INTRODUCTION
The use of activity monitors has transformed physical activity (PA) research over the past decade as more registered clinical studies are using consumer monitors instead of research monitors (1). Wearable activity monitors provide a unique opportunity to improve PA surveillance, to refine scientific knowledge of the relationships between PA and health, and to quantify the recommended dose of PA for the public (2). Because of the increased use of activity monitors, it is important that they are validated to accurately count steps (3–6), a commonly used PA metric that is easily understood by the public and positively associated with cardiometabolic health (7–9).
An advantage of step count validation under laboratory conditions (e.g., treadmill walking/running and structured bouts of activities) is that monitors can be easily validated against a gold standard (i.e., hand-counted steps). Although the findings from laboratory-based studies are useful in understanding monitor performance under specific conditions (e.g., differing speeds and walking surfaces), they may not capture all types of activities that are performed throughout the day. Free-living validation studies provide greater ecological validity, although it is difficult to obtain hand-counted steps per day because of the difficulties of observing individuals outside the laboratory for long periods of time.
It was not until recently that a method was established for validating research and consumer activity monitors against hand-counted steps, across all waking hours of one day (10). In this study, participants wore a GoPro video camera that captured steps taken throughout the day. The steps were later hand-counted by researchers. In this study, the StepWatch (SW), an ankle-worn research monitor, yielded step counts within 5% of hand-counted steps. This supports its consideration as an alternative criterion measure of steps per day that is similar to direct observation. Limited studies have assessed the step count accuracy of monitors under free-living conditions, using SW as the criterion, and several of these studies have included consumer monitors (11–16).
Numerous consumer and research monitors have been used to examine the association between step count and health outcomes. However, the use of different monitors across studies obscures the dose–response relationship because of diverse monitor validity. Therefore, there is an urgent need to harmonize steps from numerous monitors to the SW, which is a validated criterion for assessing steps per day (17). Proposals and efforts have been made to harmonize device-based estimates of physical behaviors; however, there has been limited work that concurrently calibrates many devices against a single criterion measure. Therefore, the purpose of this study was 1) to determine the step count accuracy of multiple consumer and research monitors compared with SW across all waking hours of 1 d, under free-living conditions, and to 2) develop correction methods to calibrate all monitor step estimates to the SW.
METHODS
Participants
A total of forty-eight healthy adults (mean age ± SD, 28 ± 12 yr) participated in the study. Participants were recruited via word of mouth and flyers at the University of Tennessee, Knoxville, and in the surrounding community. Exclusion criteria included contraindications to exercise (determined by administering the PA Readiness Questionnaire), pregnancy, and participation in stationary cycling or bicycling. The study was reviewed and approved by the Institutional Review Board at the University of Tennessee, Knoxville.
Protocol
The study was conducted over 3 d, and all participants signed an informed consent on the first day. On the first day, participants reported to the laboratory for anthropometric measurements and study instructions. Height and weight were measured using a stadiometer and electronic weight scale, respectively, in light clothing with no shoes. Participants were shown how to properly affix monitors to the corresponding body locations (i.e., wrist, hip, thigh, and ankle) and were given an information sheet that displayed how to wear each monitor correctly. In addition, participants were provided with a PA diary to record the monitor wear times (time on and off) and the step counts displayed at those times for monitors with screens.
For the second and third days, participants were instructed to wear the monitors while going about their daily routine, except during swimming or bathing. On both days, participants wore two wrist, two hip, and one ankle monitors. A thigh monitor was worn on only one of the 2 d. With the exception of the ankle monitor, a different brand of monitor was worn each day.
Wrist Monitors
Each day, two wrist-worn monitors were worn on the nondominant wrist. Wrist monitors included Apple Watch Series 2 (ApW; firmware version 4.0, 15R372), Fitbit Alta (FA; firmware version 21.40.2), Garmin vivofit 3 (GV; firmware version 4.2.2.1), and ActiGraph GT9X (GT9X). Only two monitors were worn each day as a result of a previous study that explored the effect of simultaneously wearing four identical monitors (e.g., 4 ApW) on the wrist/forearm on steps per day (18). Because step counts were attenuated as the monitor was positioned further away from the reference position (i.e., proximal to the ulnar styloid process), the current study aimed to place monitors as close to the wrist as possible. Based on the two monitors that would result in the smallest displacement from the reference position, ApW and GV were worn on one day, whereas GT9X and the FA were worn on the other day.
Hip Monitors
Each day two different hip-worn monitors were worn. Hip monitors included Yamax Digiwalker SW-200 (SW200), Omron HJ-325 (HJ325), GT9X, and Fitbit Zip (FZ; firmware version 90). The monitors were worn on either the anterior axillary line or the midclavicular line on the right hip except the HJ325, which was always worn on the anterior axillary line (18).
