Wearable activity trackers are a multi–billion-dollar industry with expectations of future growth (1,2). In 2015, device shipments increased by 20%, equaling 20.5 million shipments (3). The use of these devices is forecasted to increase further over the next 5 yr, and during this period, a shift from fitness trackers to wearable medical devices is expected (2).
Currently, activity trackers are used by consumers and researchers to quantify daily physical activity (4). For the general public, these devices are used for personal motivation to be active, goal setting, and self-interest in recording aspects of daily life (i.e., the “quantified self” movement) (5). Medical researchers use activity trackers for physical activity surveillance (6–8), to examine associations between physical activity and health outcomes (9,10), to monitor compliance with interventions (11,12), and to track rehabilitation programs (13).
Most activity trackers provide estimates of steps, distance, and caloric expenditure, and some include features such as heart rate, sleep duration, activity recognition, pulse oximetry, and mobile phone connectivity. Among all the metrics available to users, the most recognizable is step counts. Steps are a common metric for physical activity assessment (14) because they are intuitive (15), objective (16), and motivating (11), correlated with cardiometabolic biomarkers (11,17,18), and linked to positive health outcomes (11,18). In addition, walking and running are commonly reported forms of physical activity for Americans (19–21), both of which are easily measured in terms of steps.
Recent studies have examined the step count accuracy of consumer-grade activity trackers during bouts of continuous, moderate-intensity walking (22–26). For example, the mean absolute percent errors (MAPE) for the Lumoback, Jawbone Up, Misfit Shine, Withings Pulse, Fitbit Zip (FZ), Digi-Walker SW-200, Garmin Vivofit, and Fitbit Charge (FC) HR were within 2% of hand-counted steps for continuous treadmill walking at 80 m·min−1 (23,24,26). In addition, Nelson et al. (22), found MAPE values ranging from 2% to 11% for the FZ, Fitbit Flex, Jawbone UP24, and Omron HJ-720ITC when compared with hand-counted steps for walking (at an average speed of 78 m·min−1) (22,23).
Step-counting errors increase during slow walking (<80 m·min−1) and short, intermittent walking bouts (<6 to 8 steps) (23,27–30). At walking speeds of <80 m·min−1, consumer and some research-grade devices tend to underestimate steps (23,25,27,28,30,31). Recently, Chen et al. (31) reported that four wrist-worn consumer-grade trackers (Fitbit Flex, Jawbone UP, and Garmin Vivofit) significantly underestimated steps at four speeds ranging from 54 to 134 m·min−1 when compared with hand-counted steps. The Fitbit Flex and Jawbone UP had the greatest errors at a slow speed (54 m·min−1), but accuracy improved as speed increased. Hickey et al. (28) reported similar findings for research and hip-worn consumer-grade devices. In that study, the ActivPAL, ActiGraph GT3X (without low-frequency extension (LFE)), Yamax Digi-Walker SW-200 (YX), and Omron HJ-720ITC significantly underestimated steps (by 46%–89%) during treadmill walking at 40 m·min−1 and accuracy improved as speed increased to 162 m·min−1. Activities with nonrhythmic and multidirectional stepping (e.g., cleaning, vacuuming, folding laundry, gardening) also increased step count error (22,28). Hickey et al. (28) found that these devices significantly underestimated steps during cleaning (range, 20%–92% of criterion), dusting (range, 59%–26% of criterion), and vacuuming (range, 20%–79% of criterion). Similarly, Nelson et al. (22) found that the Omron HJ-720ITC, Fitbit One, FZ, and Jawbone UP24 significantly underestimated steps by 35% to 74% (compared with hand-counted steps) during household activities.
Brief stepping bouts are a common occurrence in the free-living environment, with more than one third of bouts consisting of eight or fewer steps (32). Because some trackers use step-counting algorithms that discard steps taken in brief intermittent walking bouts, an underestimation of step counts during intermittent activities is possible (29,33). Thus, the purpose of this study was to test the equivalence of step count estimates from six consumer-grade activity trackers (Garmin Vivofit 2 (GV), FC, Withings Pulse Ox (WT), YX, FZ, and Omron HJ-322U (OM)) and four research-grade activity trackers (two ActiGraph WGT3X-BT (AG) and two StepWatch (SW) devices) to hand-counted steps, during 2-min bouts of treadmill locomotion, overground walking, and activities of daily living (ADL). Research-grade activity trackers (i.e., SW and AG) were defined as those that store raw data and require specific software for data extraction. Consumer-grade activity trackers (i.e., GV, FC, WT, YX, FZ, and OM) were defined as those that store and display data in user friendly terms (e.g., steps, calories, and distance) and often sync to smartphone applications via Bluetooth connectivity.
