In recent years, the consumer marketplace has been flooded with an array of activity monitors designed to enhance self-monitoring and behavior change. Accelerometry-based activity monitors have been widely used in research applications for many years (11,30), but the availability of these devices to consumers is a relatively recent phenomenon. According to ABI Research, a market research and intelligence firm, more than 32 million wearable activity and health devices were sold in 2013, and the number will increase to approximately 42 million in 2014. In addition, several research studies have adopted consumer monitors to serve as self-assessment and monitoring tools for clinical and research purposes (3,4,12,25). The trend has been toward wrist-worn activity tracking devices that link to cell phone and social media applications for personalized monitoring. Some consumer activity monitors utilize online platforms that allow users to share progress with friends as a means of peer support, but nearly all provide direct estimates of time spent in physical activity (e.g., minutes of moderate and vigorous activities) and estimates of total energy expenditure (EE), physical activity EE, or steps. Some monitors [e.g., BodyMedia Core (BMC) and Jawbone Up24 (JU24)] segment data into different intensity zones, whereas others include ancillary outcome measures, including sleep time, distance, physical activity time, sedentary time, and heart rate.
Wearable monitors have been shown to have utility as effective assessment and monitoring tools in a variety of settings (2,3,4,12,21,23–25). Bassett and John (5) highlighted the monitors’ convenient display and utility as key features for clinical populations. Appelboom et al. (3) recently summarized key features of smart wearable sensors in a variety of clinical applications (patients with cardiopulmonary, vascular, and neurological dysfunctions, as well as physical therapy and rehabilitation). The rapid technological advancements in consumer-based monitors offer considerable potential for clinical applications and may also serve as more cost-effective and appealing intervention methods for behavior change applications.
Despite the widespread adoption and potential applications of these devices, there is relatively little information about their accuracy. A previous report by Lee et al. (22) demonstrated consumer monitors’ reasonable accuracy in estimating EE. BodyMedia FIT and Fitbit (i.e., Zip and One) were found to be the most accurate (compared to criterion estimates from a portable metabolic analyzer), with mean absolute percent error (MAPE) values of 9.3%, 10.1%, and 10.4%, respectively. A number of other studies have reported on the accuracy of Fitbit in estimating steps counts (1,29) and EE (13,17,22,28), but other consumer monitors are not well represented in the literature. Limitations of past studies include the use of controlled laboratory activities and the lack of direct comparisons with other objective monitors. Preliminary evidence suggests that the accuracy of the current generation of consumer monitors may be comparable to findings from more established research monitors such as the ActiGraph or the BodyMedia armband (22); however, direct comparisons are needed to empirically test this (31). Moreover, new monitors have been released into the market; they need to be tested relative to other consumer monitors to evaluate their validity. The present study fills this gap by examining the relative validity of the most recent wrist-worn consumer monitors and research-based monitors in estimating EE concurrently against a criterion measure (i.e., portable metabolic analyzer). The findings from this study may inform consumer choices but also have important implications for the potential adoption and utilization of these devices in research applications. The use of a naturalistic design that includes sedentary activity and different exercise modalities (aerobic and resistance) provides novel insights into the monitors’ utility for typical consumer use while also providing a sound comparison of the relative validity of different consumer monitors for researchers interested in applying these devices to field-based research applications.
Fifty-two (28 male and 24 female) individuals (age 18–65 yr) volunteered to participate in the study. Individuals were recruited within the university and adjacent community via posted statements, E-mails, laboratory website, and word of mouth. The study protocols were approved by the institutional review board of Iowa State University. A self-reported health history questionnaire was employed to screen for participants who were unable to safely perform 25 min of self-selected aerobic activity and 25 min of resistance exercises.
Oxycon Mobile (OM) version 5.0 (Viasys Healthcare Inc., Yorba Linda, CA) was used as a criterion measurement in this study. EE was estimated from a direct measurement of oxygen consumption and carbon dioxide production. OM is a portable metabolic analyzer that is widely used as a criterion reference in validation studies examining EE from wearable monitors (18). OM volume and gas were calibrated before each trial.