Thigh Monitor
One activPAL (AP; firmware version 7.1.2.142) was worn on the anterior midline of the right thigh, midway between the top of the patella and the inguinal crease. It was wrapped in parafilm and attached to the thigh with Tegaderm film (3M Health Care, St. Paul, MN).
Ankle Monitor
One SW was worn on the right ankle just above the lateral malleolus. It was held on by an elastic strap with Velcro® closures. Using the “Easy Start” menu within the Modus software (Modus Health, version 3.4), each SW was initialized with a preprogrammed setting (“both extremes of walking speed”) and participant characteristics (i.e., height, weight, and sex). In a previous study, this procedure was shown to yield daily step counts within 5% of hand-counted steps (10).
Data Processing
GT9X monitors were initialized at 30 Hz with the limb placement setting designated as the monitor wear location (i.e., waist or wrist) using ActiLife 6 software (version 6.13.1; ActiGraph, Pensacola, FL).
After all monitors and PA diaries were returned, total daily steps for monitors with screens (including the GT9X processed with moving average vector magnitude [MAVM]) were obtained by subtracting the steps recorded each morning (when monitors were first affixed) from the steps recorded each evening (when monitors were removed before bed) on the PA diaries.
For SW, AP, and GT9X step methods, data were downloaded for subsequent analysis using Modus Health (version 3.4), activPAL process and presentation (version 7.2.28), and ActiLife 6 (version 6.13.1), respectively. For the GT9X, raw data were processed with and without the low-frequency extension (ActiGraph GT9X low-frequency extension [AGL] and ActiGraph GT9X normal filter [AG], respectively) and exported in 1-min, time-stamped epochs along with steps. For SW, AP, and GT9X step methods (i.e., AGL and AG), total daily steps were determined by summing all steps according to participant’s wear times indicated from the PA diary.
Data Analysis
Descriptive (mean ± SD) and statistical summaries (i.e., percent of SW steps [monitor estimate / SW × 100%], mean difference [monitor estimate − SW], mean absolute percent error [MAPE], and root-mean-square error [RMSE]) were computed using total daily step counts for each monitor. Separate linear regression models were used to estimate the overall monitor statistical summary and the 95% confidence interval (CI). Pearson product–moment correlation coefficients were computed for each monitor.
Specific correction factor regressions were fit and evaluated using leave-one-subject-out cross validation. For each monitor, one participant’s data are held out, and a linear regression model was fit to estimate SW steps using monitor-specific step estimates. The resultant linear model was applied to the holdout participant’s data to produce their corrected step count estimate. This process was repeated until each participant served as the holdout participant. The resultant monitor-specific correction method regressions are the average slope and intercepts across participants. Percentages of SW steps were computed with the corrected step estimates for each monitor. Modified Bland–Altman plots were constructed to display the individual variability of total daily steps between the original (observed data) and the correction method for each monitor. This process was repeated to investigate whether corrected step estimates were improved using other statistical models (quadratic regression, multivariate regression with individual-level covariates, and random forest). Correction factor regressions and statistical analyses were developed and evaluated using cranR (version 3.6.1).
RESULTS
A total of 48 healthy participants participated in the study (Table 1), and only one was left-hand dominant. The average wear time was 13.2 ± 1.7 h (mean + SD), which was calculated based on the start and stop times on the PA diaries.
TABLE 1 -
Participant Characteristics.
|
All (N = 48) |
Female (n = 27) |
Male (n = 21) |
| Age (yr) |
28 ± 12 |
28 ± 12 |
28 ± 11 |
| Height (cm) |
169.8 ± 8.7 |
164.2 ± 6.1 |
177.0 ± 5.7 |
| Weight (kg) |
72.3 ± 18.3 |
63.5 ± 16.0 |
83.6 ± 14.7 |
| BMI (kg·m−2) |
24.9 ± 5.3 |
23.6 ± 5.8 |
26.6 ± 4.0 |
| SW steps per day |
9634 ± 3800 |
9355 ± 3731 |
9996 ± 3869 |
Data are presented as mean ± SD.
BMI, body mass index; SW steps per day, average of both days of wear.
Mean difference, RMSE, and MAPE for monitors are displayed according to placement site (Table 2). Missing data across monitors were due to device malfunctions and/or participant noncompliance with instructions on wearing them. Mean difference ranged widely across all monitors (−3452 to 8772 steps). FA had the lowest MAPE (14.8%, 95% CI = 10.5%, 19.2%), RMSE (2130.0, 95% CI = 1393.2, 2670.7), and mean difference (−855 steps, 95% CI = −1432, −279). FZ (r = 0.902), AGL (r = 0.930) on the hip, and AP (r = 0.910) displayed the highest correlation with SW (Table 3).