Twenty-one adults (mean ± SD age, 26 ± 9 yr) were recruited from the University of Tennessee and surrounding community. Individuals with heart or lung problems, orthopedic injuries, or other conditions that could be exacerbated by exercise, and those who reported difficulty walking and/or used assistive devices when walking were excluded from the study. The research protocol was approved by the University of Tennessee Institutional Review Board. All participants provided written informed consent and completed a health history questionnaire before participating in the study.
Participants were asked to perform 15 activities that fell into three general categories while wearing the devices:
- Treadmill locomotion
- Walking at 27 m·min−1
- Walking at 54 m·min−1
- Walking at 80 m·min−1
- Walking at 80 m·min−1 while holding onto handrails
- Jogging at 161 m·min−1
- Walking while holding onto backpack straps
- Walking while holding an umbrella in nondominant hand
- Walking with hands in coat pocket
- Walking while pushing a stroller
- Brushing teeth
- Brushing hair
- Eating a snack
- Meal preparation (cooking an egg)
- Folding laundry
- Sweeping a room (using a broom)
For this study, treadmill locomotion was completed at 0% grade. The overground walking conditions were performed in a hallway at self-selected speeds. Participants completed all of the activities during one session in the order listed. ADL tasks were performed in a laboratory setting. Each activity was performed continuously for 2 min, separated by a 1-min rest. The total duration of activities was 30 min and the total time of participation including rest was approximately 1 h. Steps taken during each activity were hand counted by a researcher and were used as the criterion for statistical comparisons. Start and stop times and step counts from devices with digital displays were recorded at the beginning/end of the activity. For devices without displays, the time-stamped data were downloaded to a laptop computer after each trial. Steps were then summed over the duration of the activity, using the start and stop times recorded.
A trained investigator put the devices on each participant to ensure proper placement. Four activity trackers were simultaneously worn on the nondominant wrist. These devices included one research-grade tracker (AG-wrist; ActiGraph, LLC, Pensacola, FL)) and three consumer-grade trackers (FC; Fitbit, San Francisco, CA), GV (Garmin International, Inc., Olathe, KS), and WT (Cambridge, MA)). The first device was placed proximal to the styloid process. The remaining three devices were placed just proximal to the previous device, without making contact with any other devices. The order of devices was randomized for each participant.
Four activity trackers were worn on the hip. These devices included one research-grade tracker (AG-hip) and three consumer-grade trackers (YX; Yamax; Yamasa Corporation, Tokyo, Japan), OM (Omron, Omron Healthcare, Lake Forest, IL), and FZ (Fitbit). The AG-hip and YX were worn on the right hip in line with the anterior axillary line and the midline of the thigh, respectively. The OM and FZ were worn along the same respective landmarks on the left hip. The AG-hip was secured with an elastic band, and the other devices were clipped to the waistband of the participant's pants or shorts.
Two research-grade SW activity monitors (Modus Health, LLC, Washington, DC) were worn on the right ankle directly above the lateral and medial malleoli, and were secured with an elastic strap with a Velcro® closure. The SW-default was programmed with default settings (cadence determined by the participant's height, and sensitivity set at 13) (14) and the other (SW-modified) was programmed with modified settings (cadence was set at 70% of default cadence and the sensitivity was set at 16) (15). Medial and lateral placement of the SW devices was randomized for each participant.
SW and AG data were downloaded after each participant with Modus Health (version 3.4) and ActiLife 6 (version 6.13.1), respectively. Data were downloaded into time-stamped epochs and summed across the 2-min trials. ActiGraph data were processed with two different step-counting algorithms: (a) the default algorithm in ActiLife software and (b) a beta-version of the moving average vector magnitude (MAVM) algorithm. The ActiLife algorithm can be applied to the acceleration data with or without the LFE. The LFE extends the band-pass filter which increases the sensitivity for recording low-intensity movements. The LFE allows more low-frequency accelerations to be included as “counts” (16), and it also dramatically increased the number of steps counted (17). The MAVM algorithm was designed to perform on-board step count estimation using reduced processing power; however, this algorithm was applied to the raw data during postprocessing in the present study. The algorithm features wear site–specific filters for the hip and wrist. Continuous stepping must occur for at least 2 s for the hip wear location and 10 s for the wrist location before steps are added to the aggregated step counts (18).
