Secondary Logo

Journal Logo

Accelerometer-assessed Physical Activity in Epidemiology: Are Monitors Equivalent?

ROWLANDS, ALEX, V.1,2,3; MIRKES, EVGENY, M.4; YATES, TOM1,2; CLEMES, STACEY5; DAVIES, MELANIE1,2; KHUNTI, KAMLESH1,2,6; EDWARDSON, CHARLOTTE, L.1,2

Medicine & Science in Sports & Exercise: February 2018 - Volume 50 - Issue 2 - p 257–265
doi: 10.1249/MSS.0000000000001435
EPIDEMIOLOGY
Free

Purpose Accelerometers are increasingly being used to assess physical activity in large-scale surveys. Establishing whether key physical activity outcomes can be considered equivalent between three widely used accelerometer brands would be a significant step toward capitalizing on the increasing availability of accelerometry data for epidemiological research.

Methods Twenty participants wore a GENEActiv, an Axivity AX3, and an ActiGraph GT9X on their nondominant wrist and were observed for 2 h in a simulated living space. Participants undertook a series of seated and upright light/active behaviors at their own pace. All accelerometer data were processed identically using open-source software (GGIR) to generate physical activity outcomes (including average dynamic acceleration (ACC) and time within intensity cut points). Data were analyzed using pairwise 95% equivalence tests (±10% equivalence zone), intraclass correlation coefficients (ICC) and limits of agreement.

Results The GENEActiv and Axivity could be considered equivalent for ACC (ICC = 0.95, 95% confidence interval (CI), 0.87–0.98), but ACC measured by the ActiGraph was approximately 10% lower (GENEActiv/ActiGraph: ICC = 0.86; 95% CI, 0.56–0.95; Axivity/ActiGraph: ICC = 0.82; 95% CI, 0.50–0.94). For time spent within intensity cut points, all three accelerometers could be considered equivalent to each other for more than 85% of outcomes (ICC ≥0.69, lower 95% CI ≥0.36), with the GENEActiv and Axivity equivalent for 100% of outcomes (ICC ≥0.95, lower 95% CI ≥0.86).

Conclusions GENEActiv and Axivity data processed in GGIR are largely equivalent. If GENEActiv or Axivity is compared with the ActiGraph, time spent within intensity cut points has good agreement. These findings can be used to inform selection of appropriate outcomes if outputs from these accelerometer brands are compared.

1Diabetes Research Centre, University of Leicester, Leicester General Hospital, Leicester, UNITED KINGDOM; 2NIHR Leicester Biomedical Research Centre, Leicester, UNITED KINGDOM; 3Alliance for Research in Exercise, Nutrition and Activity (ARENA), Sansom Institute for Health Research, Division of Health Sciences, University of South Australia, Adelaide, AUSTRALIA; 4Department of Mathematics, University of Leicester, Leicester, UNITED KINGDOM; 5School of Sport, Exercise and Health Sciences, University of Loughborough, Loughborough, UNITED KINGDOM; and 6NIHR Collaboration for Leadership in Applied Health Research and Care East Midlands, Leicester General Hospital, Leicester, UNITED KINGDOM

Address for correspondence: Alex V. Rowlands, Ph.D., Diabetes Research Centre, University of Leicester, Leicester General Hospital, Leicester, LE5 4PW, United Kingdom; E-mail: alex.rowlands@le.ac.uk.

Submitted for publication July 2017.

Accepted for publication September 2017.

Pooling data from multiple surveys has facilitated robust and generalizable estimates of risk factors (e.g., smoking and overweight) for cardiovascular events, cardiovascular disease, and mortality that have informed clinical and public health practice (1,2). Physical inactivity is also an established risk factor for chronic disease (3), but until recently, physical activity measurement in epidemiological and surveillance studies has relied on self-report. This is imprecise, which has complicated comparison or aggregation of data across populations.

Over the past few years, it has become feasible to move to large-scale objective measurement of physical activity with wrist-worn accelerometers worn 24 h·d−1, 7 d·wk−1. Because the latest generation of accelerometers measure acceleration in SI units, there is great potential for aggregation of measures of physical activity into very large multinational databases. Data harmonization would facilitate a step change in our ability to (a) compare prevalence or levels of activity/inactivity across populations, (b) quantify dose–response associations between activity and health, and (c) identify the factors that affect these associations. A key advantage would be the ability to address these questions in very large samples across countries and/or populations for a wide range of health outcomes.

