This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mHealth and uHealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
Continuous tracking of ambulatory activity in real-world settings using step activity monitors has many potential uses. However, feasibility, accuracy, and correlation with performance measures in stroke patients have not been well-established.
The primary study objective was to determine adherence with wearing a consumer-grade step activity monitor, the Fitbit Charge HR, in home-going ischemic stroke patients during the first 90 days after hospital discharge. Secondary objectives were to (1) determine accuracy of step counts of the Fitbit Charge HR compared with a manual tally; (2) calculate correlations between the Fitbit step counts and the mobility performance scores at discharge and 30 days after stroke; (3) determine variability and change in weekly step counts over 90 days; and (4) evaluate patient experience with using the Fitbit Charge HR poststroke.
A total of 15 participants with recent mild ischemic stroke wore a Fitbit Charge HR for 90 days after discharge and completed 3 mobility performance tests from the National Institutes of Health Toolbox at discharge and Day 30: (1) Standing Balance Test, (2) 2-Minute Walk Endurance Test, and (3) 4-Meter Walk Gait Speed Test. Accuracy of step activity monitors was assessed by calculating differences in steps recorded on the step activity monitor and a manual tally during 2-minute walk tests.
Participants had a mean age of 54 years and a median modified Rankin scale score of 1. Mean daily adherence with step activity monitor use was 83.6%. Mean daily step count in the first week after discharge was 4376. Daily step counts increased slightly during the first 30 days after discharge (average increase of 52.5 steps/day; 95% CI 32.2-71.8) and remained stable during the 30-90 day period after discharge. Mean step count difference between step activity monitor and manual tally was –4.8 steps (–1.8%). Intraclass correlation coefficients for step counts and 2-minute walk, standing balance, and 4-meter gait speed at discharge were 0.41 (95% CI –0.14 to 0.75), –0.12 (95% CI –0.67 to 0.64), and 0.17 (95% CI –0.46 to 0.66), respectively. Values were similarly poor at 30 days.
The use of consumer-grade Fitbit Charge HR in patients with recent mild stroke is feasible with reasonable adherence and accuracy. There was poor correlation between step counts and gait speed, balance, and endurance. Further research is needed to evaluate the association between step counts and other outcomes relevant to patients, including patient-reported outcomes and measures of physical function.
Stroke is a leading cause of long-term disability in adults [
The ability to passively capture and track continuous ambulatory activity over time in real-world settings for patients with stroke has many potential uses. For example, such tracking may be able to detect relevant changes in a patient’s ability to ambulate more accurately and efficiently than other common strategies, which include self-reported ambulation questionnaires or mobility performance test [
Step counts can be monitored using step activity monitors. The devices developed for research settings have been expensive and difficult to use [
Despite the theoretical potential for the use of easy-to-wear step activity monitors to improve the clinical care and outcomes of patients poststroke, the feasibility and utility of using consumer-oriented step activity monitors in stroke patients are poorly known. The consumer-oriented step activity monitors, Fitbit Charge HR and Garmin Vivosmart, were found to accurately measure step count in 37 patients attending either inpatient or outpatient therapy for stroke who wore the step activity monitors 5 to 10 hours per day for 2 days [
Therefore, we performed a prospective cohort pilot study to assess feasibility of monitoring step counts in ambulatory patients with recent mild ischemic stroke using a consumer step activity monitor, the Fitbit Charge HR. The primary study objective was to determine adherence with wearing a Fitbit Charge HR in home-going ischemic stroke patients during the first 90 days after hospital discharge. Secondary objectives were to (1) determine accuracy of step counts of the Fitbit Charge HR compared with a manual tally; (2) calculate correlations between the Fitbit step counts and the mobility performance scores at discharge and 30 days after stroke; (3) determine variability and change in weekly step counts over 90 days; and (4) evaluate patient experience with using the Fitbit Charge HR poststroke.
All admissions to the stroke service were monitored by a trained research coordinator to identify potentially eligible patients. Prior to hospital discharge, participants were given a Fitbit Charge HR step activity monitor and instructed to wear this continuously throughout the day for 90 days. Participants were asked to complete a diary of their physical activity for the first 7 days postdischarge (See
Informed consent was obtained from all patient participants. The study was approved by the Cleveland Clinic Institutional Review Board.