Research activity monitors
Two research-based activity monitors were included in the current study. ActiGraph GT3X+ (GT3X+; Actigraph, Pensacola, FL) is a triaxial accelerometer commonly used to objectively quantify physical activity under free-living conditions. GT3X+ is small (4.6 × 3.3 × 1.5 cm) and has been shown to provide reasonably accurate estimates of physical activity (8). GT3X+ was initialized with 100 Hz, and data were exported in 60-s epochs. Herein, it is compared with other consumer-based monitors.
BMC (BodyMedia Inc., Pittsburgh, PA) is a commercially available multisensor device that utilizes a triaxial acceleromete, heat flux, skin temperature, and galvanic skin response to estimate EE and other activity metrics. The device is small (40 × 65 × 11 mm) and weighs 30 g. It is worn with an adjustable strap on the back of the triceps. It tracks EE (total EE and activity-specific EE), duration and intensity of different types of activities, and sleep efficiency.
Wrist-worn consumer activity monitors
Five wrist-worn consumer activity monitors were tested, including Fitbit Flex (FBF; Fitbit Inc., San Francisco, CA), JU24 (Jawbone, San Francisco, CA), Misfit Shine (MS; Misfit Wearables, Burlingame, CA), Nike+ Fuelband SE (NFS; Nike Inc., Beaverton, OR), and Polar Loop (PL; Polar Electro, Kempele, Finland). All five monitors use a triaxial accelerometer to track steps and to estimate EE, except for NFS, which only records physical activity EE instead of total EE. Additionally, JU24 also tracks resting and active EE separately. Three monitors (i.e., FBF, JU24, and MS) track distance travelled, whereas four monitors (all except for NFS) provide indicators of sleep duration and quality. A recent report in the Journal of the American Medical Association compared the step counts accuracy of several consumer monitors (9), but studies to date have not evaluated EE estimates, which provide a more robust and relevant outcome indicator and serve as the underlying basis for estimates of time spent in different intensities of physical activity.
Participants signed the informed consent form and were encouraged to ask any questions before the beginning of the study protocol. Next, participants were asked to complete the screening assessment (i.e., self-reported health history questionnaire). Anthropometric measures, including height, weight, and percent body fat, were taken at the onset of data collection. Standing height was measured to the closest 0.1 cm using the Harpenden stadiometer (Harpenden, London, UK). Body mass was measured to the nearest 0.1 kg using an electronic scale (Seca 770). A handheld bioimpedance analysis device (Omron, Shelton, CT) was used to assess participants’ percent body fat.
Following this, each participant was fitted with seven different monitors (two research units and five consumer-based monitors) and a portable metabolic system (i.e., OM). FBF, PL, and MS were positioned on the left wrist, and NFS and JU24 were worn on the right wrist. GT3X+ was placed on a belt around the left waist, and BMC was positioned over the triceps of the nondominant arm. All devices (including OM) were initialized before the beginning of data collection.
Each participant completed an 80-min protocol consisting of three distinct components: 1) 20 min of sedentary activity (e.g., reading, working with cell phone, typing, watching videos, listening to music/radio, or any other activities requiring sitting without speaking); 2) 25 min of aerobic exercise on a treadmill at self-selected speeds (walking, jogging, or running); 3) 25 min of resistance exercise (self-selected weights, repetition, and sets from any of the 12 machine—rotary torso, abdominal crunch, lower back, vertical traction, shoulder press, arm curl, arm extension, leg press, leg curl, leg extension, multihip, and chest press) on a TechnoGym training equipment. Five-minute break times between each session were included to facilitate transitions between activities and data recording. The participants were allowed to switch activities under the same categories in each session to more directly simulate real-world conditions. For example, a participant could switch from reading to typing during the sedentary session or alter treadmill speeds during the aerobic exercise session.
All of the consumer monitors provide real-time monitoring of EE; thus, these values were recorded at the beginning and end of each session. EE readings were taken from each of the monitor-specific apps on the iPad (e.g., FBF, JU24, and MS) or directly from the device (e.g., NFS and PL). EE data from OM, GT3X+, and BMC were downloaded and processed for every minute after the 80-min session had been completed.