There are three brands of accelerometers providing acceleration data in SI units being used in large surveys: the Axivity, ActiGraph, and GENEActiv. For example, UK Biobank, a large-scale prospective epidemiological resource containing baseline phenotypic and genotypic data on 500,000 participants, has recently used the Axivity wrist-worn accelerometer in more than 100,000 participants (4), and the Breakthrough Generation Study (5) has used the Axivity on 4800 women to date. The US National Health and Nutrition Examination Survey used the ActiGraph wrist-worn accelerometer in cycles 2011–2012 and 2013–2014 (approximately 9000–10,000 participants examined per cycle). The Pelotas Birth Cohort (6), the Melbourne Child Health Checkpoint (7), the Cork Children’s Lifestyle Study (8), and the British Whitehall II study (9) used the GENEActiv in approximately 10,000, 4000, 1000, and 3750 participants, respectively. Before outcomes from studies using different brands of accelerometer can be pooled into multinational databases, it is necessary to demonstrate comparability of data outputs between brands.

Pooling accelerometer data from these studies is more viable now than ever before. Earlier accelerometers processed data into counts using proprietary algorithms (10); this complicated the interpretation and comparison of data from studies using different devices (11). However, evidence suggests that, despite the outcome being nonproprietary accelerations, data may not be equivalent between brands (12–16).

To facilitate transparent processing of these raw data, the generation and use of open-source resources is encouraged (16,17). GGIR is an open-access package in R (http:/cran.r-project.org) that can be used to process and analyze raw accelerations from the GENEActiv and the Actigraph using identical methods (17–20). Because it is open source and an efficient method for processing and analyzing raw data to obtain the key outcomes required for characterizing habitual physical activity, it has been used widely to analyze GENEActiv and ActiGraph data (e.g., Refs. [6,9,20–26]). We have shown that the outcomes from GENEActiv and ActiGraph processed through GGIR are broadly comparable, although comparisons at lower magnitudes of acceleration may be problematic (20). More focused assessment of specific sedentary and light activities is needed to examine this further.

The Axivity and GENEActiv measured similar magnitudes of accelerations when tested in a mechanical shaker (27), but there are no data comparing the Axivity and ActiGraph or comparing physical activity outcomes from the Axivity to the GENEActiv or ActiGraph during actual wear. This article will introduce a function that converts raw Axivity files to a format that facilitates identical processing and analysis of Axivity files in GGIR. It is important to consider the accelerometer, processing, and analysis together when establishing whether outcomes can be considered equivalent or not, because each of these steps can affect the final outcome variable (16). Given the widespread use of (a) these accelerometers to assess physical activity and (b) GGIR for processing and analysis of the data, establishing which outcomes can be considered equivalent between accelerometers would be a significant step toward capitalizing on the increasing availability of accelerometry data in epidemiological research.

We aimed to establish whether the Axivity, GENEActiv, and ActiGraph GT9X result in equivalent physical activity outcomes when data are processed and analyzed identically with GGIR. We considered the magnitude of acceleration, time spent above and below published intensity cut points, and the distribution of time across an incremental acceleration range, both for specific activities and for the duration of a simulated free-living situation in a laboratory.

Back to Top | Article Outline

METHODS

A convenience sample of 20 adult participants was recruited from Loughborough University and University of Leicester (staff and students) via e-mail and word of mouth. All participants provided written informed consent, and the study was approved by the Ethics Committee of Loughborough University. Data were collected between January and April 2016.

Height and body mass were measured to the nearest 0.5 cm and 0.1 kg, respectively. Each participant was fitted with an Axivity AX3 (Axivity Ltd, Newcastle, United Kingdom), GENEActiv (ActivInsights Ltd, Cambridgeshire, United Kingdom), and ActiGraph GT9X Link (ActiGraph LLC, Pensacola, FL) on their nondominant wrist. This was part of a larger study using further activity monitors; to reduce the need for multiple wrist-straps, the Axivity was taped to the GENEActiv and the ActiGraph was worn immediately proximal to the GENEActiv and Axivity. In our previous studies comparing output from the GENEActiv and the ActiGraph, differences in output were consistent whether the monitors were taped together (12,28) or worn adjacent on the wrist (20).

