CES-D) [36], and Fried Frailty criteria [37]. The FES-I was used to measure self-report concern about falling. A cutoff of FES-I score of 23 or greater was used to identify participants with high concern about falling [35]. The CES-D scale was used to measure self-reported depression symptoms. A cutoff of CES-D score of 16 or greater was used to identify participants at risk for clinical depression [36].

Determination of Physical Frailty and Cognitive-Frailty

We used Fried Frailty criteria to determine physical frailty [37]. In summary, the presence or absence of physical frailty phenotypes, including unintentional weight loss, weakness (grip strength), slow gait speed (15-foot gait test), self-reported exhaustion, and self-reported low physical activity, were used to assess physical frailty [37]. Participants without phenotype presence were classified as physically robust. Participants with 1 or more positive phenotypes were considered physically frail [37]. In this study, instead of using frailty categories (robust, pre-frail, and frail), we use the term of frailty severity, which is defined by number of phenotypes present [38]. The status of cognitive impairment was confirmed by comprehensive neurological exams and cognitive screens or comprehensive neuropsychological evaluations by board-certified neurologists. The cognitive-intact participants were defined as those without any clinical diagnosis of cognitive impairment and MMSE of 24 or greater [39]. Participants presenting with both cognitive impairment and presence of any physical frailty phenotype were classified into a CF group. Others were classified into a non-CF (NCF) group, which includes all robust participants with or without cognitive impairment.

Gait Test

For all participants, five wearable sensors (LegSys™, BioSensics, Newton, MA, USA) were attached to the lower back, both the left and right thighs and lower shins to quantify gait metrics of interest (Figure 1). Participants were asked to walk with their habitual gait speed for 10 meters without any distraction (single-task walking), using the protocol described in our previous studies [40, 41]. Then they were asked to repeat the test while loudly counting backward from a random two-digit number (dual-task walking: walking + working memory test) [2]. The two walking tests together took less than 1 minute. We used validated algorithms [42–45] to identify the gait initiation and steady state walking. All gait metrics were calculated during the steady-state phase of walking, using validated algorithms [46–49]. All gait metrics of interest were averaged during the entire steady state phase. Normalized gait metrics were calculated by the average gait metrics divided by body height. The evaluated gait metrics in this study included normalized gait speed, normalized stride length, gait cycle time, double support, dual-task cost, gait speed unsteadiness (coefficient of variability of gait speed), and stride length unsteadiness (coefficient of variability of stride length). Dual-task cost was defined as the percentage of normalized gait speed reduction between single-task walking and dual-task walking.

Figure 1:

Gait was assessed using five wireless wearable sensors attached to the lower back, both the left and right thighs and lower shins. We assessed gait under two walking conditions: 1) Single-task walking: Participants were asked to walk with their habitual gait speed for 10 meters without any distraction 2) Dual-task walking Participants were asked to walk with their habitual gait speed for 10 meters, while loudly counting backward from a two-digit random number

Data and Statistical Analysis

All continuous data were presented as mean ± standard deviation. All categorical data were expressed as percentages. Analysis of variance (ANOVA) was used for between-group comparison of continuous demographics and clinical data. Analysis of Chi-square was used for comparison of categorical demographics and clinical data. Analysis of covariance (ANCOVA) was employed to compare differences between groups for gait metrics, adjusting for gender. A 2-sided p<0.050 was considered to be statistically significant. The effect size for discriminating between groups was estimated using Cohen’s d effect size and represented as d in the Results section. Values were defined as small (0.20–0.49), medium (0.50–0.79), large (0.80–1.29), and very large (above 1.30) [50]. Values less than 0.20 were classified as having no noticeable effect [50]. Linear regression analysis was conducted for the four outcome measures, including MMSE, frailty phenotype, normalized single-task and dual-task gait speeds, to examine correlations between each outcome measure; correlation analyses (Pearson correlation or Spearman’s correlation, depending on data normality) for pairs of these outcome measures were observed. Binary logistic regression analysis was employed to examine the relationship between gait metrics and CF. Four models were built for prospective identification of CF. In Model 1, we used only normalized single-task gait speed as the independent variable. In Model 2, we used only normalized dual-task gait speed as the independent variable. In Model 3, we used both normalized single-task and dual-task gait speed as independent variables. Then, to examine additional values of detailed gait metrics, in Model 4, independent variables included normalized gait speed and gait speed unsteadiness, normalized stride length and stride length unsteadiness, gait cycle time, and double support, in both single-task and dual-task walking. The receiver operating characteristic (ROC) curve and area-under-curve (AUC) were calculated for the four models. All data and statistical analyses were performed using MATLAB (The MathWorks, Natick, MA, USA) and IBM SPSS Statistics (IBM, Armonk, NY, USA).

