Original Research. Validity of Different Activity Monitors to Count Steps in an Inpatient Rehabilitation Setting

Transcription

1 Validity of Different Activity Monitors to Count Steps in an Inpatient Rehabilitation Setting Daniel Treacy, Leanne Hassett, Karl Schurr, Sakina Chagpar, Serene S. Paul, Catherine Sherrington Original Research Background. Commonly used activity monitors have been shown to be accurate in counting steps in active people; however, further validation is needed in slower walking populations. Objectives. To determine the validity of activity monitors for measuring step counts in rehabilitation inpatients compared with visually observed step counts. To explore the influence of gait parameters, activity monitor position, and use of walkers on activity monitor accuracy. Methods. One hundred and sixty-six inpatients admitted to a rehabilitation unit with an average walking speed of 0.4 m/s (SD 0.2) wore 16 activity monitors (7 different devices in different positions) simultaneously during 6-minute and 6-m walks. The number of steps taken during the tests was also counted by a physical therapist. Gait parameters were assessed using the GAITRite system. To analyze the influence of different gait parameters, the percentage accuracy for each monitor was graphed against various gait parameters for each activity monitor. Results. The StepWatch, Fitbit One worn on the ankle and the ActivPAL showed excellent agreement with observed step count (ICC 2,1 0.98; 0.92; 0.78 respectively). Other devices (Fitbit Charge, Fitbit One worn on hip, G-Sensor, Garmin Vivofit, Actigraph) showed poor agreement with the observed step count (ICC 2, ). Percentage agreement with observed step count was highest for the StepWatch (mean 98%). The StepWatch and the Fitbit One worn on the ankle maintained accuracy in individuals who walked more slowly and with shorter strides but other devices were less accurate in these individuals. Limitations. There were small numbers of participants for some gait parameters. Conclusions. The StepWatch showed the highest accuracy and closest agreement with observed step count. This device can be confidently used by researchers for accurate measurement of step counts in inpatient rehabilitation in individuals who walk slowly. If immediate feedback is desired, the Fitbit One when worn on the ankle would be the best choice for this population. D. Treacy, MHM, The George Institute for Global Health, Sydney Medical School, The University of Sydney, PO Box M201, Missenden Road, New South Wales 2050, Australia. Address all correspondence to Mr Treacy at: daniel. treacy@health.nsw.gov.au. L. Hassett, PhD, The George Institute for Global health, Sydney Medical School, The University of Sydney. K. Schurr, MAppSc, Physiotherapy Department, Bankstown-Lidcombe Hospital, Bankstown New South Wales, Australia. S. Chagpar, MSc(PT), The George Institute for Global Health, Sydney Medical School, The University of Sydney. S.S. Paul, PhD, The George Institute for Global Health, Sydney Medical School, The University of Sydney. C. Sherrington, PhD, MPH, The George Institute for Global Health, Sydney Medical School, The University of Sydney. [Treacy D, Hassett L, Schurr K, et al. Validity of different activity monitors to count steps in an inpatient rehabilitation setting. Phys Ther. 2017;97: ] 2017 American Physical Therapy Association Published Ahead of Print: March 2, 2017 Accepted: February 1, 2017 Submitted: Xxxx XX, XXXX Post a comment for this article at: May 2017 Volume 97 Number 5 Physical Therapy 581

2 Decreased mobility is one of the most common reasons for admission to a rehabilitation unit. A significant dosage of repetitive task-specific practice is provided to maximize walking abilities among patients in a rehabilitation unit. 1-4 Knowing the amount of exercise that each patient completes in the rehabilitation setting assists therapists intervention decisions and review of patients goal achievement. Accurate feedback about exercise dosage and subsequent improvement are likely to be motivating factors for patients. 5 Therefore, accurate recording of the dosage of exercise that participants complete has significant clinical implications. 6 While time spent in therapy has been the most common measure of exercise dosage, dosage of practice performed in a fixed time period has been shown to be highly variable between patients. 7 Counting each repetition of exercise a patient completes is a more reliable record of exercise dosage, as it does not include rest time. Several published studies have used repetitions to accurately measure the actual dosage each rehabilitation patient completes. 