Back to Top | Article Outline

Protocol

The study took place in a laboratory mocked up as a living space with items of furniture and lasted approximately 2 h with participants tested in groups of two or three. Participants were asked to undertake a series of seated activities (watching television, using the computer, eating, and reading) in any manner, at their own pace and in any order they chose. Minimal instructions were given with participants simply asked to ensure they undertook each activity at least once and for a minimum of 10 min. The aim was to mimic free-living postures/behaviors as closely as possible. A researcher observed the participants continuously and recorded their activity and posture (seated or standing) minute-by-minute. Participants then performed six upright light and active behaviors in a randomized order for 5 min each: standing still, standing up to work on a computer, dusting, sweeping, washing pots, and walking.

Back to Top | Article Outline

Accelerometers

The Axivity AX3, GENEActiv, and ActiGraph GT9X Link (from herein: Axivity, GENEActiv, and ActiGraph) are triaxial accelerometry–based activity monitors with a dynamic range of ±8g, where g is equal to the Earth’s gravitational pull. All accelerometers were set to capture and store accelerations at their maximum sampling frequency of 100 Hz. The “idle sleep mode” in the ActiGraph software (Actilife v. 6.13.0) was disabled. Axivity data were downloaded using OmGui open-source software (OmGui Version 1.0.0.28; Open Movement, Newcastle University, Newcastle upon Tyne, United Kingdom) and saved in raw format as .cwa files. GENEActiv data were downloaded using GENEActiv PC software version 2.2 and saved in raw format as.bin files. ActiGraph data were downloaded using ActiLife v. 6.13.0, saved in raw format as .gt3x files, and converted to .csv format for data processing.

Back to Top | Article Outline

Data processing and outcome measures

Raw cwa Axivity files were converted to .bin files using our new function, “AccelerometerCWA2BINConverter,” to enable analysis in GGIR, thus enabling identical processing and analysis as GENEActiv and ActiGraph files. This function includes resampling of the data to a standard frequency as specified in the header of the cwa file; this is necessary because the Axivity sample frequency is unreliable and varies over time. This function is available at: https://github.com/Mirkes/AccelerometerCWA2BINConverter.

All accelerometer files were analyzed using R-package GGIR version 1.4 in R (http://cran.r-project.org) (17,19). This included autocalibration using local gravity as a reference (19), detection of sustained abnormally high values, calculation of the average magnitude of dynamic acceleration (i.e., resultant vector magnitude, corrected for gravity and expressed as Euclidean Norm Minus One (ENMO) in milligravitational units (mg) averaged over 1-s epochs) and generation of participant-specific csv files with accelerometer output in 1-s epochs. Where insufficient nonmovement periods were available for autocalibration, we used back-up calibration coefficients derived from free-living data collected with the same accelerometer unit.

A number of prespecified outcomes were assessed: average acceleration (mg); % time accumulated within cut points for sedentary, light, and moderate-to-vigorous physical activity (MVPA); distribution of time across acceleration levels in 40-mg resolution (0–40, 40–80,…, >200 mg). Mean acceleration was calculated for each activity separately, seated activities only, upright activities only, and over the total time period (i.e., including all activities and transitions, approximately 2 h). Cut point and distribution of time outcomes were calculated for seated activities only, upright activities only, and over the total time period.

Time spent sedentary and time in light activity were calculated using cut points of 30, 40, and 50 mg to enable evaluation of the equivalency of a range of sedentary cut points (14,15). The accuracy of these cut points was further calculated for seated activity (i.e., percent seated time classified as sedentary) and for upright activity (i.e., percent upright time classified as not sedentary). MVPA was calculated using an acceleration cut point of 100 mg (13). All outcomes were calculated for all three devices.

Back to Top | Article Outline

Data analysis

Descriptive statistics (mean ± SD) were calculated for all outcomes. The level of agreement between outputs from the three brands of accelerometer was determined pairwise using intraclass correlation coefficients (ICC; single measures, absolute agreement) with 95% confidence intervals (CIs) and limits of agreement (LoA) (29). We used pairwise 95% equivalence tests to determine whether the 95% CI for the mean of one accelerometer fell within a proposed equivalence zone of the second accelerometer (30). We selected ±10% of the mean as our proposed equivalence zone as in previous studies comparing activity monitors (31,32). Equivalence results are presented with the reference accelerometer selected according to the following hierarchy: GENEActiv, Axivity, and ActiGraph. However, because no accelerometer can be considered the gold standard, this was arbitrary; consequently, all equivalence analyses were repeated with the alternate accelerometer in each pairing selected as the reference to test whether this affected the conclusions.