TABLE 2 -
Total Daily Steps Across All Waking Hours of One Day in a Free-Living Environment as Measured by Six Wrist Methods, Six Hip Methods, and One Thigh Method, with the Criterion of SW steps (
N = 48).
|
Method |
n
|
Method (Mean ± SD) |
SW (Mean ± SD) |
Mean Difference (95% CI) |
MAPE (%) (95% CI) |
RMSE (95% CI) |
| Wrist |
Research |
AGL** |
45 |
18,631 ± 5,192 |
9,858 ± 3,674 |
8,772 (7767 to 9778) |
89.0 (78.8 to 99.2) |
9408.8 (8359.6 to 10352.3) |
| AGM** |
42 |
6,540 ± 2,939 |
9,772 ± 3,539 |
−3,452 (−4117 to −2787) |
34.6 (27.9 to 41.2) |
4078.5 (3217.0 to 4787.4) |
| AG** |
45 |
11,274 ± 3,414 |
9,858 ± 3,674 |
1,416 (746 to 2037) |
20.3 (15.4 to 25.2) |
2589.9 (199.4 to 3072) |
|
Consumer |
ApW** |
45 |
8,229 ± 3,552 |
9,357 ± 4,093 |
−1,128 (−1696 to −559) |
18.5 (14.1 to 22.9) |
2229.8 (1691.3 to 2661.5) |
| FA* |
45 |
9,029 ± 3,817 |
9,884 ± 3,653 |
−855 (−1432 to −279) |
14.9 (10.3 to 19.5) |
2130.0 (1393.2 to 2670.7) |
| GV** |
46 |
7,590 ± 3,487 |
9,278 ± 4,126 |
−1,664 (−2262 to −1067) |
22.2 (17.0 to 27.4) |
2636.0 (1984.2 to 3156) |
| Hip |
Research |
AGL** |
46 |
12,562 ± 3,943 |
9,765 ± 3,779 |
2,797 (2377 to 3216) |
28.6 (24.3 to 32.9) |
3143.9 (2604.3 to 3603) |
| AGM** |
43 |
7,821 ± 3,870 |
9,674 ± 3,679 |
−2,074 (−2700 to −1449) |
23.2 (17.6 to 28.7) |
2928.5 (2092.1 to 3574.3) |
| AG** |
46 |
7,506 ± 3,063 |
9,765 ± 3,799 |
−2,259 (−2808 to −1709) |
24.0 (18.7 to 29.3) |
2939.3 (2174.0 to 3542.9) |
|
Consumer |
SW200* |
38 |
8,004 ± 4,059 |
9,343 ± 4,101 |
−1,339 (−2001 to −677) |
19.3 (13.6 to 25.0) |
2452.7 (1548.9 to 3103.7) |
| FZ** |
46 |
7,623 ± 3,584 |
9,489 ± 3,975 |
−1,867 (−2363 to −1371) |
22.2 (18.0 to 26.5) |
2523.7 (1971.3 to 2975.2) |
| HJ325** |
46 |
6,652 ± 2,792 |
9,274 ± 3,627 |
−2,623 (−3269 to −1976) |
29.7 (23.3 to 36.2) |
3431.1 (2543.3 to 4132.4) |
| Thigh |
Research |
AP** |
37 |
8,290 ± 3,822 |
9,857 ± 4,147 |
−1,567 (−2122 to −1011) |
20.8 (17.3 to 24.4) |
2312.6 (1893.9 to 2666.3) |
Mean: total daily steps. Mean difference: method estimate − SW estimate.
Significant difference between method and criterion (SW) indicated by * P < 0.05 and ** P < 0.01.
AGM, ActiGraph GT9X MAVM; SW200, Yamax Digiwalker SW-200.
TABLE 3 -
Pearson Correlation Coefficients for Step Counting Methods during One Waking Day.