3,8-12 Counting the number of steps taken during walking practice can be considered an objective measure of the amount of mobility training a patient has completed. Encouraging individuals in the community to reach a certain number of steps each day is a common strategy used in the community to promote improved health. Visually counting the number of steps patients take during their therapy would be labor intensive and unlikely to count steps taken outside of therapy time. Fortunately, there are many activity monitors currently available that can be used to count steps. Different activity monitors have advantages and disadvantages, including cost, comfort, ease of use, and immediacy of feedback provided. Some monitors are likely to be more suited for research and others for clinical purposes such as motivation. A number of commonly used activity monitors have been found to accurately count steps in active populations of younger and older people Several studies have identified the need for further validation of activity monitors in slower walking populations. No studies have compared the accuracy of multiple activity monitors in a very slow walking population, as is often seen in an inpatient rehabilitation setting. There is currently no information to guide which activity monitors accurately measure step counts in people who walk slowly. Nor has the optimal placement of the monitors been determined or whether monitor accuracy is influenced by specific gait parameters or the use of walking aids. Our research questions were: 1. What is the validity of activity monitors for measuring step counts in rehabilitation inpatients compared with visually observed step counts? 2. Do specific gait parameters influence the accuracy of these activity monitors? 3. Does the position of the monitor influence its accuracy? 4. Does the use of a walker influence activity monitor accuracy? Methods Participants Patients were recruited from the General Rehabilitation and Stroke Rehabilitation Units at Bankstown-Lidcombe Hospital, Sydney, Australia. All patients on these wards were screened weekly for study eligibility, and those meeting the eligibility criteria were invited to participate. Participants were eligible if they (1) had a gait speed <1.2 m/s or an observable gait abnormality thought by their physical therapist as likely to impact the accuracy of the activity monitor; (2) were able to walk at least 10 meters (m) with or without a gait aid or with or without assistance of one person; and (3) were medically stable to participate in a 6-Minute Walk Test (6MWT). All participants provided written informed consent prior to data collection. For those participants with a Mini-Mental State Examination score <18, consent was obtained from a person responsible. For participants whose primary language was other than English, an interpreter was used when explaining the process and completing the consent form. Procedures Each participant wore 16 activity monitors (7 different models in different positions) simultaneously while completing 2 different walking tests as described below. The 7 different models used were the Garmin Vivofit (Garmin Ltd, Schaffhausen, Switzerland) and Fitbit Charge (Fitbit Inc, San Francisco, California), each worn on the wrist like a watch; the Fitbit One was worn on each hip as well as on each ankle; the G-Sensor 2026 Accelerometer Pedometer (Pedometers Australia, Perth, Australia) was worn on each hip; the ActiGraph GT3X+ accelerometer ( ActiGraph Corp, Pensacola, Florida) was worn on each hip; the ActivPAL (Pal Technologies, Glasgow, Scotland) was worn on both thighs; and the StepWatch Activity Monitor (Modus Health, Washington DC) was worn on each ankle. Where possible, participants age, sex, and approximate height were programmed into the device software (ie, all devices except the ActivPAL and G-Sensor Pedometers). During the study period 4 Garmin Vivofits, 4 Fitbit Charges, 8 Fitbit Ones, 4 ActiGraph GT3X+ accelerometers, 4 G-Sensor Pedometers, 4 ActivPALS, and 4 StepWatch Activity Monitors were used. These activity monitors were rotated randomly across body location to ensure varied use of each device. These devices were chosen because they represent activity monitors currently used in research and clinical practice or are popular in the general population. Each device has various advantages and disadvantages, such as cost, ease of attachment, and feedback timing. All participants completed a 6MWT on the same track wearing the monitors described above. A rectangular walking track (120 m) was used to minimize the number of turns required to complete the distance. Standardized instructions were given to all participants. Participants were instructed to stand still for 10 seconds prior to and after the 6MWT. At the end of the test, the count on all the activity monitors was recorded. The number of steps recorded by the Fitbit 582 Physical Therapy Volume 97 Number 5 May 2017