On the basis of our previous work comparing GGIR physical activity outputs from the GENEActiv and the ActiGraph (20), we anticipated that the SD of the differences between the log transformed outputs from the two accelerometers would be less than 0.05 mg and the ratio between the mean outputs from the two accelerometers would be within 1 ± 0.05. Log transforming the data enables hypotheses about ratios to be analyzed in terms of differences. Given this effect size, using Minitab (v17), we determined that a sample size of 12 was required to provide 90% power (alpha = 0.05) to conclude that the difference between physical activity outcomes from a pair of accelerometers was within 10% of the mean when this was in fact true.

Descriptive statistics, ICC values, and LoA were conducted in IBM SPSS Statistics v22.0. Equivalency testing and power analyses were carried out in Minitab (v17). Alpha was set at 0.05.

Back to Top | Article Outline

RESULTS

Twenty participants (13 women, 7 men; mean ± SD (age), 23.2 ± 5.9 yr; body mass index, 25.2 ± 3.6 kg·m−2) took part. The Axivity accelerometers were unavailable for one testing session (three participants), one GENEActiv file and two ActiGraph files did not process, and one participant did not complete the seated activities. Therefore, the sample included 17 Axivity, 19 GENEActiv, and 18 ActiGraph files for the overall time period and upright activities and 16, 18, and 18, respectively, for seated activities. Pairwise N’s were 16 for GENEActiv/Axivity, 17 for GENEActiv/ActiGraph, and 15 for Axivity/ActiGraph, exceeding the sample size of 12 required to achieve 90% power. Participant characteristics were similar for included and excluded files. Running the analyses with listwise deletion did not change the results (N = 14), so pairwise analyses were retained to maximize sample sizes.

The total testing period lasted approximately 2 h and included 1 h 20 min of seated activities, 24 min of upright activities, and 16 min of transition between activities. Descriptive statistics are presented in Table 1 and agreement statistics in Table 2.

TABLE 1

TABLE 1

TABLE 2

TABLE 2

Back to Top | Article Outline

Acceleration (mg)

Agreement between pairs of accelerometers was largely good, with ICC values of 0.82–0.95 for the total period, 0.73–0.85 for all seated activities combined (except the Axivity/ActiGraph pairing, ICC = 0.59), and 0.75–0.97 all upright activities combined (Table 2). Results for specific activities suggested that the poorest agreement between pairs of accelerometers was obtained when using the computer (sitting or standing) and reading. Although the mean biases between pairs of accelerometers tended to be low, some of the 95% LoA were relatively large, particularly for the Axivity/ActiGraph pairing.

Overall, the highest agreement was between the GENEActiv and Axivity devices (ICC = 0.95; 95% CI, 0.87–0.98; mean bias −0.1 mg 95% limits, ±6.8 mg), which could be considered equivalent (i.e., the 95% CI for the mean of the Axivity fell within ±10% of the mean of the GENEActiv) for the total period (Fig. 1A), upright activities combined (Fig. 1A), and 4 of the 10 specific activities (not for seated activities, standing still, or standing computer; Fig. 1B, C). In contrast, the ActiGraph could not be considered equivalent to either the GENEActiv or the Axivity at all (Fig. 1B, C).

FIGURE 1

FIGURE 1

Back to Top | Article Outline

Sedentary, light intensity, and MVPA cut points

Agreement between accelerometers was good for classification of sedentary time (ICC ≥0.84, mean bias ≤±1.5 percentage points) for all accelerometer pairings, irrespective of cut point, and all could be considered equivalent (Table 3, Fig. 1D). The 95% LoA values were narrowest for the 50-mg point (Table 3). All accelerometers could also be considered equivalent for classification of light-intensity activity (Fig. 1D), with pairings weakest for the 30-mg cut point (ICC ≥0.69). As for classification as sedentary, the highest ICC values (≥0.83), lowest mean biases (≤±0.3 percentage points), and narrowest 95% LoA were found for the 50-mg point. Although ICC values for classification of MVPA were high (≥0.84), LoA values were large (±2.3 to 4.1 percentage points) relative to the means (approx. 13%), and only the GENEActiv/Axivity pairing could be considered equivalent (Fig. 1D).