|
Ankle |
Nondominant Wrist |
Hip |
Thigh |
|
|
SW |
FA |
GV |
ApW |
AG |
AGL |
AGM |
HJ325 |
FZ |
SW200 |
AG |
AGL |
AGM |
AP |
| Ankle |
SW |
1 |
0.861 |
0.866 |
0.88 |
0.804 |
0.75 |
0.791 |
0.787 |
0.902 |
0.87 |
0.868 |
0.93 |
0.829 |
0.91 |
| Nondominant wrist |
FA |
|
1 |
– |
– |
0.781 |
0.701 |
0.915 |
0.902 |
0.881 |
0.97 |
0.903 |
0.824 |
0.929 |
0.909 |
| GV |
|
|
1 |
0.963 |
– |
– |
– |
0.974 |
0.979 |
0.941 |
0.991 |
0.991 |
0.986 |
0.925 |
| ApW |
|
|
|
1 |
– |
– |
– |
0.965 |
0.971 |
0.906 |
0.989 |
0.999 |
0.981 |
0.913 |
| AG |
|
|
|
|
1 |
0.95 |
0.629 |
0.98 |
0.818 |
0.721 |
0.741 |
0.866 |
0.631 |
0.897 |
| AGL |
|
|
|
|
|
1 |
0.552 |
0.668 |
0.563 |
0.506 |
0.676 |
0.858 |
0.57 |
0.806 |
| AGM |
|
|
|
|
|
|
1 |
0.951 |
0.843 |
0.951 |
0.93 |
0.742 |
0.903 |
0.832 |
| Hip |
HJ325 |
|
|
|
|
|
|
|
1 |
0.984 |
0.971 |
0.865 |
0.712 |
0.977 |
0.891 |
| FZ |
|
|
|
|
|
|
|
|
1 |
0.959 |
0.982 |
0.938 |
0.966 |
0.985 |
| SW200 |
|
|
|
|
|
|
|
|
|
1 |
0.971 |
0.88 |
0.993 |
0.885 |
| AG |
|
|
|
|
|
|
|
|
|
|
1 |
0.866 |
0.924 |
0.883 |
| AGL |
|
|
|
|
|
|
|
|
|
|
|
1 |
0.799 |
0.912 |
| AGM |
|
|
|
|
|
|
|
|
|
|
|
|
1 |
0.822 |
| Thigh |
AP |
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
AGM, ActiGraph GT9X MAVM; SW200, Yamax Digiwalker SW-200.
All monitor step estimates were significantly different from SW (P < 0.05). Steps were overestimated using AG (wrist) and AGL (wrist and hip), ranging from 114% to 189% of SW steps. All other monitors and step count methods underestimated steps, ranging from 67% to 91% of SW steps. Consumer hip-worn monitors yielded 72%–86% of SW steps, whereas consumer wrist-worn monitors yielded 82%–91% of SW steps. Research monitors (and their accompanying algorithms) displayed wide variability in step counts and appeared to be influenced by several factors, including placement site, digital band-pass filter, and step counting algorithm used to analyze acceleration data.
Corrected step estimates were significantly different from the original steps (P < 0.04). However, corrected step estimates within each monitor and step counting method were not significantly different among correction factor models (P > 0.05) (Fig. 1). Thus, only the simple linear regression correction factors are presented (Table 4), but graphical representations of the other correction factor models’ performance (i.e., mean difference, RMSE, and MAPE) can be found in the Supplementary Material Figures 1–3, https://links.lww.com/TJACSM/A133. Modified Bland–Altman plots show that the correction methods decreased the group level error (mean difference) but did not impact individual error (Fig. 2).
;)
Figure 1: Mean difference, MAPE, and RMSE before and after applying specific correction factor method. AGM, ActiGraph GT9X MAVM; SW200, Yamax Digiwalker SW200.
TABLE 4 -
Correction Method for Step Count Methods to SW From Linear Regression.
| Step Counting Method |
Wear Location |
Correction Method Regression Equation |
| FA |
Wrist |
0.825 × (FA steps per day) + 2439 |
| GA |
1.025 × (GV steps per day) + 1478 |
| ApW |
1.014 × (ApW steps per day) + 1013 |
| AG |
0.853 × (AG steps per day) + 271 |
| AGL |
0.531 × (AGL steps per day) – 30 |
| AGM |
0.982 × (AGM steps per day) + 3570 |
| HJ325 |
Hip |
1.023 × (HJ325 steps per day) + 2471 |
| FZ |
1 × (FZ steps per day) + 1863 |
| SW200 |
0.879 × (SW200 steps per day) + 2310 |
| AG |
1.077 × (AG steps per day) + 1683 |
| AGL |
0.897 × (AGL steps per day) – 1497 |
| AGM |
0.828 × (AGM steps per day) + 3418 |
| AP |
Thigh |
0.987 × (AP steps per day) + 1676 |
AGM, ActiGraph GT9X MAVM; SW200, Yamax Digiwalker SW-200.
Figure 2: Modified Bland–Altman plots depicting differences in total daily steps between observed and specific correction factor method. AGM, ActiGraph GT9X MAVM; SW200, Yamax Digiwalker SW200. Solid line represents mean differences, and dotted lines represent 95% prediction intervals (mean differences ± 1.96 times the SD of the differences).
DISCUSSION
The SW has been used as a criterion measure in previous research to evaluate monitor step count accuracy under free-living conditions. However, these studies were limited primarily to research monitors (i.e., ActiGraph GT3X, AP) and older consumer monitors (i.e., numerous models of the YX and OM) (11–16). One study validated newer consumer monitors against hand-counted steps, but it only included one wrist-worn monitor (i.e., Fitbit Charge) (10). It is important to include more consumer wrist monitors because they are a major component of the wearable industry market and are now frequently being used in behavioral intervention studies, clinical trials, and epidemiological research (2,19).