Equivalence tests were completed for this investigation to assess the agreement between activity tracker output and hand-counted steps. Equivalence testing reverses the null and alternative hypotheses of traditional tests of mean difference (e.g., t-test, ANOVA) to provide direct evidence of statistical equivalence. Thus, whereas rejection of the null hypothesis in mean difference tests provides evidence of group-level differences, rejection of the null for equivalence testing shows evidence of group-level equivalence (19,20). This is important, because rejection of the null hypothesis is the only way to make a valid empirical inference within hypothesis tests. Potential problems of using tests of mean difference to demonstrate agreement have been detailed elsewhere (19,21).
Equivalence testing was carried out by performing two one-sided t-tests with α = 0.05. The equivalence zone was defined as within ±10% of the mean hand-counted steps. Thus, the first t-test evaluated whether the estimate was ≥90% of the mean hand-counted steps, and the second evaluated whether it was ≤110% of hand-counted steps. The larger of the two resulting P values was the P value for the equivalence test, and general significance (i.e., P ≤ 0.05) was shown visually if the estimate's 90% confidence interval (CI) fell entirely within the equivalence zone. Equivalence tests were completed for three categories of activities: ADL, treadmill locomotion, and overground walking. For these tests, units were steps per minute. A separate equivalence test was performed to compare total steps for all activities combined.
Percent of hand-counted steps, MAPE, mean bias, and 95% prediction intervals were calculated for treadmill locomotion and overground walking. Because some ADL items resulted in zero steps taken, the percent of hand-counted steps and MAPE could not be calculated; therefore, only mean bias and 95% prediction intervals are displayed for ADL. Mean bias, percent of hand-counted steps ((device estimate/hand count) × 100%), and MAPE were computed for all activities combined. Significance was set at α = 0.05. All statistics were analyzed using SPSS Version 24 for Windows (SPSS Inc., Chicago, IL).
Twenty-one participants completed this study (Table 1). All participants were right-hand dominant. Mean steps per minute, mean bias, and 95% prediction intervals for treadmill locomotion, overground walking, and ADL are displayed in Table 2.
In regard to treadmill locomotion, two research-grade devices, SW-modified (P = 0.002) and AG-hip (LFE; P = 0.04), were statistically equivalent to hand count (Fig. 1A). The SW-modified captured 104.7% of hand-counted steps (MAPE, 6.4%) and AG-hip (LFE) captured 94.2% of hand-counted steps (MAPE, 6.7%). The nonequivalent devices generally underestimated the actual steps taken in treadmill locomotion (Fig. 2).
For self-paced overground walking, the SW-default and SW-modified, FZ, and YX were all statistically equivalent to hand-counted steps (P < 0.001; Fig. 1B). The statistically equivalent research-grade devices were the SW-default and SW-modified, which captured 98.2% and 102.2% of actual steps, respectively. The statistically equivalent consumer-grade trackers were the FZ and YX, and they captured 99.7% and 99.4% of actual steps, respectively. The nonequivalent consumer-grade trackers captured at least 80% of actual steps, with the exception of WT, which underestimated to a greater extent (Fig. 2).
For ADL, the 90% CI for hand-counted steps (9.4–14.2 steps per minute) was outside the equivalence zone of ±10% mean hand-counted steps (10.6–13.0 steps per minute). Thus, ±10% was not an appropriate range for detecting equivalence in ADL, and the equivalence zone was expanded to enclose the CI for hand-counted steps, resulting in a lower bound of 9.3 steps per minute and an upper bound of 14.3 steps per minute (mean hand count, ±~21%). The subsequent results are shown in Figure 1C. The AG-hip (LFE) was the only statistically equivalent device to hand count (P = 0.02). Most of the other devices fell outside the equivalence zone. Only the GV and WT had means that fell within the equivalence zone, but were not equivalent due to large 90% CI (Table 2).