3 and Garmin trackers was calculated as the difference between the step count displayed by the tracker at the start and at the end of the 6MWT. The number of steps recorded by the ActiGraph was extracted in 1-second epochs with ActiLife 6 software. The number of steps recorded by the ActivPAL was extracted with ActivPAL v software. The number of steps recorded by the Step- Watch was extracted in 3- second epochs with Modus Trex software. An assessor (D.T., a senior physical therapist with >10 years experience in rehabilitation) concurrently counted the steps each individual took using a hand-held counter. This is referred to as the observed step count. To determine the reliability of the observed step count, a second assessor simultaneously counted the number of steps taken by 17 (10%) participants using a handheld counter. The second assessors included physiotherapy students, physiotherapy assistants, junior and senior physiotherapy staff, and physiotherapy research staff. Interrater reliability between observed step counts by the 2 assessors was excellent, with an intraclass correlation coefficient (ICC 2,1 ) of (95% CI: ). The percentage agreement between the 2 observed step counts was 99.0% (SD 0.8%). Participants then performed 2 walking trials along a 14-foot GAITRite Gold walkway (CIR Systems Inc, New York). GAITRite is a portable gait analysis system embedded with pressure sensors that detect footfalls as the participant walks the length of the mat. The software enables the measurement of multiple gait parameters, including walking speed, cadence, step length, and duration of swing and stance phases. GAITRite has been shown to be highly reliable in measuring most gait parameters. 24 The GaitRite Gold walkway contained half inch square sensors that were laid out in 2 2 feet pads. The walkway had 7 pads, for a total of sensors covering an active area 2 feet wide and 14 feet long. The walkway had a sampling frequency of msec. Participants started walking 1 m before and finished 1 m beyond the gait mat. Participants sat down on a chair after each trial to rest. Participants were asked to walk at their normal speeds for each test. The assessor counted the number of steps as per the process for the 6MWT test. For both walking tests participants walked unaided or used their current gait aid. The assessor walked slightly behind and to the side of the participant so as not to influence the participant s walking speed. Data Analysis Data were analyzed using Stata v13 (StataCorp LP, College Station, Texas). ICC 2,1 were used to examine criterion validity between step counts taken from each of the devices compared with the observed step count during the 6MWT. An ICC 0.75 was considered excellent, good, fair, and <0.40 poor. 25 The most accurate step count between the left and right sides (ie, the step count closest to the observed step count) was used for the primary analysis of each device at each body position. The percentage agreement for each device compared with the observed step count was calculated as: (activity monitor measured step count / observed step count) 100. The accuracy within a close agreement (within 6 steps of the observed step count) and within 10% and 20% agreement was assessed for each activity monitor. Average absolute error was calculated as the difference between the activity monitor count and the observed step count for each activity monitor. The percentage of occasions that each device did not record any steps taken was recorded. The influence of different gait parameters on the percentage agreement between measured and observed step count was visually assessed by graphing the percentage agreement for each activity monitor against the following gait parameters assessed by the GAITRite: walking speed, cadence, stride length, swing speed, double support stance phase percentage, and gait variability. Subgroup analyses were performed on participants with a unilateral physical impairment and comparing those using and not using a walker. ICC 2,1 and percentage agreements were used to examine agreements between step counts taken from each device compared with the observed step count on both the affected side and the unaffected side. The effect of using a walker on device accuracy was assessed by graphing the percentage agreement of the observed step count for each device against walking speed for individuals who did and did not use a walker. Results During