Out of the 13 monitors (and corresponding step count algorithms) studied, the FA produced step estimates within 10% of SW steps. Fitbit is the most popular consumer activity monitor and has more than 70% of the market share in fitness trackers (20). In addition, Fitbit monitors are being used more in NIH-funded research cohort studies and clinical trials (1,2). Despite capturing more than 90% of SW steps, it still yielded MAPE of 14.9%, showing that there is some individual variability in the step estimates.
The current study supports the findings of previous laboratory-based and free-living research, where ActiGraph step counting methods display a wide range of results (10,12,13). The current study found that the AGL on both wear locations overestimated steps. These large overestimations in steps that result from enabling the low-frequency extension (LFE) in the ActiGraph GT3X have been documented in previous step count validation studies (15,16). The LFE was originally designed to make the accelerometer counts of the MEMS accelerometer in the newer ActiGraph models (GT1M, GT3X, and GT9X) align with those of the older, analog ActiGraph 7164 (21). However, an unforeseen consequence of widening the band-pass filter to allow for lower-frequency accelerations to “pass through” was that it reduced the attenuation of low-frequency signals, resulting in a large overestimation of steps (22). One possibility is that the AGL algorithm may be recording extraneous, nonambulatory movements as steps. Currently, only one study has examined the most recent ActiGraph accelerometer model (GT9X) and found similar results to the present study (10). In our study, there was an underestimation of steps with the normal filter (77% of SW steps) but an overestimation of steps with the LFE filter (129% of SW steps) on the hip. On the wrist, both the normal filter and the LFE filter resulted in an overestimation of steps (114% and 189% of SW steps, respectively). Our results are consistent with previous findings of large overestimations when enabling the LFE filter (23,24).
The current study also processed raw GT9X data from monitors worn on the nondominant wrist and hip with the beta version of ActiGraph’s MAVM step algorithm. ActiGraph GT9X MAVM on the hip and wrist recorded 81% and 67% of SW steps, respectively. Previously, we assessed the step accuracy of the MAVM step algorithm across 1 d on both wrists and the hip. Interestingly, the findings from their study showed that MAVM steps were within 10% of SW steps for both wear locations (10). Differences in the implementation of the MAVM algorithm may have contributed to the inconsistencies between the current study and the aforementioned study. In the current study, rather than postprocessing the data after downloading, MAVM step counts were obtained directly off the device screen, similar to how steps were acquired from consumer monitors (the steps displayed on the screen are based on the MAVM algorithm; Wyatt, J., ActiGraph Chief Executive Officer, 2018, personal communications, 7 May).
The other consumer monitors displayed values that were 72% to 91% of SW steps. Findings of previous studies that used the SW as a criterion have found a similar magnitude of undercounting in the SW200 and HJ325 (11,13,14,16), which could be partially attributable to their underestimations of steps observed during slow walking and activities of daily living (16,25,26). The newer consumer monitors (i.e., ApW, GV, and FZ) yielded 88%, 83%, and 80% of SW steps, respectively, with similar MAPE (19%–22%). Although there are limited studies that have validated various models of these activity monitors, previous studies show that faster ambulation speeds increase step count error for the Garmin worn on the wrist, whereas slower speeds were found to increase step count error for the Fitbit worn on the wrist and hip (4,5,27–31). Step count validation studies on the ApW are limited and so far have only been conducted in laboratory settings (32–34). Findings suggest that the ApW displays low MAPE (<5%) over a wide range of ambulation speeds (i.e., 54–107 m⋅min−1), with increasing reliability at faster speeds (ICC = 0.38–0.80) (32). In addition, the ApW displayed less step counting error (MAPE 6%) during aerobic exercise (i.e., walking or jogging) than during light intensity PA (i.e., folding laundry, sweeping, moving light boxes, stretching, and slow walking) (MAPE 161%) (33). Although these studies highlight specific activities that can lead to decreased accuracy in step counting, they do not quantify the magnitude of error across an entire day of wearing the monitor.
Steps accumulated from brief, intermittent walking bouts and sporadic activities of daily living may be a source of error attributing to the underestimation of steps in the current study. A previous study conducted in our laboratory assessed step count accuracy during brief, intermittent walking bouts and found that for the FZ, steps taken in bouts that were limited to only six steps (or approximately 3 s) did not get added to the total step count (35). Most consumer monitors store the steps in a temporary cache and require several seconds of continuous ambulation before those steps are added to the aggregate total step count. If a person stops walking before that period, the steps will not be added to the aggregate total. In addition, ActiGraph’s MAVM algorithm requires more than 2 s of continuous walking when worn on the hip, and 10 s of continuous stepping when worn on the wrist, in order for steps to be recorded (Wyatt, J. 2017, personal communications, 3 May) (35). Thus, one of the reasons that many step count methods underestimate the SW is that the SW is designed to capture each step, whereas other methods only count steps taken in continuous stepping bouts of a certain minimum duration.