TABLE 3

TABLE 3

Although all accelerometer pairings could be considered equivalent for six of seven outcomes, the highest agreement was found for the GENEActiv and Axivity device, irrespective of cut point (ICC >0.95, mean bias ≤±0.4 percentage points, 95% limits <±5 percentage points).

Back to Top | Article Outline

Accuracy of sedentary cut points

Accuracy of classification of seated activities as sedentary was high for all cut points (>87%; Table 1) and equivalent between accelerometers (Table 4, Fig. 1E). However, the upright activities were misclassified approximately 60% of the time with the 30-mg point and only equivalent for the GENEActiv/Axivity pairing. When applying the 40-mg or 50-mg point, the upright activities were still misclassified approximately 50% of the time, but all accelerometers could be considered to have equivalent accuracy.

TABLE 4

TABLE 4

Again, the highest agreement was found for the GENEActiv and Axivity device, irrespective of cut point (ICC >0.83, mean bias ≤±1.3 percentage points, 95% limits ≤±5.5 percentage points).

Back to Top | Article Outline

Incremental acceleration ranges

Agreement was high for all accelerometer pairings for percent time in 40-mg increments (ICC ≥0.71, mean bias <±1.6 percentage points). As with average mg and time spent in cut point categories, the strongest agreement was for the GENEActiv and Axivity pairing (≥0.93, mean bias ≥±0.25 percentage points), which could be considered equivalent for all categories, except 160–200 mg, which was borderline equivalent (Table 5, Fig. 1F). The GENEActiv/ActiGraph and Axivity/ActiGraph pairings could be considered equivalent in the following categories: 0–40 mg, 40–80 mg, and 80–120 mg (Fig. 1F).

TABLE 5

TABLE 5

The highest agreement was found for the GENEActiv and Axivity device, irrespective of 40-mg range (ICC >0.93, mean bias ≤±0.25 percentage points, 95% limits ≤±4.0 percentage points).

Rerunning the equivalence analyses with the alternate accelerometer selected as the reference did not affect whether accelerometer brands were considered equivalent or not, except two cases that were previously borderline but could be considered equivalent when the alternate accelerometer was the reference. These were 1) accuracy of the sedentary cut point, 30 mg, for classification as upright by the ActiGraph and the Axivity (ratio, 0.93 (95% CI, 0.90–1)), and 2) time spent between 160 and 200 mg by the Axivity and the GENEActiv (ratio, 0.96, (95% CI, 0.90–1.03); Fig. 1E–F, Tables 2–5).

Back to Top | Article Outline

DISCUSSION

The Axivity, GENEActiv, and ActiGraph wrist-worn raw acceleration accelerometers are widely used to assess physical activity in large-scale surveys (16); consequently, the generation of equivalent physical activity outcomes from these tools would aid epidemiological comparisons. We used an open-source software package (GGIR) to identically process and analyze data from the three accelerometer brands to establish the degree of equivalence and agreement across specific activities, types of activities, and the entire semistructured pseudo–free-living period. To the best of our knowledge, this is the first time physical activity outcomes from the Axivity accelerometer have been compared with those from the GENEActiv or the ActiGraph.

The GENEActiv and Axivity accelerometers had excellent equivalence and agreement across most outcome measures including acceleration (mg). However, acceleration was around 11% lower in the ActiGraph data. Despite this, time spent in sedentary and light intensity could be considered equivalent between all three accelerometers, irrespective of cut point used. The GENEActiv and Axivity could also be considered equivalent for MVPA; agreement with the ActiGraph was also high for MVPA, although not within the proposed 10% equivalence zone. The higher agreement between accelerometer brands evident when considering variables derived from acceleration; that is, time accumulated in cut-point categories or acceleration ranges, rather than the acceleration itself, is consistent with our previous research comparing the GENEActiv and ActiGraph GT3X+ (20). The high correspondence between accelerometer brands for intensity categories is important because quantities of time accumulated within sedentary, light, and moderate-to-vigorous intensity cut points are arguably the most commonly cited physical activity outcome measures.

The observation protocol we used enabled us to comprehensively evaluate the accuracy of three sedentary cut points (14,15). Sedentary/light cut points of 40–50 mg were the most accurate at classifying sedentary and upright time and had higher agreement between accelerometers compared with cut points of 30 mg. However, all cut points, irrespective of accelerometer brand, were poor at classifying upright time. The inability to differentiate between postures using magnitude of acceleration alone has been previously reported (14,15) with use of further features from the acceleration signal recommended for classification of posture (e.g., Refs. [33,34]). When using cut points or the distribution of time across acceleration ranges, it is perhaps best to think of classifying a spectrum of inactive to active time rather than referring to sedentary time, which infers posture (35).