Results

One hundred and twenty-one participants were recruited. According to their cognitive status and physical frailty status, 41 (34%) participants were classified into the CF group and 80 (66%) participants were classified into the NCF group. As summarized in Table 1, the CF group had a lower prevalence of women (p=0.018) and were taller (p=0.045) than the NCF group. The average MMSE score of the CF group was significantly lower than the NCF group (p<0.001). The CF group had a higher prevalence of high concern about falling (p=0.021), higher CES-D score (p=0.013), and significantly higher prevalence of low activity (p=0.015) than the NCF group. No between-group difference was observed for other characteristics including age, weight, body mass index, prevalence of using walking assistance/aids, pain level, FES-I score, prevalence of at risk for clinical depression, or fall history (p>0.050).

Table 1.

General characteristics of the study groups.

	NCF (n=80)	CF (n=41)	p-value
Demographics
Age, years	78.7 ± 8.5	79.2 ± 7.5	0.725
Sex (female), %	66%	44%	0.018
Height, m	1.63 ± 0.10	1.67 ± 0.11	0.045
Weight, kg	71.7 ± 17.3	74.1 ± 18.5	0.474
BMI, kg/m2	26.8 ± 5.8	26.2 ± 4.8	0.592
Using walking assistance, %	30%	17%	0.123
Clinical characteristics
MMSE, units on a scale (max 30)	29.0 ± 1.3	24.7 ± 4.8	<0.00
Pain level, units on a scale (max 10)	1.8 ± 4.3	1.8 ± 2.2	0.996
Concern for fall (FES-I score), units on a	26.4 ± 10.5	22.2 ± 5.8	0.069
High concern about falling (FES-I≥23), %	51%	29%	0.021
Depression (CES-D score), units on a scale	8.2 ± 7.2	4.9 ± 6.1	0.013
At risk for clinical depression (CES-D≥16), %	21%	7%	0.051
Low activity, %	15%	34%	0.015
Had fall in last 12-month, %	35%	32%	0.717

NCF: non cognitive frailty group

CF: cognitive frailty group

FES-I: Fall Efficacy Scale - International

CES-D: Center for Epidemiologic Studies Depression

Significant difference between groups were indicated in bold

As summarized in Table 2, during single-task walking, all gait metrics, except double support, reached statistical significance to identify older adults with CF (d=0.42–0.79, p<0.050), with the largest effect size observed in gait speed unsteadiness (d=0.79, p<0.001). Overall, the effect sizes for between group differences were larger during dual-task walking compared to single-task. All dual-task gait metrics reached statistical significance to identify older adults with CF (d=0.62–0.97, p<0.010), with the largest effect size observed in normalized gait speed (d=0.97, p<0.001). In addition, dual-task cost reached a statistically significant level when compared between the NCF group and the CF group (d=0.68, p=0.001).

Table 2.

Gait performance for groups with and without cognitive frailty.