the recruitment period, 402 patients were admitted to the 2 rehabilitation wards at Bankstown-Lidcombe Hospital. One hundred and sixty-six of these individuals participated. Of the 236 patients who did not participate, 39 were ineligible because they were unable to walk 10 m despite the assistance of one person and/or a walker, and 187 participants declined to participate in the study. The majority of participants were male with slower than normal walking speeds admitted for rehabilitation with a mix of musculoskeletal and neurological health conditions (Table 1). Data were collected from all 16 devices for all participants except for one participant who had data collected from only 14 devices due to a previous left upper limb amputation (no data were collected from the Garmin Vivofit and Fitbit Charge, both located on the left wrist). Concurrent Validity and Accuracy Compared with Observed Step Count Table 2 summarizes the main results. More detailed results are available in the online supplement (online supplement 1: Full results table). The StepWatch (ICC 2, ), Fitbit One worn on the ankle (ICC 2, ), and the ActivPAL (ICC 2, ) showed excellent agreement with the observed step count. All other devices showed poor agreement with the observed step count (ICC 2, ). The StepWatch had the highest percentage agreement compared with the observed step count (98%, SD 12%), followed by the Fitbit One worn on the May 2017 Volume 97 Number 5 Physical Therapy 583

4 Table 1. Participant Characteristics a Characteristic (n = 166) Mean (SD; range) or n (%) Age, y 80 (11; 26 98) Sex, male 91 (55%) Affected side Bilateral or neither 66 (40%) Left 50 (30%) Right 50 (30%) 6MWT distance, m 123 (112; ) 6MWT observed step count 413 (152; ) Walking speed, m/s 0.42 (0.22; ) Double support phase 44 (41; 20 87) Walking aid Nil 55 (33%) Cane, s 32 (19%) Walker 79 (48%) Diagnosis Fractured hip, pelvis, or other lower limb orthopedic surgery 52 (31%) Stroke, transient ischemic attack 35 (21%) Post fall with no lower limb fracture 20 (12%) Decreased mobility post medical or non-orthopedic surgical event 35 (21%) Neurological event (nonstroke) 6 (4%) Other 16 (10%) a 6MWT=6-Minute Walk Test. Walkers used included: no wheels; with 2 wheels on the front and skis on the back; with 4 wheels and handbrakes. ankle (84%, SD 27%) and the ActivPAL (76%, SD 26%). The StepWatch achieved close agreement (within 6 steps of the observed step count) for 70% of participants. The Fitbit One worn on the ankle achieved close agreement for 27% of participants (Table 2). There were no occasions where the Step- Watch did not record any steps on either side. The Garmin Vivofit did not record any of the steps taken during the 6MWT on 49% of participants. There were a few occasions where the ActivPAL (0.6%) and the Actigraph (2.4%) did not record any steps taken on either side (Table 2). The Actigraph showed the highest absolute error during the 6MWT (299 steps per participant), followed by the Garmin Vivofit (269), G-Sensor (233), Fitbit One worn on the hip (201), and Fitbit Charge (182). Influence of Gait Parameters on Accuracy of Activity Monitors For participants who walked with a walking speed between 0.8 to 1.2 m/s, all devices except the Actigraph showed 90% agreement with the observed step count (Figure 1). Accuracy of all devices decreased as walking speed decreased. The StepWatch maintained close to 100% accuracy with the observed step count at all speeds except below 0.2 Table 2. Counts of Agreement, Accuracy, and Error for Each Device for the 6MWT (n = 166) Compared with Observed Step Count, Reported as Percentages, in Order of Best to Worst Based on ICC( 2,1 ) a Activity Monitor ICC( 2,1 ) (95% CI) Percentage Agreement (SD) b StepWatch (worn on ankle) Close 10% Agreement c Agreement 20% Agreement Recorded Zero Over 6MWT Average Absolute Error (SD) d ( ) 98% (12%) 70% 92% 95% 0% 11 (27) Fitbit One (worn on ankle) ( ) 84% (13%) 27% 68% 80% 5% 41 (55) ActivPAL (worn on thigh) ( ) 76% (25%) 8% 42% 58% 1% 81 (80) Fitbit Charge (worn on wrist) ( ) 52% (43%) 4% 28% 39% 28% 182 (165) Fitbit One (worn on hip) ( ) 44% (39%) 4% 23% 30% 23% 201 (153) G-Sensor (worn on hip) ( ) 38% (39%) 4% 16% 25% 14% 233 (167) Garmin Vivofit (worn on wrist) ( ) 29% (39%) 4% 14% 21% 49% 269 (176) Actigraph (worn on hip) ( ) 26% (25%) 1% 2% 7% 2% 299 (155) a ICC=intraclass correlation coefficient b The percentage agreement for each device compared with the observed step count was calculated as: (activity monitor measured step count / observed step count) 100. c Close agreement: within 6 steps of the observed step count. d Steps per participant (whole numbers). 584 Physical Therapy Volume 97 Number 5 May 2017