The equivalency results suggest that equivalence is worse at low accelerations (seated activities), for time spent at accelerations greater than 120 mg, and for time spent in MVPA, most notably for the ActiGraph. However, because the proposed equivalence zone is ±10% of the output magnitude, this is most likely a function of the low numbers involved, that is, accelerations of <20 mg and ≈4% of time spent at accelerations of >120 mg. For example, when comparing accelerations during seated activities, the distribution of time across 40-mg increments in acceleration and time spent in MVPA between accelerometer brands, despite sometimes not reaching equivalence, fairly high ICC values, low mean bias, and relatively narrow LoA, were evident.

It is possible that the closer agreement between the GENEActiv and Axivity than for either accelerometer with the ActiGraph is due to the taping together of the GENEActiv and Axivity accelerometers, whereas the ActiGraph was worn proximal to the GENEActiv and Axivity. Although this may have contributed, the 11% higher acceleration from the GENEActiv relative to the ActiGraph (across the total period) in the current study is not dissimilar to the 13%–16% higher acceleration observed in our earlier studies where the GENEActiv and ActiGraph GT3X+ were taped together at the wrist (28) and hip (12). These consistent differences we and others have observed may relate to technical differences between the brands (12,20,36); specifically, it seems that there is some onboard processing of the raw acceleration signal of the ActiGraph device, but details of this are proprietary (36,37).

The most recent version of the ActiGraph, the GT9X Link, was used in this study. How this compares with the previous version, the GT3X+, which is the accelerometer that has been deployed in large surveys, for example, National Health and Nutrition Examination Survey, is important. The accelerometer sensor in the GT9X is the same as the sensor in GT3X+ so good comparability between the two would be anticipated; perhaps more importantly, there are differences in the design of the devices that may affect orientation when worn. However, this should affect individual axis output rather than resultant metrics, such as the ENMO average acceleration metric. Recently, Montoye and colleagues (38) presented a comparison of the GT9X and GT3X+ ActiGraphs. They reported that raw acceleration data were highly correlated between models. Furthermore, our results comparing the GT9X with the GENEActiv are very similar to those from our previous comparison of the GT3X+ with the GENEActiv (20).

Strengths of this study include the large range of sedentary and light activities incorporated in a simulated free-living protocol designed to encourage a range of natural self-paced behaviors common in normal daily life. This elicited an average acceleration across the total period similar to the daily average acceleration observed in free-living individuals (6), strengthening the ecological validity of the results. The observation facilitated the evaluation of specific activities and the accuracy and comparability of sedentary/light thresholds. The evaluation of the agreement of key physical activity outcomes between accelerometer brands will facilitate selection of the most appropriate outcomes to use when comparing studies that have used different accelerometer brands. Critically, the GGIR accelerometer processing package used in this study (17–19) can be easily applied to large data sets and is available open-source, as is the function we created to convert Axivity raw files to a format that can be analyzed in GGIR. Limitations of the study include the following: the small sample size, although the study was powered appropriately; the self-selected homogenous young fit cohort; and the relatively small amount of time spent in MVPA, given the focus of our protocol design on sedentary and light activities. Furthermore, the short duration of the study precluded the examination of sleep and the range of behaviors across a 24-h day.

In conclusion, this study suggests that key physical activity outcomes from GGIR (acceleration and time spent in intensity cut points) can be considered equivalent between the Axivity and GENEActiv accelerometers. Comparability with the ActiGraph is reduced, with acceleration and MVPA approximately 9%–11% lower; however, time spent sedentary and time in light-intensity activity can be considered equivalent. It should be noted that these results are generalizable only to studies using the same wear location (nondominant wrist) and processing the data in GGIR. To ensure the comparability of the accelerometer brands was tested and not confounded by differences in wear location, all accelerometer brands were worn on the nondominant wrist. This is common practice in most studies (e.g., Refs. [6,9,21–24]), but a notable exception is UK Biobank, where the Axivity was worn on the dominant wrist (4). The next step is to examine the agreement between accelerometer brands because they are commonly used, that is, in a true free-living environment with the GENEActiv and ActiGraph again on the nondominant wrist, but the Axivity on the dominant wrist, as in UK Biobank.