Participants were recruited through daily screening of the inpatient stroke service at Cleveland Clinic. All patients who met eligibility criteria were approached for consent. Initial inclusion criteria were the following: (1) admission to the Stroke Service with an admission diagnosis of ischemic stroke; (2) reside within Cuyahoga County or 5 surrounding counties; (3) informed consent obtained from the patient, caregiver, or a legal representative; (4) discharged home; (5) mild stroke operationalized as mild disability at discharge as defined by a modified Rankin scale score of 1 to 2; and (6) ambulatory at the time of hospital discharge. The modified Rankin scale is a 1-item clinician-reported scale with scores ranging from 0 (representing no symptoms) to 6 (death) [
Most patients discharged home after stroke from the inpatient stroke service had few neurological deficits as assessed by the clinician-reported National Institutes of Health Stroke Scale [
A total of 55 individuals were approached to participate in the study. Of these, 11 declined participation and 29 no longer met the inclusion criteria after initial interest in participating (
Study enrollment and retention.
The Fitbit Charge HR was provided to patients at the time of enrollment. It is a wrist-worn device with a triaxial accelerometer marketed as a consumer device that displays step count and wirelessly synchronizes with smartphones. Fitbit devices have been shown to provide step counts similar to those provided by research grade accelerometers [
We administered 3 mobility performance tests that are part of the National Institutes of Health (NIH)Toolbox at discharge and at a return visit on Day 30: (1) Standing Balance Test, (2) 2-Minute Walk Endurance Test, and (3) 4-Meter Walk Gait Speed Test. The NIH Toolbox is comprised of numerous standardized royalty-free measures in neurological and behavioral health. It is designed for use in people ages 3 to 85 to allow for comparison of data across multiple studies [
The NIH Toolbox balance test involves the participant maintaining 5 poses for 50 seconds each. Postural sway is recorded for each pose using an accelerometer worn by the participant which synchronizes with the NIH Toolbox app. Normative scores are provided that adjust for age, sex, ethnicity, and education. The average score of the normative population is 50 (SD 10), with higher scores indicating better balance.
The 4-Meter Gait Speed Test measures the time required to walk 4 meters, which is then transformed into gait speed in meters per second. The 2-Minute Walk Endurance Test records the distance that the participant is able to walk back and forth on a 50-foot course in 2 minutes, which is converted to normative scale scores. As with balance scores, higher scores indicate better performance. To determine accuracy of the step activity monitor, we manually counted the number of steps during the 2-Minute Walk Endurance Test using a hand counter at both discharge and Day 30 assessments.
The National Institutes of Health Stroke Scale [
Questions were adapted from those used in previously published research [
The MyInertia system used to access step count data for this study only provided daily step counts. Participant adherence was defined as having a daily step count of 100 or more. This threshold was chosen to minimize the chance of incorrectly classifying participants as nonadherent. Available data from the literature indicates that stroke survivors are sedentary [
Adherence was summarized descriptively and modeled using a logistic Generalized Estimating Equation. A first-order auto-regressive (AR(1)) correlation structure was used to take into account the dependent nature of consecutive days. The intercept of the Generalized Estimating Equation model provided the population-level probability of adherence. This approach to calculating adherence allowed more accurate estimates of error around adherence rate [
To determine accuracy of the Fitbit, we performed a manual count of the number of steps taken during the 2-Minute Walk Endurance Test using a hand tally counter at both discharge and Day 30 assessments. We performed an analysis of differences in the number of steps recorded on the Fitbit during 2-Minute Walk Endurance Tests and manual tally by calculating mean and median differences and mean absolute difference. Differences between device and manually counted steps were graphically displayed using Bland-Altman plots for repeated measures [
Spearman rank correlations were used to examine the relationship between step counts and mobility performance test scores for the weeks following discharge and the Day 30 visit. The mean daily steps of the 7-day period following assessments was used to summarize step counts. Confidence intervals were estimated using the bootstrap with 10,000 repetitions. As per standard NIH Toolbox scoring procedures, these mobility performance tests were adjusted for age, sex, and education to provide normative data for patients falling within similar age, sex, and educational attainment categories [
Variability in steps-per-day was measured over the 90-day study period and for each 7-day period. Previous studies have found that a 7-day monitoring period is required to obtain a stable and representative average of levels of walking activity in healthy people [
Descriptive statistics were used to summarize participant survey responses.