For GT3X+, Freedson’s 98 regression (an age-specific activity EE prediction equation) was used if activity counts were greater than or equal to 1952 counts per minute; the Work–Energy Theorem was applied to estimate EE for activity counts less than 1952 counts per minute (16). OM, FBF, JU24, MS, PL, and BMC provided total EE values, but GT3X+ and NFS yielded estimates of activity EE without accounting for resting metabolic rate. To facilitate direct comparisons across the various monitors, we calculated resting EE for each participant using the following Mifflin–St. Jeor method (14):
Estimated resting EE values were added to measured activity EE values for both GT3X+ and NFS to calculate the total EE.
Descriptive analyses were conducted to summarize the demographic characteristics of the participants, including age, weight, height, body mass index (BMI), and percent body fat. Absolute MET values based on measured OM and estimated resting EE using the Mifflin–St. Jeor method were calculated in order to determine individualized MET values. Pearson’s correlations were calculated to examine the overall association among OM and consumer/research-grade monitors for total EE. Mean bias (criterion − monitors) point estimate and 95% confidence intervals were computed. Limits of agreement were computed as mean bias ± 1.96 SD to evaluate individual-level agreement. MAPE was calculated for each of the monitors by dividing absolute bias (criterion − monitor) by the criterion measure.
Mean bias provides the overall underestimation or overestimation of EE by each monitor compared to OM, but MAPE provides a more conservative estimation of individual-level error. The total EE on sedentary activity, aerobic exercise, and resistance exercise, as measured by OM and monitors, were calculated separately to further explore measurement errors. Mean bias, MAPE, and effect size (ES) from the three sessions were also computed separately for each activity monitor.
In addition to applying the traditional inferential statistics of mean differences, this study employed equivalence testing in order to more directly determine whether the activity monitors agree with OM at group level. Specifically, we tested the null hypothesis (“There is a difference between criterion measure and monitor”) to enable a direct and interpretable conclusion if the null hypothesis is rejected (i.e., that the two methods are equivalent to each other). To facilitate this statistical testing, we need to predefine an equivalence zone. In the current study, the equivalence zone was defined as ±10% of the mean of OM. A physical activity monitor is equivalent to OM with, on average, a 95% precision if the 90% confidence interval from the activity monitor falls within the defined equivalence zone.
Participants’ mean BMI and percent body fat were 24.0 kg·m−2 and 21.2%, respectively. BMI ranged from 17.6 to 39.9 kg·m−2, percent body fat ranged from 6.2% to 43.7%, and age ranged from 18 to 60 yr. The mean MET values for sedentary activity, aerobic exercise, and resistance exercise were 1.62, 7.11, and 3.18 MET, respectively.
Data from each monitor were processed according to the manufacturers’ instructions. However, the estimated EE from PL had to be removed from the analysis due to the failure of the sync features designed to send information from the Polar Flow web service and the mobile app to PL. The web service failed to upload data to PL, and the demographic variables inputted through the mobile app did not sync with PL in real time. This caused discrepancies between the data on PL band and those reported on the mobile app. It was not possible to determine which estimates were correct, so the data from the PL were removed from the analysis.
Descriptive statistics of measured EE and mean bias between OM and activity monitors for the full protocol are summarized in Table 1. The mean EE for the overall study protocol was 316.8 kcal (measured by OM). The estimated EE from the activity monitors ranged from 274.5 kcal (NFS) to 395.5 kcal (MS). BMC provided the narrowest 95% limits of agreement (−122.7 to 51.7 kcal), followed by FBF (−126.3 to 85.5 kcal), NFS (−65.7 to 150.3 kcal), and GT3X+ (−104.3 to 128.1 kcal), whereas values for MS were the widest (−243.3 to 98.5 kcal). In addition, Bland–Altman plots (Fig. 1) revealed more information on the distribution of individual-level errors and possibilities of systematic bias compared to measured EE. The plots indicated a less evident systematic bias for FBF and GT3X+ due to well-distributed data clustering. In regards to the mean bias, GT3X+, JU24, and NFS underestimated the total EE by 11.9, 23.1, and 42.3 kcal, respectively. For the rest of the monitors, however, overestimation of EE was observed, ranging from 20.4 kcal (FBF) to 72.4 kcal (MS). BMC yielded the smallest 95% confidence interval of mean bias (−48.3 to −22.7), whereas MS had the largest 95% confidence interval (−97.2 to −47.6).
