Technologies for objectively monitoring physical activity have advanced considerably over recent decades from humble mechanical pedometers (3) to more sensitive time-stamped accelerometers with greater data collection capacity (4). Regardless of the instrument employed, however, a simple rendering of steps taken continues to be an enduring output of interest to researchers, practitioners, and the lay public (15). This uncomplicated unit of measurement indicative of a basic and familiar human locomotor movement pattern facilitates translation between studies and communication between different users, but only to the extent that different instruments provide similar estimates of steps taken.
A potential source of interinstrument variation in step detection capability is the attachment site. For example, a waist-worn attachment site is sensitive to locomotor patterns affected by the stride cycle of both legs (2), whereas an ankle-worn device’s output must be doubled to obtain similar estimates (17). In an effort to improve participant compliance to extended wearing regimens (5), there has been recent interest in moving the accelerometer to a wrist-worn attachment site (11), and this is true for commercial products as well (10). However, preliminary data suggest that methods for estimating energy expenditure and identifying activity intensity thresholds may perform better using data from waist-worn accelerometers than from wrist-worn devices (11). We are not aware of any study at this time that has specifically reported the comparative accuracy of step outputs obtained from waist-worn accelerometers versus wrist-worn devices of the same brand and model. Therefore, we undertook two studies, one in the laboratory and one under a free-living condition, to compare step outputs obtained from waist- and wrist-worn accelerometer attachment sites.
All study procedures were approved by the Pennington Biomedical Research Center Institutional Review Board, and participants provided written informed consent before participating. As part of a larger study with multiple measurement objectives, 15 participants (67% female, age = 27.5 ± 2.5 yr, body mass index = 23.0 ± 3.4 kg·m−2) completed a 3-h controlled laboratory-based assessment and subsequent monitoring of their free-living behaviors for a 1-wk period. Under both conditions, participants were asked to wear (among other devices that are not the focus of these specific analyses) one ActiGraph GT3X+ accelerometer (ActiGraph LLC, Pensacola, FL) attached to a manufacturer-provided band at their right hip in line with the midaxillary line and a second one (also using a provided band) on their nondominant wrist. The ActiGraph accelerometer is the focus of this analysis because this is the brand that has been previously used as a waist-worn device in the 2003–2006 National Health and Nutrition Examination Survey (NHANES) (13) and has also been deployed as a wrist-worn device in the 2011–2014 NHANES cycles (11).
Under the laboratory condition, participants were asked to walk/run on a 0% grade treadmill for 14 incrementally faster 5-min bouts (with a 2-min rest period between bouts). Table 1 displays all tested treadmill speeds. The slowest speed tested was 14 m·min−1, and the fastest speed was 188 m·min−1. During the between-bout rest periods, participants straddled the treadmill belt, which continued to operate at 14 m·min−1. The treadmill speed was promptly increased at the beginning of each bout to the designated speed. Steps were visually counted during each bout, and a redundant video recording was used for later confirmation of visually counted steps as required. Observed steps per minute were determined by dividing the total steps counted in each bout by 5 reflecting the 5-min bout duration. Accelerometer-determined steps per minute was calculated from the average of the second to fourth minutes within a bout to eliminate the potential error associated with asynchronous bout starting times and accelerometer time stamps.
Under the free-living condition, participants were asked to wear both accelerometers for seven consecutive days, 24 h·d−1, removing the devices only for extended contact with water (e.g., bathing and swimming). Only data from 7:00 a.m. to 10:00 p.m. daily were considered for this free-living analysis to focus explicitly on daytime ambulatory behaviors assessed by both the waist- and wrist-worn accelerometers. A day was considered valid when a participant achieved at least 1500 steps on both the waist and wrist accelerometers with the default frequency filter. The average number of valid days per participant was 5.5 ± 1.8.