The ankle-worn SW records a step every time the leg it is attached to is “unweighted,” and the foot is moved to a new location. The foot does not need to be lifted totally off the ground, and even slow, shuffling (or sliding) movements are recorded as steps. The SW also counts only the steps taken with one leg and then multiplies this number by two to estimate the steps taken by both legs. Because of the location of the SW on the ankle, it has a high “specificity” for step counting. The wrist monitors appear to use rhythmic accelerations/decelerations of the wrist to count steps. Because of the fact that accelerations/decelerations of the wrist occur even when steps are not being taken, most wrist monitors only record steps in response to repeated, rhythmic accelerations (35). This is necessary to increase the specificity for step counting with wrist monitors, and even some waist-worn monitors use this approach. As a result, most consumer monitors yield lower step counts than the SW.
The use of wearable monitors to assess steps per day using monitors has increased within the general population, research settings, and clinical trials. The benefits and limitations of using research- and consumer-grade monitors are well documented (1,2,7,36) and should be considered by researchers and clinical trials according to the purpose of their study. Briefly, consumer monitors (e.g., Fitbit, Garmin, Omron, and Yamax) are generally lower cost, have greater participant acceptance and compliance, have acceptable accuracy for steps, and have an interactive display allowing for self-monitoring (and social media platforms). However, because the companies that manufacture them have proprietary algorithms, step count accuracy could change over time from routine firmware updates. In addition, the interactive display could be bad for surveillance (due to “reactivity”). Research monitors (e.g., ActiGraph GT series, StepWatch, and activPAL) can store raw data, yield estimates of MVPA, and assess “wear time.” Downsides of using research monitors are the generally higher cost for the device and data management platform, the plethora of algorithms to process data, no agreed upon reference standard for validation, MVPA estimates that vary greatly, and lack of interactive display (in some models), which does not allow for self-monitoring.
The correction method established in the current study for each monitor may help to achieve comparability by converting steps per day from each monitor to that of SW steps per day, but this should be further investigated in an independent sample. In addition, the validity of the developed correction factor regressions is dependent on several factors, such as behavioral and movement patterns, and updates to device firmware that may influence the underlying step count algorithm.
A limitation of our correction method is that it does not account for the behavioral composition of how steps are accumulated. Generally, laboratory-based studies have shown that most monitors have greater step count accuracy when worn during continuous walking compared with intermittent walking bouts or activities of daily living (17,35). As a result, the magnitude of monitor-based step count error may differ depending on how much time spent in various activity types. Future studies should investigate the validity of the correction factors in individuals with varying compositions of behavioral patterns.
The current study provides support for the use of consumer monitors in healthy adult populations, although more research needs to be conducted in clinical populations that may have irregular gait patterns and/or slower walking speeds. Our findings add to the step count validation literature by assessing the accuracy of popular consumer and research monitors across all waking hours of 1 d, under free-living conditions, and provide correction methods for the comparison of step counts from several monitors. This is important because steps per day is a variable of great interest to researchers, and activity monitors are being widely used by researchers, clinicians, and the public.
DB was a member of the scientific advisory board of ActiGraph, LLC, within the 3-year window preceding publication of the manuscript, but he has not served on the board since 2019. Remaining authors 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.
REFERENCES
1. Wright SP, Hall Brown TS, Collier SR, Sandberg K. How consumer physical activity monitors could transform human physiology research.
Am J Physiol Regul Integr Comp Physiol. 2017;312(3):R358–r67. Epub 2017/01/06. doi: 10.1152/ajpregu.00349.2016. PubMed PMID: 28052867; PubMed Central PMCID: PMCPMC5401997.
2. Strath SJ, Rowley TW. Wearables for promoting physical activity.
Clin Chem. 2018;64(1):53–63. Epub 2017/11/10. doi: 10.1373/clinchem.2017.272369. PubMed PMID: 29118062.
3. Chow JJ, Thom JM, Wewege MA, Ward RE, Parmenter BJ. Accuracy of step count measured by physical activity monitors: the effect of gait speed and anatomical placement site.
Gait Posture. 2017;57:199–203. Epub 2017/07/01. doi: 10.1016/j.gaitpost.2017.06.012. PubMed PMID: 28666177.
4. Chen MD, Kuo CC, Pellegrini CA, Hsu MJ. Accuracy of wristband activity monitors during ambulation and activities.
Med Sci Sports Exerc. 2016;48(10):1942–9. Epub 2016/05/18. doi: 10.1249/mss.0000000000000984. PubMed PMID: 27183123.