The suitability of the definition of daily adherence used in our analysis was evaluated using self-reported daily diaries during the first 7 days of monitoring. Daily diary entries were considered adherent if a patient reported wearing the device for more than 12 hours per day. Diary entries indicating that the Fitbit was not worn for 12 or more hours were categorized as nonadherent. Agreement between the diaries and adherence was summarized as a percentage. In a sensitivity analysis, the percentage of participants identified as “adherent” was calculated for different daily step count thresholds.
In addition, a sensitivity analysis was conducted to evaluate the effect of varying adherence thresholds on correlations between mobility performance scores and step counts. This was done by computing a series of average daily step counts over the 7-day period immediately following the performance measure assessment using only the days where the total step count was at or above specific thresholds (no step count restriction, ≥100, ≥500, ≥1000, ≥2000, ≥3000, ≥4000, and ≥5000 steps). Correlations between the mobility performance scores and each series of daily step counts was then calculated.
Sample size calculations were based on simulations to determine power required for the primary objective. We assumed that the percentage of adherent days could vary from 50% to 96% [
Power for secondary objective 2, the correlation between mobility performance scores and step counts, was estimated using 1000 simulations of bivariate normal distributed data with correlations of 0.65. Power for a sample size of 15 was estimated to be greater than 60% to detect a correlation greater than zero.
All computations were done in R, version 3.4.1 [
Participants had a mean age of 54.4 years (SD 12.1, range 34-81 years) and 60% (9/15) were male. Most had at least some college education (11/15, 73.3%). Participants were 60% (9/15) non-Hispanic White. Discharge National Institutes of Health Stroke Scale score was 0 or 1 in 73% (11/15) of participants (
Scores for the Standing Balance Test, 2-Minute Walk Endurance Test, and 4-Meter Gait Speed Test measured at discharge and Day 30 are presented in
Participant characteristics (N=15).
Variable | Value | |
|
||
|
Mean (SD) | 54.4 (12.1) |
|
Range | 34-81 |
|
||
0 | 6 (40) | |
1 | 5 (33.3) | |
3 | 3 (20) | |
4 | 1 (6.7) | |
Male, n (%) | 9 (60) | |
White Non-Hispanic, n (%) | 9 (60) | |
|
||
|
Married | 9 (60) |
|
Divorced | 1 (6.7) |
|
Widowed | 1 (6.7) |
|
Single | 4 (26.7) |
|
||
|
0 | 2 (13.3) |
|
1 | 8 (53.3) |
|
2 | 5 (33.3) |
Received PTb and/or OTc post-discharge, n (%) | 9 (60) | |
Use of ambulatory assistance deviced, n (%) | 3 (20) | |
|
||
|
High School | 4 (26.7) |
|
Some college | 6 (40) |
|
Associate's degree | 3 (20) |
|
Bachelor’s degree | 1 (6.7) |
|
Master's degree | 1 (6.7) |
Right-handed, n (%) | 15 (100) |
aNIHSS: National Institutes of Health Stroke Scale.
bPT: physical therapy.
cOT: occupational therapy.
dUsed a cane at Day 30 visit.
Mobility performance tests at Discharge and Day 30. Scores adjusted for age, sex, race/ethnicity, and education. Average T-Score is 50; higher scores indicate better performance. 2-Minute Walk Test is a measure of endurance; 4-Meter Distance Test is a measure of gait speed.
Measure | Discharge | Day 30 follow-up | Difference | ||||
|
Mean (SD) | Range | Mean (SD) | Range | Mean (SD) | Range | |
2-minute walk |
27.94 |
14.52 to 45.16 | 33.53 (8.09) | 19.39 to 48.75 | 5.60 |
–3.52 to 19.71 | .01 |
Balance T-Score | 40.75 (13.67) | 24.47 to 64.26 | 40.55 (12.98) | 22.96 to 63.68 | 0.64 |
–19.6 to 17.4 | .83 |
4-meter distance |
43.2 |
24.95 to 61.94 | 50.15 (15.23) | 27.68 to 72.67 | 6.95 |
–16.34 to 28.34 | .04 |
2-minute distance (m) | 105.43 (40.9) | 31.9 to 164.59 | 130.92 (33.08) | 79.25 to 180.98 | 25.49 (29.31) | –20.42 to 83.82 | .004 |
4-meter speed (m/sec) | 0.94 |
0.2 to 1.43 | 1.16 (0.36) | 0.67 to 1.79 | 0.22 |
–0.24 to 0.75 | .01 |
a
On average, participants performed worse than the general population after adjustment for education, sex, and age on all 3 tests, especially on the 2-Minute Walk test. Participant scores for both mobility tests were significantly improved at the Day 30 follow-up; balance test scores were similar at both time points.