For all activities combined, the SW-modified and AG-hip (LFE) were statistically equivalent to hand-counted steps (SW: P < 0.001, AG: P = 0.01; Fig. 3). The SW-modified captured 99.5% of hand-counted steps (MAPE, 3.6%) and the AG-hip (LFE) captured 94.4% of hand-counted steps (MAPE, 6.0%; Table 3). The nonequivalent research and consumer-grade devices all underestimated steps; however, the consumer devices (with the exception of the WT) had more consistent results than the research-grade devices (Fig. 3).
The results of this study show that across 15 activities, only two research-grade devices, the SW-modified and AG-hip (LFE), were statistically equivalent to hand-counted steps. These devices had the smallest step-counting errors out of all devices in the study. The device with the greatest error was the AG-wrist. However, the ActiLife algorithm was designed for hip wear and was not optimized for the wrist; therefore, it is not surprising that it performed poorly. Interestingly, across most of the consumer-grade trackers (excluding WT), the step count error of was very consistent, capturing approximately 87.0% to 92.7% of hand-counted steps. The extreme consistency of consumer-grade trackers could be the result of recent guidelines developed by the Consumer Technology Association, accredited by the American National Standards Institute, which outlines standardization guidelines and procedures for wearable fitness trackers (34). In many cases, the step-counting error of consumer-grade devices was small enough that the devices could be useful for facilitating behavior change. However, the most accurate methods (SW and AG-hip LFE) are better suited for researchers who want to capture every step, rather than only the steps taken during continuous walking.
When activities were grouped into three general categories (treadmill locomotion, overground walking, and ADL), some interesting differences between activity trackers were observed. For the treadmill locomotion category (i.e., walking and running), the most accurate research grade devices were the SW-modified and AG-hip (LFE). The least accurate research-grade device was the AG-wrist, and the least accurate consumer-grade device was the WT. For overground walking, several research and consumer-grade devices provided accurate step counts, indicating that these devices performed well during self-paced walking under various conditions (i.e., holding an umbrella, hands in pockets, pushing a stroller, holding backpack straps). Once again, the least accurate research-grade device was the AG-wrist and the least accurate consumer-grade device was the WT. For ADL, the research-grade AG-hip (LFE) was the only device that was equivalent to hand count. Most devices undercounted steps during ADL, with the exception of AG-wrist, AG-wrist LFE, and FC.
In this study, we observed a general trend for devices to underestimate steps. This was especially apparent during slow treadmill walking (<80 m·min−1) and intermittent activities (i.e., meal preparation and sweeping a room) included in the protocol. Previous research has shown that slow, continuous walking and short, intermittent bouts result in undercounting of steps by most activity trackers (23,25,27–29,31,33). Slow walking, less than 80 m·min−1, results in insufficient peak accelerations needed to register a step (35) in several consumer-grade devices (23,27,28,31) and some research-grade devices (25,28). Thus, when examining populations that walk slowly (e.g., older individuals), a research-grade device like the SW should be considered for use.
The number of steps counted by the research and consumer-grade wrist worn devices was greatly reduced when the hands were stabilized on an external object (e.g., holding treadmill rails or pushing a stroller), rather than freely swinging. Chen et al. (31) also found that wrist-worn consumer-grade trackers (Fitbit Flex, Jawbone UP, and GV) underestimated steps when pushing a stroller. Our study shows that the step counts from hip and ankle-worn devices are less likely to be affected by holding onto fixed, external objects. Thus, individuals who regularly push strollers or use rolling-assistive walking devices (i.e., rollators) should consider a hip- or ankle-worn device.
Undercounting of steps can also occur with brief, intermittent walking bouts (29,33). A newer method of reducing false-positive steps is the application of a continuous stepping requirement to step-counting algorithms. For steps to be counted toward the aggregate total, continuous stepping for a specified number of seconds (or steps) must occur. These devices temporarily hold the steps taken at the start of a walking bout in a cache. If the walking bout continues beyond a certain point, the steps from the cache are retroactively added to the aggregated total. If the walking ceases before the time or step threshold is met, the steps in the cache are deleted. An example of this method is the 4-s stepping requirement in Omron pedometers (36). With the Omron, steps taken in bouts lasting less than 4 s will not be added to the aggregated step total. A continuous stepping requirement is not limited to the OM, and other manufacturers (Fitbit, Garmin, and Withings ), as well as ActiGraph's MAVM algorithm, use a similar approach (J. Wyatt, 2017, personal communication, May 3).