Table 2 summarizes the EE values obtained from OM and the activity monitors for each of the three activity segments (i.e., sedentary activity, aerobic exercise, and resistance exercise), as well as the mean bias and ES. Relatively small differences were observed for the sedentary activity session (EE values ranged from −5.8 to 11.0 kcal) compared to the aerobic and resistance exercise segments. NFS (9.2 kcal; ES = 0.17), BMC (−16.3 kcal; ES = 0.22), and GT3X+ (−20.5 kcal; ES = 0.32) had the smallest mean bias and ES in EE during the aerobic exercise, whereas BMC (−15.3 kcal; ES = 0.53), FBF (25.5 kcal; ES = 1.31), and NFS (26.3 kcal; ES = 1.34) performed better than the other monitors in terms of mean bias during the resistance exercise. BMC consistently overestimated EE across the three types of activity segments, whereas NFS consistently underestimated EE on the same segments. In contrast, FBF, JU24 and MS all overestimated EE in aerobic exercise and underestimated EE in sedentary activity and resistance exercise.
Table 3 presents a correlation matrix for OM and the activity monitors. BMC had the highest correlation with OM (r = 0.90), followed by FBF (r = 0.78) and JU24 (r = 0.77). All other monitors were also highly correlated with OM, each exceeding 0.70. The activity monitors had moderate to high correlations with one another (r = 0.47–0.89).
MAPE values for total EE and different activity segments EE from the various monitors are presented in Figure 2. Five monitors had MAPE values lower than 20% (BMC, 15.3%; GT3X+, 16.7%; FBF, 16.8%; NFS, 17.1%; JU24, 18.2%), whereas one had considerably higher MAPE (MS, 30.4%). MAPE values were generally larger and more variable when examined separately for the three activity segments (Fig. 2). MAPE in sedentary behavior was lowest for BMC (15.7%), followed by MS (18.2%) and NFS (20.0%). BMC (17.2%) and NFS (18.5%) also outperformed the other monitors in aerobic exercise. BMC yielded the lowest MAPE for resistance exercise (29.2%), with values from the other monitors ranging from 31.6% to 52.6%.
Equivalence test results are displayed in Figure 3. The equivalence zone calculated from OM was 285.1 to 348.5 kcal. The calculated 90% confidence interval from GT3X+ (90% confidence interval, 287.1, 324.8) fell within the equivalence zone, indicating group-level equivalence with the criterion measure. The zones for FBF, JU24, and BMC partially overlapped with the equivalence zone but could not be considered as providing equivalent group-level estimates as OM (i.e., criterion).
The present study investigated the concurrent validity of multiple research-based and consumer-based activity monitors for estimating EE during three simulated free-living activities (a sedentary period, an aerobic exercise session, and a resistance exercise session). The results indicated that some consumer monitors (i.e., FBF, JU24, and NFS), but not all, provide comparable accuracy for total EE estimation as research-grade monitors. A unique feature of the study was the naturalistic design that sought to replicate free-living conditions. In contrast with traditional validation studies, the participants were not instructed to perform specific activities but were given the option to select the type and intensity of activity they preferred for three different contexts. For instance, in the sedentary segment, participants were free to select from listening to music, watching video, reading, and typing interchangeably (and in any order) during the 20-min session. They could also stand up to stretch or walk around briefly intermittently to simulate a predominantly sedentary (but naturalized) period of time. This unique design retained elements of free-living behavior while still enabling us to explore the relative sources of measurement error in different contexts. To date, this has not been possible with consumer monitors due to these devices’ limited capability for time segmentation.
Overall error estimates were similar to results from a previous evaluation of consumer monitors (22), although different models and monitors were used, with overall MAPE values typically ranging from 17% to 18% for three of the consumer monitors (FBF, JU24, and NFS). However, the present results demonstrate that these reasonable error rates for the full protocol are largely due to cancelling of overestimation and underestimation from the other activities. For example, some monitors overestimate EE in certain activities but underestimate EE during others (e.g., FBF, JU24, and MS). This cancellation of error led to an illusion of improved accuracy compared to monitors that exhibited a more consistent pattern of overestimation or underestimation of EE (e.g., BMC and NFS).