The authors thank Hannah Goodes, Carly Kingdon, and Megan Waters for their assistance with data collection and the participants for their involvement with this study. The research was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at the University Hospitals of Leicester and Loughborough University, the NIHR for Leadership in Applied Health Research and Care—East Midlands, and the Leicester Clinical Trials Unit. The views expressed are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health. No external sources of funding were accessed.

The results of the present study do not constitute endorsement by the American College of Sports Medicine. We declare that the results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.

Back to Top | Article Outline

REFERENCES

1. Chen Y, Copeland WK, Vedanthan R, et al. Association between body mass index and cardiovascular disease mortality in East Asians and South Asians: pooled analysis of prospective data from the Asia Cohort Consortium. BMJ. 2013;347:f5446.
2. Mons U, Müezzinler A, Gellert C, et al. Impact of smoking and smoking cessation on cardiovascular events and mortality among older adults: meta-analysis of individual participant data from prospective cohort studies of the CHANCES consortium. BMJ. 2015;350:h1551.
3. Lee IM, Shiroma EJ, Lobelo F, et al. Effect of physical inactivity on major noncommunicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet. 2012;380(9838):218–29.
4. Doherty A, Jackson D, Hammerla N, et al. Large scale population assessment of physical activity using wrist worn accelerometers: the UK Biobank Study. PLoS Med. 2017;12(2):e0169649
5. Swerdlow AJ, Jones ME, Schoemaker MJ, et al. The Breakthrough Generations Study: design of a long-term UK cohort study to investigate breast cancer aetiology. Br J Cancer. 2011;105(7):911–7.
6. da Silva IC, van Hees VT, Ramires VV, et al. Physical activity levels in three Brazilian birth cohorts as assessed with raw triaxial wrist accelerometry. Int J Epidemiol. 2014;43(6):1959–68.
7. Wake M, Clifford S, York E, et al. Introducing Growing Up in Australia’s Child Health CheckPoint: A physical and biomarkers module for the Longitudinal Study of Australian Children. Fam Matters 2014;94:15–23.
8. Li X, Kearney PM, Keane E, et al. Levels and sociodemographic correlates of accelerometer-based physical activity in Irish children: a cross-sectional study. J Epidemiol Community Health. 2017;71(6):521–7.
9. Menai M, van Hees VT, Elbaz A, Kivimaki M, Singh-Manoux A, Sabia S. Accelerometer assessed moderate-to-vigorous physical activity and successful ageing: results from the Whitehall II study. Sci Rep. 2017;8:45772.
10. Troiano RP, McClain JJ, Brychta RJ, Chen KY. Evolution of accelerometer methods for physical activity research. Br J Sports Med. 2014;48(13):1019–23.
11. Wijndaele K, Westgate K, Stephens SK, et al. Utilization and harmonization of adult accelerometry data: review and expert consensus. Med Sci Sports Exerc. 2015;47(10):2129–39.
12. Rowlands AV, Fraysse F, Catt M, et al. Comparison of measured acceleration output from accelerometry-based activity monitors. Med Sci Sports Exerc. 2015;47(1):201–10.
13. Hildebrand M, Van Hees VT, Hansen BH, Ekelund U. Age group comparability of raw accelerometer output from wrist- and hip-worn monitors. Med Sci Sports Exerc. 2014;46(10):1816–24.
14. Hildebrand M, Hansen BH, van Hees VT, Ekelund U. Evaluation of raw acceleration sedentary thresholds in children and adults. Scand J Med Sci Sports. 2017;27(12):1814–23.
15. Bakrania K, Yates T, Rowlands AV, et al. Developing and validating intensity-based thresholds on raw accelerometer data for discriminating between sedentary behaviours and light-intensity physical activities: a MAD approach. PLoS One. 2016;11(10):e0164045.
16. van Hees VT, Thaler-Kall K, Wolf KH, et al. Challenges and opportunities for harmonizing research methodology: raw accelerometry. Methods Inf Med. 2016;55;(6):525–32.
17. van Hees VT, Gorzelniak L, Dean León EC, et al. Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity. PLoS One. 2013;8(4):e61691.