5 as leg swing speed decreased. Cadence, double support stance phase percentage, and gait variability did not appear to have a strong influence on the accuracy of these devices (online supplement 3: Alternate gait parameters versus percentage accuracy of the observed step count). Figure 1. Gait velocity (m/s) from GAITRite versus percentage accuracy of the observed step count by device. Figure 2. Stride length (m) from GAITRite versus percentage accuracy of the observed step count. m/s, when its accuracy decreased to 89%. No other device maintained more than 50% accuracy with the observed step count at walking speeds below 0.2 m/s. At speeds less than 0.2 m/s, the Garmin Vivofit (70%), Fitbit One worn on the hip (60%), and Fitbit Charge (53%) all recorded zero steps on both sides for more than half the participants. For participants who walked with a stride length longer than 1.0 m, all devices except the Actigraph showed 90% agreement with the observed step count (Figure 2). The accuracy of all devices decreased with decreasing stride length. The StepWatch maintained close to 100% agreement with the observed step count at all stride lengths except those <0.4 m, when its accuracy decreased to 89%. The speed of the swinging leg appeared to influence the accuracy of the StepWatch, ActivPAL, and the Fitbit One worn on the ankle. The accuracy of these devices (online supplement 2: Leg swing velocity vs percentage accuracy of the observed step count) decreased Location of Activity Monitor and Use of Walkers on Accuracy of Activity Monitor Table 3 compares the accuracy of each device when worn on the unaffected side compared with the affected side for participants with a unilateral physical impairment (n = 100). For the Step- Watch, ActivPAL, and Fitbit One worn on the ankle, wearing the device on the unaffected side on average resulted in greater accuracy than wearing the device on the affected side. However, there were occasions when the affected side had greater accuracy than the unaffected side. For all other devices, there was no difference in accuracy when wearing the device on the unaffected or affected side. The accuracy of the leg-worn devices was similar for those who used (n = 87) and did not use (n = 79) a walker (online supplement 4: Alternate graphs walker versus no walker), but the accuracy of the wrist-worn devices (Garmin Vivofit and Fitbit Charge) was worse at slower speeds among walker users (Figure 3). Between speeds of 0.2 m/s and 0.8 m/s, the monitors were less accurate for those who used a walker. At speeds below 0.2 m/s, the accuracy of the devices was very poor regardless of walker use. Discussion At walking speeds between 0.8 and 1.2 m/s, all devices except the Actigraph showed high accuracy, and the choice of device at these speeds should be determined by the purpose of using the device, patient preference, and features of the device such as those described in the etable (available at academic.oup.com/ptj) for instance, cost, charging requirements, parameters reported. At these speeds, the use of a walker may not affect the accuracy of these devices. However, at May 2017 Volume 97 Number 5 Physical Therapy 585

6 Table 3. Comparison of Accuracy for Unaffected Versus Affected Sides for Participants with an Observable Affected Side (n = 100), in Order of Best to Worst Based on ICC( 2,1 ) a Unaffected Side Affected Side Activity Monitor ICC( 2,1 ) (95% CI) Percentage agreement (SD) b slower speeds and shorter step lengths the accuracy of these devices varies considerably. Additionally the use of a walker and/or the side on which the device is attached can affect accuracy. The StepWatch showed the highest accuracy and closest agreement with the observed step count of all the devices across all gait parameters for individuals assessed in this study. Count of Times Closest to Observed Step Count Than Affected Side c The 3 devices with the strongest agreement (the StepWatch, Fitbit One worn on the ankle, and ActivPAL) were the ones worn lower down the leg. These results are similar to findings by Klassen et al, 26 who found that placement of the Fitbit One at the ankle was more accurate than at the waist in slow walking community-dwelling survivors of stroke. All devices use an accelerometer ICC( 2,1 ) (95% CI) Percentage Agreement (SD) * Count of Times Closest to Observed Step Count Than Unaffected Side c StepWatch (worn on ankle) ( ) 98% (12%) ( ) 87% (27%) 29 Fitbit One (worn on ankle) ( ) 77% (30%) ( ) 71% (35%) 35 ActivPAL (worn on thigh) ( ) 68% (28%) ( ) 56% (33%) 29 Fitbit Charge (worn on wrist) ( ) 35% (41%) ( ) 36% (42%) 35 Fitbit One (worn on hip) ( ) 34% (37%) ( ) 