A novel finding of the study is that none of the monitors accurately estimated EE in resistance exercise. Resistance exercise is a common form of exercise—recommended for all Americans to be performed at moderate or high intensity two or more days a week (26), but few studies have evaluated error in the estimation of EE during this activity (7). Not surprisingly, none of the monitors in the current study were able to accurately capture the energy cost of resistance exercise. Most monitors are likely designed to recognize locomotive movement and cannot capture the increased energy cost associated with lifting activities. The magnitude of error is noteworthy because participants were only lifting for a relatively short span of time during the 25-min interval.
An advancement over past validation studies is that both individual-level and group-level accuracy in EE estimation were evaluated. MAPE values and 95% limits of agreement provide a useful indicator of individual-level validity and reflect the error that individuals could expect if they were tracking their activity patterns. BMC, an established research device, had the smallest MAPE for total EE (15.3%), the lowest values for all three individual activity categories, and the most precise estimate based on 95% limits of agreement. This finding is consistent with past studies that evaluated multisensor armband technology (20,22). However, it is noteworthy that three consumer monitors (FBF, JU24, and NFS) yielded comparable errors for individual EE estimation as GT3X+ (16.8%), which is the most widely used research monitor. Mean bias, equivalence testing, and correlations provide alternative indicators of group-level accuracy. The mean bias and equivalence testing results favored GT3X+, but FBF, JU24, and BMC provided similar group-level validity. It is surprising that the group-level estimate from BMC was outside the group equivalence zone, despite having the smallest individual error. However, closer examination of the data revealed that BMC yielded extremely tight estimates for approximately half of the participants but larger errors for others. Similar patterns were evident for FBF and JU24, which further justifies the need to examine both individual-level and group-level estimates.
The current study was constrained to wrist-worn consumer monitors to standardize the study design and to minimize the measurement variation introduced by the placement of the monitors. Studies investigating research-based accelerometers showed that different placements of the monitors caused varied estimations of raw accelerometer output (19), activity EE (27), total EE (10), and MET values (27). Contemporary consumer monitors have emphasized wrist locations for their less obstructive placement and users’ convenience in checking their progress throughout the day (some monitors have display screens for this purpose, including NFS and PL). Wrist locations facilitate integration with telecommunications features (i.e., smart watch), enable sleep detection, and promote participant compliance (15). With some exceptions, the results demonstrate that some consumer-based monitors provide comparable accuracy (both individual and group) to more established research-grade devices. This is an important finding because it provides documentation supporting the utility of these new devices for both personal use and research applications. GT3X+ is the only monitor that fell within the equivalence zone, but it is not clear if this was due to cancellation of error, the type and nature of the activities, or the position on the hip (instead of the wrist or arm). It is possible, for example, for wrist-worn monitors to have considerable error for some free-living activities, particularly those not involving upper-body movements.
The ability to monitor behavior in real time and to link data to phone and social media apps may help to directly facilitate individual behavior change. A recent study (23) examined the behavior change strategies incorporated in 13 wearable monitors (including six of the monitors tested in the present study). The review compared the extensive feedback delivered through the consumer monitor companion computer or mobile apps based on both theory-based (e.g., social–cognitive theory) criteria and adoption of evidence-based effective behavior change techniques. The results indicated that self-monitoring, feedback, environmental change, goal-setting, and emphasizing discrepancy between current behavior and goal were used in all of the reviewed monitor apps. In addition, the behavior change techniques, such as behavioral goals, social support, social comparison, prompts/cues, rewards, and focus on past success, were also included in more than half of the monitors.