Devices that require continuous stepping do not capture all steps taken throughout a day, but they can track steps taken in continuous walking bouts lasting more than 4–8 s. Therefore, they could be sufficient for walking interventions, or for those who are using activity trackers to motivate themselves to become more active. Because devices with continuous stepping requirements do not capture steps in short bouts, they would tend to underestimate steps in people who perform very short bouts, interspersed with rest periods (e.g., store clerks, wait staff).
Although step counts were almost universally underestimated in this study, overcounting remains a concern because of extraneous body movements that can register a step. Over the course of a day, these “false-positive” step counts could accumulate to meaningful levels. In the current study, overestimation of steps mainly occurred in wrist-worn devices during activities that involved repetitive hand movements such as hair brushing and folding laundry. If trackers are to be used with individuals who spend a large part of their day performing tasks that involve repetitive arm movements, hip- and ankle-worn devices should be considered.
The strengths of this study include the use of 10 step-counting devices (using 14 step-counting methods) and a wide range of activities. In addition, the use of equivalence testing provides a newer statistical approach to examining the accuracy of wearable devices compared with a criterion method of step counting. Limitations of the study include the use of laboratory-based ADL and the fact that the activities were only performed for 2 min each. Thus, the data are not representative of free-living conditions.
In summary, this study investigated the accuracy of research and consumer-grade step-counting devices. Fifteen different activities across three activity categories (treadmill locomotion, overground walking, and ADL) were examined. The SW and AG-hip (LFE), both research-grade devices, were found to be the most accurate devices overall, recording 99.5% and 94.4% of actual steps, respectively, across all activities. Consumer-grade devices (FZ, YX, OM, FC, GV) underestimated steps by approximately 10%. The other AG methods worn on the hip and wrist (with the exception of the AG-wrist) and the WT underestimated by 20%–30%. The AG-wrist had the greatest step-counting error.
Consumers, clinicians, and researchers need to be aware of the errors in consumer and research-grade step-counting devices. Although the present study focused solely on validity, other device characteristics such as cost, ease-of use, wearability, and data management software should be considered. In addition, the population being examined must also be considered. For older individuals and clinical populations who have altered or slow gait, it may be necessary to use the SW. In addition, when the greatest accuracy in step counting is, the ankle-worn SW is recommended. However, consumer-grade trackers are suitable for tracking physical activity during behavioral interventions, where the outcome of interest is the change in physical activity. Finally, when drawing comparisons between studies, one must keep in mind that there are differences in the percent of steps recorded by various devices.
This study was not funded.
D. B. is a member of scientific advisory board of ActiGraph, LLC. S. C., L. T., W. P., D. S., P. H., A. M., and C. S. have no conflicts of interest. The results of the present study do not constitute endorsement by the American College of Sports Medicine. The results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.
4. Wright SP, Brown TSH, Collier SR, Sandberg K. How consumer physical activity monitors could transform human physiology research. Am J Physiol Regul Integr Comp Physiol
5. Almalki M, Gray K, Sanchez FM. The use of self-quantification systems for personal health information: big data management activities and prospects. Health Inf Sci Syst
6. Beg MS, Gupta A, Stewart T, Rethorst CD. Promise of wearable physical activity monitors in oncology practice. J Oncol Pract
7. Appelboom G, Yang AH, Christophe BR, Bruce EM, Slomian J, Bruyère O, et al. The promise of wearable activity sensors to define patient recovery. J Clin Neurosci
8. Demeyer H, Burtin C, Van Remoortel H, et al. Standardizing the analysis of physical activity in patients with COPD following a pulmonary rehabilitation program. Chest
9. Brocklebank LA, Falconer CL, Page AS, Perry R, Cooper AR. Accelerometer-measured sedentary time and cardiometabolic biomarkers: a systematic review. Prev Med
10. Biddle SJ, Bennie JA, Bauman AE, et al. Too much sitting and all-cause mortality: is there a causal link? BMC Public Health
11. Bravata DM, Smith-Spangler C, Sundaram V, et al. Using pedometers to increase physical activity and improve health: a systematic review. JAMA
12. Richardson CR, Newton TL, Abraham JJ, Sen A, Jimbo M, Swartz AM. A meta-analysis of pedometer-based walking interventions and weight loss. Ann Fam Med
13. Boone DA, Coleman KL. Use of a step activity monitor in determining outcomes. J Prosthet Orthot
14. Tudor-Locke C. Steps to better cardiovascular health: how many steps does it take to achieve good health and how confident are we in this number? Curr Cardiovasc Risk Rep
15. Bassett DR, Toth LP, LaMunion SR, Crouter SE. Step counting: a review of measurement considerations and health-related applications. Sports Med
16. Bassett DR, Strath SJ. Use of pedometers to assess physical activity. In: Welk GJ, editor. Physical Activity Assessments for Health-Related Research
. Champaign (IL): Human Kinetics; 2002. pp. 163–77.