Using the adherence definition of ≥100 steps, participants were adherent 83.6% of days (approximately n=75) during the 90 day study period. An intercept-only logistic Generalized Estimating Equation using an AR(1) correlation, which adjusts for the correlation of adjacent daily observations, estimated the probability of adherence as 0.83 (95% CI 0.73-0.90). Three participants with low-adherence reported difficulties with the device: the wristband broke, step activity monitor was lost when traveling, the charger was lost. We replaced 2 of these devices. Removing these days from the calculations, probability of adherence was 0.86 (95% CI 0.78-0.92).
While individual observation step counts recorded by the Fitbit displayed variability in comparison to manual tallies, the overall mean difference was only –4.8 steps (–1.8%). Mean absolute difference was higher, at approximately 21.7 steps (10.9%). A Bland-Altman plot showing percent difference between Fitbit and manual counts versus manual count appears in
Bland Altman plot of accuracy of step activity monitor step counts compared to manual counts. Percent difference between Fitbit and manual step counts (y-axis) versus the manual step counts (x-axis). Each participant had 2 trials. The shaded region represents the “acceptable” range of error (10% in either direction), while dotted lines depict the mean percent difference and their 95% confidence intervals. Participants 2, 6, and 8 used a cane for ambulation, but no ambulatory assistance devices were used during gait testing.
Correlations between step counts and mobility performance scores at both time points were very poor. Values for 95% confidence intervals all straddled zero. Correlations with endurance (2-Minute Walk Endurance Test v2.0), balance (Standing Balance Test v2.0), and gait (4-Meter Walk Gait Speed Test v2.0) at discharge were 0.44 (95% CI –0.14 to 0.75), –0.12 (95% CI –0.67 to 0.64), and 0.17 (95% CI –0.46 to 0.66), respectively. Correlations at Day 30 were 0.22 (95% CI –0.45 to 0.71), –0.30 (95% CI –0.83 to 0.41), and 0.21 (95% CI –0.48 to 0.76) respectively.
Intraclass correlation showed moderate within-subject correlation (intraclass correlation coefficient=0.47) in step counts.
Variability in weekly step counts.
Week | Within-subject variability, intraclass correlation coefficient (95% CI)a | Between-subject variability, standard deviation (95% CI)b |
1 | 0.73 (0.49-0.84) | 3338.02 (1997.56-5218.09) |
2 | 0.48 (0.18-0.66) | 2367.04 (1539.35-3639.75) |
3 | 0.39 (0.10-0.60) | 2089.70 (1294.69-3372.90) |
4 | 0.68 (0.40-0.81) | 2967.29 (2011.54-4377.15) |
5 | 0.40 (0.13-0.61) | 2016.81 (1281.03-3175.19) |
6 | 0.54 (0.26-0.70) | 2504.30 (1665.16-3766.31) |
7 | 0.34 (0.09-0.53) | 1859.55 (1143.88-3022.99) |
8 | 0.53 (0.21-0.71) | 2222.68 (1435.35-3441.88) |
9 | 0.41 (0.10-0.61) | 2040.17 (1242.61-3349.64) |
10 | 0.48 (0.18-0.66) | 2252.09 (1445.39-3509.02) |
11 | 0.45 (0.14-0.65) | 2196.83 (1354.26-3563.62) |
12 | 0.36 (0.07-0.57) | 1978.23 (1185.71-3300.49) |
13c | 0.39 (0.04-0.62) | 1639.84 (972.28-2765.745 |
a95% confidence interval estimated using semiparametric bootstrap with 10,000 resamples.
b95% confidence interval calculated using the Wald-type test.
c6-day period from day 85 to 90. Over time, between-subject variability decreased while within-subject variability increased, as shown by lower intra-class correlations.
Of the 15 participants, 10 (67%) completed the participant experience survey. All felt comfortable wearing the Fitbit device, thought it was easy to remember to wear, and planned to continue to use it (
Participant experience survey responses (n=10).