Under both conditions, accelerometer step data were processed in 60-s epochs using ActiLife software (version 6.0 or higher, ActiGraph LLC) and applying both the manufacturer’s default and low-frequency extension filters. The low-frequency extension filter is a data processing option that increases the instrument’s sensitivity to low-force accelerations (e.g., slow walking) detected at the waist attachment site (1), but its utility for processing wrist-detected data is unknown. ActiGraph uses a proprietary algorithm to count steps, and no information about the specifics of this process is publically available. Our personal communication with the manufacturers indicates that only the vertical axis is considered in determining steps taken.
Paired sample t-tests were used to evaluate speed-specific mean differences in measured steps per minute between the criterion of direct observation and the four estimates (attachment site × filter) produced from the waist- and wrist-worn accelerometers during the laboratory study. Differences in mean steps per day detected between the waist- and wrist-worn accelerometers (considering both types of filters) in the free-living study were computed. All statistical analyses were conducted using R (version 3.1.0; R Foundation for Statistical Computing, Vienna, Austria). In an effort to control for multiple testing associated with the large number of hypothesis tests conducted in the controlled laboratory study, we set the level of significance α to 0.001 (two tailed).
Table 1 presents the data obtained in the laboratory-based study. By design, the visually counted step criterion increased with each incrementally faster treadmill speed. The error associated with the waist-worn accelerometer missing steps at slower speeds when using the manufacturer’s default filter was alleviated at speeds ≥67 m·min−1. Steps estimated using the low-frequency extension applied to the waist-worn accelerometer were not significantly different from the criterion at any speed. Although the low-frequency extension filter applied to the wrist-worn accelerometer did appear to provide a reasonable estimate of visually counted steps taken at treadmill speeds ≤40 m·min−1 (although the default filter applied to the same device and attachment site did not), neither filter provided acceptable estimates at the wrist attachment site relative to the criterion of observed steps for any other tested speed.
Under free-living conditions, the waist-worn accelerometer detected 6743 ± 2398 (default filter) and 13,029 ± 3734 (low-frequency extension) steps per day. The concurrently worn wrist accelerometer detected 9301 ± 2887 (default filter) and 15,493 ± 3958 (low-frequency extension) steps per day. The smallest difference between the attachment sites was obtained by comparing results from the default filter processing: −2558 steps per day. The largest difference of −8751 steps per day was observed between the waist-worn accelerometer processed using the default filter and the wrist-worn device processed with the low-frequency extension filter.
This is not the first study to compare outputs from accelerometers worn concurrently at the waist and wrist. Rosenberger et al. (11) reported that Wocket accelerometers (an open-source technology) located at the waist generally performed better than those worn at the wrist in terms of estimating energy expenditure and time spent in sedentary behavior and different physical activities evaluated under laboratory conditions. Lambiase et al. (8) compared 24-h × 7-d free-living outputs obtained from a waist-worn ActiGraph GT1M and a wrist-worn Actiwatch-2 (Philips Healthcare, Andover, MA) and reported that “total movement volume” (defined as any movement detected expressed as an average of activity counts over the specified period) correlated between the two devices during logged waking periods (r = 0.47, P < 0.001) and to a lesser extent during active periods (r = 0.26, P < 0.01), defined as any time >100 activity counts per minute as recorded by the waist-worn device. They concluded that although the wrist-worn Actiwatch-2 could be used to rank total movement volume, its utility for assessing physical activity was substandard. We know of only a single study (7) that has focused on step-counting accuracy of different waist- and wrist-worn devices. A deliberate focus on the step-counting output is rational because this particular metric is common to most motion-sensing devices and it is of pragmatic value to a range of end users. Fortune et al. (7) focused on step-counting accuracy of various devices including a wrist-worn Nike FuelBand (Nike, Beaverton, OR) under laboratory conditions, specifically crossing an 8.5-m walkway (with additional acceleration and deceleration space) at velocities of 0.1 to 4.8 m·s−1 (comparable with 6 to 288 m·min−1). In contrast to waist- and ankle-worn Fitbit Trackers (Fitbit, San Francisco, CA) that demonstrated a 92%–93% median percentage agreement with visually counted steps, the wrist-worn Nike FuelBand displayed only a 33% median percentage agreement. It is difficult to conclude whether the discrepancy in accuracy was due to the attachment site or the device studied. In contrast to these assembled previous studies, we purposely sought to compare step-defined ambulatory activity outputs from waist and wrist attachment sites of the same brand and model of accelerometer under both laboratory and free-living conditions.