17. Hornbuckle LM, Kingsley JD, Kushnick MR, et al. Effects of a 12-month pedometer-based walking intervention in women of low socioeconomic status. Clin Med Insights Womens Health
. 2016;9(Suppl 1):75–84.
18. Sisson SB, Camhi SM, Church TS, Tudor-Locke C, Johnson WD, Katzmarzyk PT. Accelerometer-determined steps/day and metabolic syndrome. Am J Prev Med
19. Crespo CJ, Keteyian SJ, Heath GW, Sempos CT. Leisure-time physical activity among US adults. Results from the third national health and nutrition examination survey. Arch Intern Med
20. Siegel PZ, Brackbill RM, Heath GW. The epidemiology of walking for exercise: implications for promoting activity among sedentary groups. Am J Public Health
21. Watson KB, Frederick GM, Harris CD, Carlson SA, Fulton JE. US adults' participation in specific activities: Behavioral Risk Factor Surveillance System—2011. J Phys Act Health
22. Nelson MB, Kaminsky LA, Dickin DC, Montoye A. Validity of consumer-based physical activity monitors for specific activity types. Med Sci Sports Exerc
23. Fokkema T, Kooiman TJ, Krijnen WP, Van Der Schans CP, De Groot M. Reliability and validity of ten consumer activity trackers depend on walking speed. Med Sci Sports Exerc
24. Kooiman TJ, Dontje ML, Sprenger SR, Krijnen WP, van der Schans CP, de Groot M. Reliability and validity of ten consumer activity trackers. BMC Sports Sci Med Rehabil
25. Feito Y, Bassett DR, Thompson DL. Evaluation of activity monitors in controlled and free-living environments. Med Sci Sports Exerc
26. Füzéki E, Engeroff T, Banzer W. Health benefits of light-intensity physical activity: a systematic review of accelerometer data of the national health and nutrition examination survey (NHANES). Sports Med
27. Crouter SE, Karabulut M, Bassett DR Jr. Validity of 10 electronic pedometers for measuring steps, distance, and energy cost. Med Sci Sports Exerc
28. Hickey A, John D, Sasaki J, Mavilia M, Freedson P. Validity of activity monitor step detection is related to movement patterns. J Phys Act Health
29. Toth LP, Park S, Pittman WL, Sarasaltik D, Morton AL, Bassett DR. Step count filters in wearable step counters. Med Sci Sports Exerc
30. Hasson R, Haller J, Pober D, Staudenmayer J, Freedson P. Validity of the Omron HJ-112 pedometer during treadmill walking. Med Sci Sports Exerc
31. Chen MD, Kuo CC, Pellegrini CA, Hsu MJ. Accuracy of wristband activity monitors during ambulation and activities. Med Sci Sports Exerc
32. Orendurff MS, Schoen JA, Bernatz GC, Segal AD. How humans walk: bout duration, steps per bout, and rest duration. J Rehabil Res Dev
33. Silcott NA, Bassett DR Jr, Thompson DL, Fitzhugh EC, Steeves JA. Evaluation of the Omron HJ-720ITC pedometer under free-living conditions. Med Sci Sports Exerc
34. Consumer Technology Association. Physical Activity Monitoring for Fitness Wearables: Step Counting
. Arlington (VA): Technology & Standards Department; 2016.
35. Bassett DR, Ainsworth BE, Leggett SR, Mathien CA, Main JA, Hunter DC, et al. Accuracy of five electronic pedometers for measuring distance walked. Med Sci Sports Exerc
© 2018 American College of Sports Medicine
36. Omron Healthcare Inc. Instruction Manual Pedometer Downloadable Model HJ-722U
. Lake Forest (IL): Omron Healthcare Inc.; 2012.