35% (37%) 34 G-Sensor (worn on hip) ( ) 24% (34%) ( ) 23% (31%) 41 Garmin Vivofit (worn on wrist) ( ) 18% (33%) ( ) 17% (29%) 23 Actigraph (worn on hip) ( ) 17% (19%) ( ) 18% (18%) 47 a ICC=intraclass correlation coefficient b The percentage agreement for each device compared with the observed step count was calculated: (activity monitor measured step count / observed step count) 100. c Count of times closest to observed step count may not add up to 100 as on some occasions the results were the same. Figure 3. Gait velocity (m/s) from GAITRite versus percentage accuracy of the observed step count for patients either using a walker or not using a walker when using a wrist-worn activity monitor. to detect motion and attempt to distinguish steps from other activity. Each device has a threshold for this acceleration signal to determine whether the motion detected represents a step. The slower swing-through of the leg in slower walking individuals may mean that acceleration at the hip may not be sufficient to detect a step, whereas acceleration at the thigh and ankle is greater, thus the device is more likely to detect steps. For the 3 most accurate devices (the StepWatch, the Fitbit One worn on the ankle, and the ActivPAL), wearing them on the unaffected side appeared more accurate; however, there were occasions when this was not the case. For individuals with bilaterally affected sides, it is not always possible to predict which side will be the most accurate. Therefore, it is recommended that the chosen activity monitor be trialed initially on both sides and compared with a manual count to identify which side is most accurate for that individual. A key reason for using activity monitors in clinical settings is to provide feedback to patients to aid motivation to improve mobility. The high percentage of recorded zeroes for the Garmin Vivofit, Fitbit Charge, and Fitbit One 586 Physical Therapy Volume 97 Number 5 May 2017

7 worn on the hip is likely to be a significant problem if these devices are being used for motivation. Fortunately the Fitbit One worn on the ankle showed the highest accuracy of all the devices that provided immediate feedback. However, the ability of this population to use this feedback to improve mobility requires further investigation. This study did not investigate other factors such as comfort, usability, or participant preference. Future qualitative research may provide insight into these issues. The main limitation of the study is the small number of participants for some gait parameters. There were only 2 participants who walked with a stride length >1.2 m and only 1 participant who walked with a walker at a speed >0.8 m/s, making it difficult to determine the accuracy of these activity monitors for those parameters. This study is the first to compare multiple devices on participants walking with slow to very slow walking speeds in an inpatient rehabilitation setting. It is the only study to date to examine the influence of a range of gait parameters on the accuracy of multiple devices in a very slow walking population. The StepWatch showed the highest accuracy and closest agreement with the observed step count of the 7 devices assessed in this study and can be used by researchers for accurate measurement of the number of steps taken in this population. However, the StepWatch does not provide immediate feedback. The next most accurate device was the Fitbit One when worn on the ankle, and this device can provide immediate feedback about step counts. This may be of benefit for patients and therapists in a rehabilitation setting who require immediate objective measures and feedback of the amount of activity. Author Contributions Concept/idea/research design: D. Treacy, L. Hassett, K. Schurr, C. Sherrington Writing: D. Treacy, L. Hassett, K. Schurr, S.S. Paul, C. Sherrington Data collection: D. Treacy, S. Chagpar Data analysis: D. Treacy, L. Hassett, C. Sherrington Consultation (including review of manuscript before submitting): K. Schurr, S. Chagpar, S.S. Paul Ethics Approval The South West Sydney Local Health District Human Research Ethics Committee approved this study. Funding Support Dr Sherrington s salary is supported by a fellowship from the Australian National Health and Medical Research Council. Bankstown-Lidcombe Hospital and The George Institute for Global Health provided facilities/equipment. Disclosures and Presentations The authors declare that they have no competing interests. Activity monitors were purchased by The George Institute for Global Health, and no financial support was received from any commercial company. Part of this manuscript was presented at the World Congress for Active Ageing in Melbourne, Victoria, Australia, on July 1, DOI: /ptj/pzx010 References 1 Bütefisch C, Hummelsheim H, Denzler P, Mauritz K-H. Repetitive training of isolated movements improves the outcome of motor rehabilitation of the centrally paretic hand. J Neurol Sci. 1995;130: Classen J, Liepert J, Wise S, Hallett M, Cohen L. Rapid plasticity of human cortical movement representation induced by practice. J Neurophysiolog. 