We found that the wrist attachment site detected consistently fewer visually counted steps than the waist attachment site at most treadmill speeds during laboratory testing. In contrast, the wrist attachment site produced a higher average step count (ranging from approximately 2500 to 8700 more steps per day in free living, depending on the filter processing applied) than the waist attachment site under free-living conditions. These contrasting findings from the laboratory and free-living conditions clarify that the movements enacted in these contexts are distinct as well. The laboratory, specifically the treadmill, restricts the individual to a rhythmic directionally focused ambulatory movement although there is no such focus or restriction under the free-living condition. Furthermore, it is known that continuous walking behaviors are actually rare against the more prevalent backdrop of sedentary and light-intensity activities inherent to the free-living condition (12). Even in seated postures we actively move our hands (and thus wrists) during eating, conversing, screen-based behaviors, etc. (11) Although we found that the wrist site generally detected fewer true ambulatory steps compared with the waist site at most treadmill speeds, the relatively higher wrist-worn accelerometer step counts observed under the free-living condition compared with the waist-worn values likely reflect these more commonly disjointed wrist and waist movement patterns.
The manufacturers of ActiGraph accelerometers introduced the optional low-frequency extension filtering algorithm to increase sensitivity to low-intensity movements (1). We observed that application of this optional filter did improve step-counting accuracy of the waist-worn accelerometer at treadmill speeds of ≤54 m·min−1 and of the wrist-worn accelerometer at ≤40 m·min−1 under our laboratory condition. This finding is similar to the findings by Feito et al. (6). Although we did not use a criterion standard for steps per day assessed under free-living conditions, application of the low-frequency extension filter to either the waist- or wrist-assessed accelerometers produced an estimate that was >6000 more steps per day than the default filter applied to the same data collected from the same attachment site. Application of the low-frequency extension filter to free-living data is known to produce inflated estimates of steps per day relative to published expected values for steps per day obtained from research grade pedometers (14). Although some may interpret these findings as evidence that people perform many low-intensity stepping movements during the day that would be otherwise missed, we have previously demonstrated that the filter-associated differences in free-living steps per day are not restricted to just these low-force movements (1), even detecting additional steps detected in excess of 100 steps per minute (considered indicative of at least moderate intensity (16)). Increased accelerometer step-counting sensitivity has also been indicated during riding in a car (9). Without an acceptable free-living criterion standard, however, we are left only to remark about the observed differences associated with the application of the different filtering processes. However, Feito et al. also demonstrated that using the low-frequency extension filter can substantially alter the step counts on a waist-worn ActiGraph accelerometer, changing from undercounting steps on the treadmill to overcounting steps in free living when compared with the StepWatch as a criterion under both conditions (6).
Accelerometers provide a broad array of physical activity and sedentary behavior outputs; however, we chose to focus on their step-counting feature because this simple metric is common across most instruments and presumably translatable between studies. We rigorously controlled and systematically collected step data using the same brand and model of accelerometer concurrently worn at both the waist and the wrist. We considered both the manufacturer’s default filter and the low extension filter during processing to provide the most completely inclusive range of step outputs provided by the instrument. During the laboratory segment, we video recorded participants’ steps to provide a back-up recording of visually tallied steps taken. We employed a criterion standard of visually counted steps taken under the laboratory condition, however, and as mentioned previously, this was not possible under the free-living condition. Although the sample size may be considered small, it was sufficient to clearly depict measurement differences attributable to attachment site and filtering processes. Conclusions are limited to the specific device studied; however, our choice of the ActiGraph is rational considering that a previous version of this device was attached to the waist in the NHANES 2003–2006 (13) and a more current version was attached to the wrist in the NHANES 2008–2011 (11). Future research may consider other types of devices, including emerging wrist-worn commercial devices that claim to provide an accurate count of daily steps.