The present study supports the incorporation of some consumer monitors into research applications, but differences among the various monitors should be considered. A strength of the study was its adherence to the best practice guidelines for the calibration and validation of wearable monitors (6). The sample was reasonably diverse, with a balance of male (n = 28) and female (n = 24) participants over a representative age range (18–60 yr) and the inclusion of both normal-weight and obese participants (BMI, 17.6–39.9 kg·m−2). The larger variation in sample characteristics may explain the larger measurement error compared to other studies (13,22), but it provides more confidence that the results can be generalized to a broader population. Another strength was the inclusion of research-grade monitors, in addition to the criterion measure (i.e., OM), because it enabled more direct comparisons with the consumer monitors. The study sought to simulate real-world applications but was not without limitations. As noted previously, NFS only provided activity EE output, and the ActiGraph algorithms were also not designed for estimating EE in low-intensity activity and sedentary activity. Estimated resting metabolic rate was added to these two monitors in order to provide a more appropriate estimate, but interpretation of the results for these two monitors requires caution. However, the limited utility of the outcome measures from these devices (i.e., lack of adjustment for resting metabolic rate) also warrants concern and should be noted by users and researchers. A limitation of the design is that the protocol did not include light-intensity free-living activities, which could also make up a large proportion of waking hours in most individuals. The design necessitated the simultaneous use of multiple monitors, which may have performed differently had they been worn individually. The placement of the monitors (right or left wrist) was fixed to standardize the protocol in the current study, but users might choose to wear it on their dominant or nondominant arm. Thus, the effects of arm placement could not be directly evaluated. Plus/minus 10% of the mean of OM-measured EE was used as a defensible range for the equivalence zone, but future research should be conducted to test the utilization of this value selection. Lastly, these estimates were obtained only for a short duration and may not reflect accuracy for whole-day estimates. Additional studies with longer duration and in real-world settings should be conducted to evaluate the validity of the consumer monitors in estimating EE.
In conclusion, the study provided evidence for the relative validity of several contemporary consumer monitors (compared to both research-based monitors and criterion measure). The research-grade monitors had the strongest results, but FBF, JU24, and NFS provided reasonable estimates for both individual-level and group-level comparisons (except for resistance exercise). The validity offers possibility that the consumer monitors could be used for physical-activity-promoting interventions at the population level. This opens up new opportunities to test the underutilized human interaction features of these new devices for behavior change applications.
This study was supported by the College of Human Sciences at Iowa State University.
None of the authors have a professional relationship with companies or manufacturers that might benefit from the results of the present study.
The results of the present study do not constitute endorsement by the American College of Sports Medicine.
1. Adam Noah J, 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
. 2013; 37 (7): 456–62.
2. Albert MV, Deeny S, McCarthy C, Valentin J, Jayaraman A. Monitoring daily function in persons with transfemoral amputations using a commercial activity monitor: a feasibility study. PM R
. 2014; 6 (12): 1120–7.
3. Appelboom G, Camacho E, Abraham ME, et al. Smart wearable body sensors for patient self-assessment and monitoring. Arch Public Health
. 2014; 72 (1): 28.
4. Appelboom G, Yang AH, Christophe BR, et al. The promise of wearable activity sensors to define patient recovery. J Clin Neurosci
. 2014; 21 (7): 1089–93.
5. Bassett DR Jr, John D. Use of pedometers and accelerometers in clinical populations: validity and reliability issues. Phys Ther Rev
. 2010; 15 (3): 135–42.
6. Bassett DR Jr, Rowlands A, Trost SG. Calibration and validation of wearable monitors. Med Sci Sports Exerc
. 2012; 44 (1 Suppl): S32–8.
7. Benito PJ, Neiva C, González-Quijano PS, Cupeiro R, Morencos E, Peinado AB. Validation of the SenseWear armband in circuit resistance training with different loads. Eur J Appl Physiol
. 2012; 112 (8): 3155–9.
8. Butte NF, Ekelund U, Westerterp KR. Assessing physical activity
using wearable monitors: measures of physical activity
. Med Sci Sports Exerc
. 2012; 44 (1 Suppl): S5–12.
9. Case MA, Burwick HA, Volpp KG, Patel MS. Accuracy of smartphone applications and wearable devices for tracking physical activity
. 2015; 313 (6): 625–6.
10. Chen KY, Acra SA, Majchrzak K, et al. Predicting energy expenditure
of physical activity
using hip- and wrist-worn accelerometers. Diabetes Technol Ther
. 2003; 5 (6): 1023–33.
11. Chen KY, Bassett DR Jr. The technology of accelerometry-based activity monitors: current and future. Med Sci Sports Exerc
. 2005; 37 (11 Suppl): S490–500.
12. Cook DJ, Thompson JE, Prinsen SK, Dearani JA, Deschamps C. Functional recovery in the elderly after major surgery: assessment of mobility recovery using wireless technology. Ann Thorac Surg
. 2013; 96 (3): 1057–61.