	NCF (n=80)	CF (n=41)	d *	p-value*
Single Task Walking
Normalized Gait Speed, height/s	0.65 ± 0.14	0.55 ± 0.13	0.73	0.001
Normalized Stride Length, % of height	71.58 ± 10.89	64.09 ± 11.71	0.66	0.002
Gait Cycle Time, s	1.14 ± 0.15	1.21 ± 0.13	0.42	0.037
Double Support, %	24.85 ± 5.21	25.60 ± 5.92	0.21	0.283
Single Task Gait Speed Unsteadiness, %	5.80 ± 3.64	8.96 ± 5.12	0.79	<0.001
Single Task Stride Length Unsteadiness, %	4.09 ± 3.19	6.75 ± 4.92	0.68	0.001
Dual Task Walking
Normalized Gait Speed, height/s	0.60 ± 0.15	0.45 ± 0.15	0.97	<0.001
Normalized Stride Length, % of height	71.71 ± 11.74	61.21 ± 11.36	0.91	<0.001
Gait Cycle Time, s	1.25 ± 0.19	1.46 ± 0.32	0.76	<0.001
Double Support, %	26.22 ± 5.31	29.91 ± 7.52	0.62	0.002
Dual Task Gait Speed Unsteadiness, %	6.40 ± 2.62	8.56 ± 3.90	0.70	<0.001
Dual Task Stride Length Unsteadiness, %	3.95 ± 1.82	6.10 ± 3.89	0.79	<0.001
Dual Task Cost, %	7.70 ± 11.38	17.82 ± 16.59	0.68	0.001

NCF: non cognitive frailty group

CF: cognitive frailty group

^*:Results were adjusted by gender

Significant difference between groups were indicated in bold

Effect sizes were calculated as Cohen’s d

Figure 2A color-maps the mutated data to normalized single-task gait speed, whereas Figure 2B color-maps the mutated data to normalized dual-task gait speed. The brighter color (red) presents the higher normalized gait speed, and the darker color (dark blue) presents the lower normalized gait speed. The NCF group is shown in the green region, and the CF group is shown in the pink region. This illustration suggests increasing in frailty severity (higher number of phenotypes) reduces gait speed during both single-task and dual-task conditions. However, only under dual-task condition, cognitive function decline leads to noticeable decline in gait speed. Results also suggest a significant negative correlation between normalized single-task gait speed and frailty severity (r=−0.569, p<0.001). A significant correlation was also observed between normalized single-task gait speed and MMSE score but with lower effect size (r=0.231, p=0.017). Under dual-task gait a similar negative correlation was observed between normalized dual-task gait speed and frailty severity (r=−0.524, p<0.001). However, the correlation between normalized dual-task gait speed and MMSE score showed a higher effect size compared to single-task gait (r=0.297, p=0.002). Moreover, the results showed a significant and strong correlation between normalized single-task and dual-task gait speeds (r=0.864, p<0.001), and a negative correlation between frailty severity and MMSE score (r=−0.231, p=0.029).

Figure 2:

A) A color-coded illustration of MMSE score (x-axis) and number of frailty phenotype presences (y-axis) with color map based on normalized single-task gait speed. With increase in frailty severity, gait speed is reduced (darker blue color). However, deterioration in cognitive function (lower MMSE) does not necessarily map to slower gait speed under single-task walking. B) A similar color-coded grant as Figure 2A, however with color map based on normalized dual-task gait speed. Unlike to single-task, deterioration in cognitive function led to decrease in normalized dual-task gait speed as well. MMSE: Mini-Mental State Exam; CF: Cognitive-frailty; NCF: Non cognitive-frailty

As illustrated in Figure 3, the AUC (accuracy) for distinguishing CF cases were 0.73 (70.2%), 0.76 (70.2%), 0.76 (69.4%), and 0.87 (76%), respectively for the Model 1 (normalized single-task gait speed), Model 2 (normalized dual-task gait speed), Model 3 (normalized single-task + dual-task gait speed) , and Model 4 (normalized gait speed and gait speed unsteadiness + normalized stride length and stride length unsteadiness + gait cycle time + double support, in both single-task and dual-task walking).