5. An HS, Jones GC, Kang SK, Welk GJ, Lee JM. How valid are wearable physical activity trackers for measuring steps?
Eur J Sport Sci. 2017;17(3):360–8. Epub 2016/12/04. doi: 10.1080/17461391.2016.1255261. PubMed PMID: 27912681.
6. Imboden MT, Nelson MB, Kaminsky LA, Montoye AH. Comparison of four Fitbit and Jawbone activity monitors with a research-grade ActiGraph accelerometer for estimating physical activity and energy expenditure.
Br J Sports Med. 2017. Epub 2017/05/10. doi: 10.1136/bjsports-2016-096990. PubMed PMID: 28483930.
7. Bassett DR Jr., Toth LP, LaMunion SR, Crouter SE. Step counting: a review of measurement considerations and health-related applications.
Sports Med. 2017;47(7):1303–15. doi: 10.1007/s40279-016-0663−1. PubMed PMID: 28005190; PubMed Central PMCID: PMCPMC5488109.
8. Wuerzner G, Bochud M, Zweiacker C, Tremblay S, Pruijm M, Burnier M. Step count is associated with lower nighttime systolic blood pressure and increased dipping.
Am J Hypertens. 2013;26(4):527–34. Epub 2013/03/08. doi: 10.1093/ajh/hps094. PubMed PMID: 23467208.
9. Zajac-Gawlak I, Pospiech D, Kroemeke A, et al. Physical activity, body composition and general health status of physically active students of the University of the Third age (U3A).
Arch Gerontol Geriatr. 2016;64:66–74. Epub 2016/03/10. doi: 10.1016/j.archger.2016.01.008. PubMed PMID: 26952379.
10. Toth LP, Park S, Springer CM, Feyerabend MD, Steeves JA, Bassett DR. Video-recorded validation of wearable step counters under free-living conditions.
Med Sci Sports Exerc. 2018;50:1315–22. Epub 2018/01/31. doi: 10.1249/mss.0000000000001569. PubMed PMID: 29381649.
11. Karabulut M, Crouter SE, Bassett DR Jr. Comparison of two waist-mounted and two ankle-mounted electronic pedometers.
Eur J Appl Physiol. 2005;95(4):335–43. Epub 2005/09/01. doi: 10.1007/s00421-005-0018-3. PubMed PMID: 16132120.
12. Tyo BM, Fitzhugh EC, Bassett DR Jr., John D, Feito Y, Thompson DL. Effects of body mass index and step rate on pedometer error in a free-living environment.
Med Sci Sports Exerc. 2011;43(2):350–6. Epub 2010/06/15. doi: 10.1249/MSS.0b013e3181e9b133. PubMed PMID: 20543755.
13. 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. 2011;43(9):1791–7.
14. Feito Y, Bassett DR, Thompson DL. Evaluation of activity monitors in controlled and free-living environments.
Med Sci Sports Exerc. 2012;44(4):733–41. Epub 2011/09/10. doi: 10.1249/MSS.0b013e3182351913. PubMed PMID: 21904249.
15. Feito Y, Garner HR, Bassett DR. Evaluation of ActiGraph's low-frequency filter in laboratory and free-living environments.
Med Sci Sports Exerc. 2015;47(1):211–7. Epub 2014/05/30. doi: 10.1249/mss.0000000000000395. PubMed PMID: 24870583.
16. Hickey A, John D, Sasaki JE, Mavilia M, Freedson P. Validity of activity monitor step detection is related to movement patterns.
J Phys Act Health. 2016;13(2):145–53. Epub 2015/06/25. doi: 10.1123/jpah.2015-0203. PubMed PMID: 26107045.
17. Toth L, Park S, Pittman W, et al. Validity of activity tracker step counts during walking, running, and activities of daily living.
Trans J Am Coll Sports Med. 2018;3(7):52–9.
18. Park S, Toth LP, Crouter SE, Springer CM, Marcotte RT, Bassett DR. Effect of monitor placement on the daily step counts of wrist and hip activity monitors.
Journal for the Measurement of Physical Behaviour. 2020;3(2):164–9.
19. Statista. Wearables Sales Revenue Worldwide by Category in 2015, 2018 and 2021 (in Billion U.S. dollars): Statista; 2018 [cited 2018]. Available from:
https://www.statista.com/statistics/641865/wearables-sales-by-category-worldwide/.
20. Rodgers A. How Fitbit's CEO Sees a Future in the Medical Industry: TIME; 2015 [cited 2017]. Available from:
http://time.com/3751693/fitbit-ceo-medical-industry/.
21. Ried-Larsen M, Brønd JC, Brage S, et al. Mechanical and free living comparisons of four generations of the ActiGraph activity monitor.