13. Dannecker KL, Sazonova NA, Melanson EL, Sazonov ES, Browning RC. A comparison of energy expenditure
estimation of several physical activity
monitors. Med Sci Sports Exerc
. 2013; 45 (11): 2105–12.
14. Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc
. 2005; 105 (5): 775–89.
15. Freedson PS, John D. Comment on “Estimating activity and sedentary behavior from an accelerometer on the hip and wrist.” Med Sci Sports Exerc
. 2013; 45 (5): 962–3.
16. Freedson PS, Melanson E, Sirard J. Calibration of the Computer Science and Applications, Inc. accelerometer. Med Sci Sports Exerc
. 1998; 30 (5): 777–81.
17. Gusmer R, Bosch T, Watkins A, Ostrem J, Dengel D. Comparison of FitBit® Ultra to ActiGraph™ GT1M for assessment of physical activity
in young adults during treadmill walking. Open Sports Med J
. 2014; 8: 11–5.
18. Hannink J, Lahaije A, Verberkt C, Dekhuijzen R, van Helvoort H, Heijdra Y. Validity of Oxycon Mobile in measuring inspiratory capacity in healthy subjects. Clin Physiol Funct Imaging
. 2010; 30 (3): 206–9.
19. Hildebrand M, VT Van Hees, Hansen BH, Ekelund U. Age group comparability of raw accelerometer output from wrist- and hip-worn monitors. Med Sci Sports Exerc
. 2014; 46 (9): 1816–24.
20. Johannsen DL, Calabro MA, Stewart J, Franke W, Rood JC, Welk GJ. Accuracy of armband monitors for measuring daily energy expenditure
in healthy adults. Med Sci Sports Exerc
. 2010; 42 (11): 2134–40.
21. Lauritzen J, Muñoz A, Luis Sevillano J, Civit A. The usefulness of activity trackers in elderly with reduced mobility: a case study. Stud Health Technol Inform
. 2013; 192: 759–62.
22. Lee JM, Kim Y, Welk GJ. Validity of consumer-based physical activity
monitors. Med Sci Sports Exerc
. 2014; 46 (9): 1840–8.
23. Lyons EJ, Lewis ZH, Mayrsohn BG, Rowland JL. Behavior change techniques implemented in electronic lifestyle activity monitors: a systematic content analysis. J Med Internet Res
. 2014; 16 (8): e192.
24. Mansfield A, Wong JS, Bayley M, et al. Using wireless technology in clinical practice: does feedback of daily walking activity improve walking outcomes of individuals receiving rehabilitation post-stroke? Study protocol for a randomized controlled trial. BMC Neurol
. 2013; 13: 93.
25. Meyer J, Hein A. Live long and prosper: potentials of low-cost consumer devices for the prevention of cardiovascular diseases. Med 2.0
. 2013; 2 (2): e7.
26. Physical Activity
Guidelines Advisory Committee. Physical Activity Guidelines Advisory Committee Report, 2008
. Washington (DC): US Department of Health and Human Services; 2008. pp. A1–H14.
27. Rosenberger ME, Haskell WL, Albinali F, et al. Estimating activity and sedentary behavior from an accelerometer on the hip or wrist. Med Sci Sports Exerc
. 2013; 45 (5): 964–75.
28. Sasaki JE, Hickey A, Mavilia M, et al. Validation of the Fitbit wireless activity tracker for prediction of energy expenditure
. J Phys Act Health
. 2015; 12 (2): 149–54.
29. Takacs J, Pollock CL, Guenther JR, Bahar M, Napier C, Hunt MA. Validation of the Fitbit One activity monitor device during treadmill walking. J Sci Med Sport
. 2014; 17 (5): 496–500.
30. Troiano RP, Berrigan D, Dodd KW, Mâsse LC, Tilert T, McDowell M. Physical activity
in the United States measured by accelerometer. Med Sci Sports Exerc
. 2008; 40 (1): 181–8.
31. Welk GJ. Principles of design and analyses for the calibration of accelerometry-based activity monitors. Med Sci Sports Exerc
. 2005; 37 (11 Suppl): S501–11.