Figure 3).

Illustration of the receiver operating characteristic (ROC) curve for Model 1 (normalized single-task gait speed), Model 2 (normalized dual-task gait speed), Model 3 (normalized single-task + dual-task gait speed) , and Model 4 (normalized gait speed and gait speed unsteadiness + normalized stride length and stride length unsteadiness + gait cycle time + double support, in both single-task and dual-task walking). While normalized dual-task gait speed improved area under curve (AUC) compared to normalized single-task gait speed, combination of both single-task and dual-task did not noticeably improve AUC. However, addition of other gait metrics during both single-task and dual-task walking (Model 4) noticeably improved AUC.

Discussion

To our knowledge, this is the first study to investigate potential digital biomarkers using gait performance measures to assess CF in older adults. The results suggest that older adults with CF have deteriorated gait performance compared to those without CF. We confirmed our hypothesis that CF can be identified using gait performance measures, while dual-task walking had a larger observed effect than single-task walking to identify CF. In addition, combining normalized gait speed and gait speed unsteadiness, normalized stride length and stride length unsteadiness, gait cycle time, and double support in both single-task and dual-task walking into the binary logistic regression model enables one to distinguish between older adults with and without CF. This combined model yields relatively higher accuracy compared to normalized gait speed alone (increasing AUC from 0.76 to 0.87 for detection of CF).

In the past decade, there has been an increased interest in identifying and validating biomarkers for early diagnosis and identification of individuals who are at risk of dementia. However, the use of biomarkers has limitations in many settings [51]. For instance, access to neuroimaging is difficult and the cost of biological biomarkers limits their use [52, 51]. Additionally, the highest prevalence and incidence of dementia in the coming years will be observed in low and intermediate income countries, where the accessibility to expensive biomarkers is limited [53]. Hence, there is a need to increase the accessibility to clinical risk assessment of dementia in community-dwelling older populations [8]. Using low-cost wearable technology to determine CF irrespective of setting (e.g., home, small clinic or residential care home) is a promising predictor of dementia in older populations.

There is increasing evidence that impaired motor capacity, such as slow gait, occurs early in dementia and may precede declines in cognitive tests [54–56]. But the effect size of this association is generally modest [14, 18]. Identifying persons who experience both motor decline and cognitive decline, or CF, may have a greater prognostic value in assessing the risk of dementia. In current practice, clinicians typically use resource-intensive methods to assess cognitive function. The current cognitive screening tools have limited ability to detect subtle changes in cognitive performance over time and the accuracy is highly dependent on the examiner’s experience and the patient’s education level [57, 23]. They also do not provide an objective assessment of the impact of cognitive decline on mobility performance. Moreover, while mounting evidence suggests that knowledge of intra-subject performance variation can significantly augment clinical judgment and care, especially in the early stages of AD, the majority of available neuropsychological assessments are ill-suited for repeat testing within relatively short periods due to the effects of practice, patient’s mood, fatigue, and other influences.

Compared with conventional methods, gait assessment is an objective and time-efficient method (usually takes less than 1 minute) to assess functional ability in older adults. A previous systematic review investigating the association between gait and frailty showed that single-task gait speed has the largest effect size to discriminate between frailty subgroups [58]. Previous studies also demonstrated that dual-task gait speed is efficient in discriminating between individuals with different frailty statuses [59] as well as individuals with different cognitive status [60]. In addition, dual-task gait speed is associated with traditional cognitive measurements [61]. A recent systematic review and meta-analysis of studies using instrumented assessment suggests that mild cognitive impairment related gait changes are most pronounced when subjects are challenged cognitively (dual-task gait) [2]. The results of this study are consistent with these studies. Figure 2A shows that participants with zero or only one positive frailty phenotype had relatively higher single-task walking normalized gait speed (brighter color on the color map). For participants that exhibited higher severity of frailty (those with 3 or 4 positive frailty phenotypes), their normalized single-task gait speeds were reduced (darker color on the color map). However, the normalized single-task gait speed is not sensitive in change along with the decreasing of MMSE score, especially among participants with cognitive impairment. On the other hand, Figure 2B shows that normalized dual-task gait speed is sensitive in reduction along with the decreasing of MMSE score, as well as the increase of presented frailty phenotypes (p<0.001). Among all gait metrics, normalized dual-task gait speed also had the largest effect to discriminate participants with CF (d=0.97, p<0.001).