1998;79: Lang CE, Macdonald JR, Reisman DS, et al. Observation of amounts of movement practice provided during stroke rehabilitation. Arch Phys Med Rehabil. Oct 2009;90(10): Carr JH, Shepherd RB. Neurological Rehabilitation Optimizing Motor Performance. London: Churchill Livingstone; Colombo R, Pisano F, Mazzone A, et al. Design strategies to improve patient motivation during robot-aided rehabilitation. J Neuroeng Rehabil. 2007;4:3. 6 Falck RS, McDonald SM, Beets MW, Brazendale K, Liu-Ambrose T. Measurement of physical activity in older adult interventions: a systematic review. Br J Sports Med. Aug ;50: Scrivener K, Sherrington C, Schurr K, Treacy D. Many participants in inpatient rehabilitation can quantify their exercise dosage accurately: an observational stud. J Physiother. 2011;57(2): Nugent JA, Schurr KA, Adams RD. A dose-response relationship between amount of weight-bearing exercise and walking outcome following cerebrovascular accident. Arch Phys Med Rehabil. 1994: Scrivener K, Sherrington C, Schurr K. Exercise dose and mobility outcome in a comprehensive stroke unit: description and prediction from a prospective cohort study. J Rehabil Med. Oct 2012;44(10): Scrivener K, Sherrington C, Schurr K. Amount of exercise in the first week after stroke predicts walking speed and unassisted walking. Neurorehabil Neural Repair. Oct 2012;26(8): Lang C, MacDonald J, Gnip C. Counting repetitions: an observational study of outpatient therapy for people with hemiparesis post-stroke. J Neurol Phys Ther. 2007;31(1): Treacy D, Schurr K, Lloyd B, Sherrington C. Additional standing balance circuit classes during inpatient rehabilitation improved balance outcomes: an assessor-blinded randomised controlled trial. Age Ageing. Jul 2015;44(4): Diaz KM, Krupka DJ, Chang MJ, et al. Fitbit(R): an accurate and reliable device for wireless physical activity tracking. Int J Cardiol. Apr ;185: Noah JA, Spierer DK, Gu J, Bronner S. Comparison of steps and energy expenditure assessment in adults of Fitbit Tracker and Ultra to the Actical and indirect calorimetry. J Med Eng Technol. Oct 2013;37(7): Ryan CG, Grant PM, Tigbe WW, Granat MH. The validity and reliability of a novel activity monitor as a measure of walking. Br J Sports Med. Sep 2006;40(9): Case MA, Burwick A, Volpp KG, Patel MS. Accuracy of smartphone applications and wearable devices for tracking physical activity data. JAMA. 2015;313(6): Paul SS, Tiedemann A, Hassett LM, et al. Validity of the Fitbitactivity tracker for measuring steps in community-dwelling older adults. BMJ Open Sport Exerc Med. 2015;1(1):e Grant PM, Dall PM, Mitchell SL, Granat MH. Activity-monitor accuracy in measuring step number and cadence in community-dwelling older adults. J Aging Phys Act. 2008;16: Storti KL, Pettee KK, Brach JS, Talkowski JB, Richardson CR, Kriska AM. Gait speed and step-count monitor accuracy in community-dwelling older adults. Med Sci Sports Exerc. Jan 2008;40(1): Van Renoortel H, Giavedoni S, Raste Y, et al. Validity of activity monitors in health and chronic disease: a systematic review. Int J Behav Nutr Phys Act. 2012;9(9):84. May 2017 Volume 97 Number 5 Physical Therapy 587

8 21 Carroll SL, Greig CA, Lewis SJ, et al. The use of pedometers in stroke survivors: are they feasible and how well do they detect steps? Arch Phys Med Rehabil. Mar 2012;93(3): Lee IM, Shiroma EJ. Using accelerometers to measure physical activity in large-scale epidemiological studies: issues and challenges. Br J Sports Med. Feb 2014;48(3): Fulk GD, Combs SA, Danks KA, Nirider CD, Raja B, Reisman DS. Accuracy of 2 activity monitors in detecting steps in people with stroke and traumatic brain injury. Phys Ther. Feb 2014;94(2): Menz HB, Latt MD, Tiedemann A, Mun San Kwan M, Lord SR. Reliability of the GAITRite walkway system for the quantification of temporo-spatial parameters of gait in young and older people. Gait Posture. 2004;20(1): Hallgren KA. Computing inter-rater reliability for observational data: an overview and tutorial. Tutor Quant Methods Psychol. 2012;8: Klassen TD, Simpson LA, Lim SB, et al. Stepping up activity poststroke: ankle-positioned accelerometer can accurately record steps during slow walking. Phys Therapy. Aug ;96(3): Physical Therapy Volume 97 Number 5 May 2017