Int J Behav Nutr Phys Act. 2012;9:113. doi: 10.1186/1479-5868-9-113. PubMed PMID: 22971175.
22. Cain KL, Conway TL, Adams MA, Husak LE, Sallis JF. Comparison of older and newer generations of ActiGraph accelerometers with the normal filter and the low frequency extension.
Int J Behav Nutr Phys Act. 2013;10:51. Epub 2013/04/27. doi: 10.1186/1479-5868-10-51. PubMed PMID: 23618461; PubMed Central PMCID: PMCPMC3641979.
23. Tudor-Locke C, Barreira T, Schuna J. Comparison of step outputs for waist and wrist accelerometer attachment sites.
Med Sci Sports Exerc. 2015;47(4):839–42.
24. Feito Y, Hornbuckle LM, Reid LA, Crouter SE. Effect of ActiGraph’s low frequency extension for estimating steps and physical activity intensity.
PLoS One. 2017;12(11):e0188242. doi: 10.1371/journal.pone.0188242.
25. Crouter SE, Schneider PL, Karabulut M, Bassett DR Jr. Validity of 10 electronic pedometers for measuring steps, distance, and energy cost.
Med Sci Sports Exerc. 2003;35(8):1455–60. Epub 2003/08/06. doi: 10.1249/01.mss.0000078932.61440.a2. PubMed PMID: 12900704.
26. Le Masurier GC, Tudor-Locke C. Comparison of pedometer and accelerometer accuracy under controlled conditions.
Med Sci Sports Exerc. 2003;35(5):867–71. Epub 2003/05/17. doi: 10.1249/01.mss.0000064996.63632.10. PubMed PMID: 12750599.
27. Leth S, Hansen J, Nielsen OW, Dinesen B. Evaluation of commercial self-monitoring devices for clinical purposes: results from the future patient trial, phase I.
Sensors (Basel). 2017;17(1):211. Epub 2017/01/25. doi: 10.3390/s17010211. PubMed PMID: 28117736; PubMed Central PMCID: PMCPMC5298782.
28. O'Connell S, ÓLaighin G, Kelly L, et al. These shoes are made for walking: sensitivity performance evaluation of commercial activity monitors under the expected conditions and circumstances required to achieve the international daily step goal of 10,000 steps.
PLoS One. 2016;11(5):e0154956. Epub 2016/05/12. doi: 10.1371/journal.pone.0154956. PubMed PMID: 27167121; PubMed Central PMCID: PMCPMC4864247.
29. Alinia P, Cain C, Fallahzadeh R, Shahrokni A, Cook D, Ghasemzadeh H. How Accurate is your activity tracker? A comparative study of step counts in low-intensity physical activities.
JMIR Mhealth Uhealth. 2017;5(8):e106. Epub 2017/08/13. doi: 10.2196/mhealth.6321. PubMed PMID: 28801304; PubMed Central PMCID: PMCPMC5572056.
30. Huang Y, Xu J, Yu B, Shull PB. Validity of Fitbit, Jawbone UP, Nike+ and other wearable devices for level and stair walking.
Gait Posture. 2016;48:36–41. Epub 2016/08/02. doi: 10.1016/j.gaitpost.2016.04.025. PubMed PMID: 27477705.
31. Tully MA, McBride C, Heron L, Hunter RF. The validation of Fibit Zip physical activity monitor as a measure of free-living physical activity.
BMC Res Notes. 2014;7:952. Epub 2014/12/30. doi: 10.1186/1756-0500-7-952. PubMed PMID: 25539733; PubMed Central PMCID: PMCPMC4307145.
32. 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. 2017;49(4):793–800. Epub 2017/03/21. doi: 10.1249/mss.0000000000001146. PubMed PMID: 28319983.
33. Bai Y, Hibbing P, Mantis C, Welk GJ. Comparative evaluation of heart rate-based monitors: Apple Watch vs Fitbit Charge HR.
J Sports Sci. 2018;36:1734–41. Epub 2017/12/07. doi: 10.1080/02640414.2017.1412235. PubMed PMID: 29210326.
34. Modave F, Guo Y, Bian J, et al. Mobile device accuracy for step counting across age groups.
JMIR Mhealth Uhealth. 2017;5(6):e88. Epub 2017/07/01. doi: 10.2196/mhealth.7870. PubMed PMID: 28659255; PubMed Central PMCID: PMCPMC5508112.
35. Toth L, Park S, Pittman W, Sarasaltik D, Morton A, Bassett D. Step count filters in wearable step counters.
Med Sci Sports Exerc. 2017;49(5S):366.
36. Bassett DR, Freedson PS, John D. Wearable activity trackers in clinical research and practice.
Kinesiology Review. 2019;8(1):11–5.