In this study, instead of using a simple stopwatch, we used a validated wearable platform to measure gait performance. The advantage of using wearable sensors is that we can easily measure detailed gait metrics, such as stride length, gait cycle time, double support, and gait unsteadiness, rather than simple gait speed alone. Although gait speed is still considered the most reliable marker of functional deterioration and physical frailty [62, 63], gait is a complex motor behavior with many measurable facets besides velocity and with an intricate relationship to different aspects of cognition [64]. Previous studies have linked other gait metrics, such as stride length and gait unsteadiness, to cognitive impairment [65, 64, 2]. Our results demonstrate that all gait metrics in dual-task walking can discriminate individuals with CF. In addition, during single-task walking, the gait speed unsteadiness has larger observed effect than normalized gait speed to identify CF. Adding additional gait metrics can increase the power to identify CF in older adults over the use of gait speed alone.

A major strength in this study is using comprehensive neuropsychological evaluations to identify those with cognitive impairment instead of just relying on MMSE to determine those who are at risk of dementia. A recent study suggested that while MMSE has high specificity (94%) to distinguish those with early-stage Alzheimer’s disease (AD) and mild cognitive impairment from age-matched cognitive intact population, it has poor sensitivity (33%) [25]. Thus, just relying on MMSE may be insufficient to accurately determine those with early stage of dementia. Similarly, in our study we used Fried frailty criteria to determine presence of physical frailty instead of using just gait speed. In the absence of complete phenotype assessment, interpretation of frailty results for prediction of prospective adverse events (e.g., risk of mortality) can be narrow and the predictive power might be reduced [66, 67].

A major limitation of this study is the relatively small sample size, which could be underpowered for the clinical conclusion. Another limitation of this study is that we did not perform feature selection or other machine learning techniques to find the optimal model for identifying CF. It is possible that some of gait metrics are dependent and may over-train the classifier. We will perform feature selection and Bootstrap techniques [68] in the future study. In this study, we did not have participants presenting with severe cognitive impairment and severe physical frailty simultaneously. Individuals presenting both severe cognitive impairment and severe physical frailty are usually excluded because of other medical conditions they have, as well as because of safety concerns. In this study, the female percentage in the NCF group was significantly higher than in the CF group, which also resulted in a significant difference in height between the two groups. We provided normalized gait speed and stride length by height in the results. We also adjusted the results by gender to provide a fair comparison. In this study, our cohort with cognitive impairment included older adults with amnestic mild cognitive impairment or cognitive impairment caused by mild dementia. However, cognitive decline may also cause by other neurological disorders such as stroke, Parkinson’s disease, Huntington’s disease, etc. Future studies are needed to confirm the observation of this study to determine CF using gait-derived digital metrics among other neurological conditions.

Conclusion

This study suggests that gait metrics measured by wearable sensors during walking tests are potential digital biomarkers of CF among older adults. Results suggest that older adults with CF have deteriorated gait performances compared to those without CF. Dual-task walking has larger observed effect than single-task walking to identify CF. Besides gait speed, adding detailed gait metrics can improve the power to identify CF. We believe that our findings can inform the future design of technologies aiming at screening digital biomarkers of CF among older adults. Future studies are encouraged to use gait metrics to detect onset of CF, as well as to track its severity changes over time. Future studies are also recommended for the potential use of gait metrics to facilitate timely intervention and evaluate treatment outcomes.