18. van Hees VT, Renstrőm F, Wright A, et al. Estimation of daily energy expenditure in pregnant and non-pregnant women using a wrist-worn tri-axial accelerometer. PLoS One. 2011;6(7):e22922.
19. van Hees VT, Fang Z, Langford J, et al. Auto-calibration of accelerometer data for free-living physical activity assessment using local gravity and temperature: an evaluation on four continents. J Appl Physiol. 2014;117(7):738–44.
20. Rowlands AV, Yates T, Davies M, Khunti K, Edwardson CL. Raw accelerometer data analysis with GGIR R-package: does accelerometer brand matter? Med Sci Sports Exerc. 2016;48(10):1935–41.
21. Ramires VV, Wehrmeister FC, Böhm AW, et al. Physical activity levels objectively measured among older adults: a population-based study in a Southern city of Brazil. Int J Behav Nutr Phys Act 2017;14:13.
22. Bell JA, Hamer M, van Hees V, et al. Healthy obesity and objective physical activity. Am J Clin Nutr. 2015;102:268–75.
23. Horta BL, Schaan BD, Bielemann RM, et al. Objectively measured physical activity and sedentary-time are associated with arterial stiffness in Brazilian young adults. Atherosclerosis. 2015;243:148–54.
24. Fairclough SJ, Noonan R, Rowlands AV, van Hees V, Knowles Z, Boddy LM. Wear compliance and activity in children wearing wrist- and hip-mounted accelerometers. Med Sci Sports Exerc. 2016;48(2):243–53.
25. Rowlands AV, Cliff DP, Fairclough SJ, et al. Moving forward with backward compatibility: translating wrist accelerometer data. Med Sci Sports Exerc. 2016;48(11):2142–9.
26. Scott JJ, Rowlands AV, Cliff D, Morgan PJ, Plotnikoff RC, Lubans DR. Testing the validity and feasibility of the GENEActiv accelerometer in free-living adolescents. J Sci Med Sport. 2017. doi.org/10.1016/j.jsams.2017.04.017.
27. Ladha C, Ladha K, Jackson D, Olivier P. Table validation of open movement Ax3 accelerometer. In: Proceedings of the 3rd International Conference of Ambulatory Monitoring Physical Activity and Movement; 2013 Jun 17–19: Amherst (MA). University of Amherst; 2013. pp. 69–70.
28. Stiles VH, Griew PJ, Rowlands AV. Use of accelerometry to classify activity beneficial to bone in premenopausal women. Med Sci Sports Exerc. 2013;45(12):2353–61.
29. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307–10.
30. Wellek S. Testing Statistical Hypotheses of Equivalence. Boca Raton (FL): Chapman & Hall/CRC; 2003. Xv, p. 284.
31. Lee JM, Kim Y, Welk GJ. Validity of consumer-based physical activity monitors. Med Sci Sports Exerc. 2014;46(9):1840–8.
32. Kim Y, Welk GJ. Criterion validity of competing accelerometry-based activity monitoring devices. Med Sci Sports Exerc. 2015;47(11):2456–63.
33. Rowlands AV, Olds TS, Hillsdon M, et al. Assessing sedentary behavior with the GENEActiv: introducing the sedentary sphere. Med Sci Sports Exerc. 2014;46(6):1235–47.
34. Sasaki JE, Hickey AM, Staudenmeyer JW, Kent JA, Freedson PS. Performance of activity classification algorithms in free-living older adults. Med Sci Sports Exerc. 2016;48(5):941–50.
35. Sedentary Behaviour Research Network. Letter to the editor: standardized use of the terms “sedentary” and “sedentary behaviours.” Appl Physiol Nutr Metab. 2012;37:540–2.
36. John D, Sasaki J, Staudenmayer J, Mavilia M, Freedson PS. Comparison of raw acceleration from the GENEA and ActiGraph™ GT3X+ activity monitors. Sensors (Basel). 2013;13:14754–63.
37. Brønd JC, Arvidsson D. Sampling frequency affects the processing of ActiGraph raw acceleration data to activity counts. J Appl Physiol (1985). 2016;120:362–9.
38. Montoye AM, Nelson M, Bock J, et al. Comparability of raw and count-based data from the ActiGraph GT9X Link and GT3X+ accelerometers. In: Proceedings of the 5th International Conference of Ambulatory Monitoring Physical Activity and Movement; 2017 June 21–23: Bethesda (MD). National Institutes of Health, Natcher Conference Center; 2017: Abstract. pp. 3–61.
Keywords:

ACTIGRAPH; AXIVITY; GENEACTIV; GGIR; GT9X; LINK

© 2018 American College of Sports Medicine