In conclusion, step outputs obtained from waist- and wrist-worn accelerometer attachment sites are generally not comparable under either laboratory or free-living conditions. Thus, we assert that steps per day data collected from the currently deployed NHANES wrist accelerometry cannot be interpreted intuitively as an absolute indicator of ambulatory activity on the same scale as the waist accelerometry collected using a similar ActiGraph model and processed using the current version of ActiLife software, regardless of the processing filter used.
We would like to thank Melissa Lupo and Stefany Achee from the Exercise Testing Center at Pennington Biomedical.
This work was performed without any specified funding support.
The authors declare no conflict of interest.
The results of the present study do not constitute endorsement by the American College of Sports Medicine.
1. Barreira TV, Brouillette RM, Foil HC, Keller JN, Tudor-Locke C. Comparison of older adults’ steps/day using NL-1000 pedometer and two GT3X+ accelerometer filters. J Aging Phys Act
. 2012; 21( 4): 402–16.
2. 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.
3. Bassey EJ, Dallosso HM, Fentem PH, Irving JM, Patrick JM. Validation of a simple mechanical accelerometer (pedometer) for the estimation of walking
activity. Eur J Appl Physiol Occup Physiol
. 1987; 56( 3): 323–30.
4. 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.
5. Colley R, Gorber SC, Tremblay MS. Quality control and data reduction procedures for accelerometry-derived measures of physical activity. Health Rep
. 2010; 21( 1): 63–9.
6. 6. Feito Y, Garner HR, Bassett DR. Evaluation of ActiGraph’s low-frequency filter in lab and free-living environments. Med Sci Sports Exerc
. 2015; 47( 1): 211–7.
7. Fortune E, Lugade V, Morrow M, Kaufman K. Validity of using tri-axial accelerometers to measure human movement—Part II: Step counts at a wide range of gait velocities. Med Eng Phys
. 2014; 36( 6): 659–69.
8. 8. Lambiase MJ, Gabriel KP, Chang YF, Kuller LH, Matthews KA. Utility of Actiwatch sleep monitor to assess waking movement behavior in older women. Med Sci Sports Exerc
. 2014; 46( 12): 2301–7.
9. Le Masurier GC, Tudor-Locke C. Comparison of pedometer and accelerometer accuracy under controlled conditions. Med Sci Sports Exerc
. 2003; 35( 5): 867–71.
10. 10. Lee JM, Kim Y, Welk GJ. Validity of consumer-based physical activity monitors. Med Sci Sports Exerc
. 2014; 46( 9): 1840–8.
11. Rosenberger ME, Haskell WL, Albinali F, Mota S, Nawyn J, Intille S. Estimating activity and sedentary behavior from an accelerometer on the hip or wrist. Med Sci Sports Exerc
. 2013; 45( 5): 964–75.
12. Tudor-Locke C, Camhi SM, Leonardi C, et al. Patterns of adults stepping cadence in the 2005–2006 NHANES. Prev Med
. 2011; 53( 3): 178–81.
13. Tudor-Locke C, Camhi SM, Troiano RP. A catalog of rules, variables, and definitions applied to accelerometer data in the National Health and Nutrition Examination Survey, 2003–2006. Prev Chronic Dis
. 2012; 9: E113.
14. Tudor-Locke C, Craig CL, Aoyagi Y, et al. How many steps/day are enough? For older adults and special populations. Int J Behav Nutr Phys Act
. 2011; 8: 80.
15. Tudor-Locke C, Craig CL, Brown WJ, et al. How many steps/day are enough? For adults. Int J Behav Nutr Phys Act
. 2011; 8: 79.
16. Tudor-Locke C, Sisson SB, Collova T, Lee SM, Swan PD. Pedometer-determined step count guidelines for classifying walking
intensity in a young ostensibly healthy population. Can J Appl Physiol
. 2005; 30( 6): 666–76.
17. White DK, Tudor-Locke C, Felson DT, et al. Do radiographic disease and pain account for why people with or at high risk of knee osteoarthritis do not meet physical activity guidelines? Arthritis Rheum
. 2013; 65( 1): 139–47.