The suitability of the definition of daily adherence used in our analysis was evaluated using self-reported daily diaries during the first 7 days of monitoring
An additional sensitivity analysis was performed to evaluate if the use of different thresholds affected the correlations between mobility performance scores and daily step counts. Altering the thresholds did not improve correlations consistently between the different mobility performance measure scores and average 7-day step count (See
Our study demonstrated that wearing a consumer-grade step activity monitor after stroke is feasible in patients with recent mild acute ischemic stroke. Participants were adherent 83.6% of days over the 90-day study period, which included several instances of device malfunction or misplacement. We defined adherence a priori as ≥100 steps per day, although varying the step-counts thresholds had little effect either on the percentage of days participants were labeled adherent or with the correlation between step counts and participant performance on mobility tests. Studies of adherence with step activity monitors are sparse, and the differences in definition of adherence, populations studied, and duration of monitoring make comparisons across studies difficult. A recent clinical trial of step activity monitor use in patients with rheumatoid arthritis [
The accuracy of step activity monitors is a critical consideration when choosing a device, especially in patients with neurological impairment. This study suggests that the Fitbit Charge HR provides reasonably accurate step counts in patients with mild stroke who have similar gait speeds to those in our study. While observed step counts of individuals displayed high variability in accuracy, overall mean error was only –4.8 steps (–2.4%). The mean gait speed of participants in our study was 0.94 m/sec and the accuracy of the Fitbit Charge HR in patients with slower gait speeds than observed in our study is unclear [
Approximately 58% (15/26) of eligible patients completed the study. Almost 20% of patients approached to participate (10/55) were ineligible because they did not have an internet-facing electronic device that could link to the Fitbit, an important factor to consider when planning future studies of step counts in stroke. This rate is likely to decline with time but could be a significant factor limiting recruitment over the next few years. Acceptability of wearing the Fitbit was quite high. The majority of participants who responded to the survey indicated they would continue to use the devices to monitor their physical activity after the study ended. The respondents also indicated they would not mind having this information be available to providers.
There are several advantages to using consumer-grade step activity monitors to monitor ambulation in patients with stroke and other conditions. Designed for the consumer, they may be more intuitive for patients to use, provide user-friendly displays, and often collect other data such as sleep time and heart rate. These factors may have contributed to the high adherence rates and acceptability of the device in our study. Healthcare wearables, such as step activity monitors, have been predicted to grow from 1 million purchases in 2015 to 97.6 million by 2021 [
Although this study did not directly assess the clinical utility of step activity monitors for patients with mild stroke, the results provide some useful insights for the design of future studies that address this question. First, most of the participants in our study were sedentary, walking less than 5000 steps per day. Our findings are consistent with others who found that many stroke survivors have low levels of physical activity [
A notable and somewhat unexpected finding in our study was the dismal correlation between step counts and performance measures for gait speed, endurance, and balance. The endurance and gait speed tests are standard performance tests in stroke [
An important strength of our study included the 90-day monitoring period. A 7-day protocol for step activity monitors is commonly used in research protocols [
This study also has several limitations. Patient participants had overall mild degree of disability and findings from our study may not be generalizable to stroke survivors with more severe deficits. Our initial target population consisted of patients discharged home after stroke admission, since ambulation in the acute rehabilitation setting can be constricted [
There are also drawbacks with using consumer-grade devices in general. The algorithms used to measure steps and other metrics are typically proprietary and may not be available to investigators [
The use of patient-generated health data in clinical care and research is part of an evolving paradigm shift in health care. Wearable technologies such as step activity monitors offer an excellent mechanism for gathering data from patients effectively, continuously, and in real time. This evaluation represents one of the first applications of patient-generated health data from consumer wearable devices in patients with stroke. The results of this study suggest that the use of the consumer-grade Fitbit Charge HR in patients with recent mild stroke is feasible with reasonable adherence, accuracy, and high level of acceptability. There was poor correlation between step counts and gait speed, balance, and endurance as assessed using the mobility performance tests in the NIH Toolbox. Further research is needed to evaluate the association between step counts and other outcomes of relevance to patients, including patient-reported outcomes and measures of physical function. This feasibility study will serve as important groundwork for further studies of the use of step activity monitors to optimize the care and outcomes of patients with stroke.
Participant activity diary.
Patient experience survey.
Spearman rank correlations between performance evaluations and different daily step count thresholds.
National Institutes of Health
This study was funded in part by the Cleveland Clinic Research Program Committee. IK led the study concept and design, interpreted the data, and drafted the manuscript. AS led the data compilation and collection, interpreted the data, and critically revised the manuscript for intellectual content. TK performed the data analysis, interpreted the data, and critically revised the manuscript for intellectual content.
None declared.