Background
Accelerometers have provided epidemiologists with an objective tool to measure movement, which overcomes many limitations by self-report measures. Still, there are many unresolved issues regarding data reduction algorithms and standardization to secure the highest possible data quality and consistency among studies [
1,
2]. While choices related to the length of epochs used, criteria for hours to constitute a valid day and a valid week, and cut points deserve attention, the definition of non-wear time has received relatively little interest in the literature. As shown by Cain et al. [
2], only half of pediatric studies applying accelerometry have reported the non-wear criterion used. This is surprising, as this criterion may have a profound influence on the data subsequently available for analyses [
3‐
5] and on PA and SED estimates [
4‐
9].
Non-wear time is the time during a measurement period where participants do not wear the accelerometer, and should be excluded from further analyses on the assumption that the remaining wear time is sufficiently representative for the whole measurement period. A challenge, however, given the considerable interest in sedentary behavior [
10‐
13], is that the detection and removal of non-wear time is based on removing continuous time (consecutive epochs) of zero counts that may be easily confused with sedentary time. This confusion may lead to erroneous estimates of participants’ PA and SED.
The optimal non-wear criterion will likely vary with the population under investigation, due to different PA and SED patterns [
8,
14‐
17], and possibly also with the type of accelerometer being applied. In children and youth, non-wear time definitions using Actigraph monitors vary from 10 to 180 min of consecutive minutes of zero counts, with 10 and 20-min criteria being the most used [
2]. Variation is also seen in studies with adults, although a 60-min criterion allowing for 0–2 min of non-zero counts are frequently used [
18‐
22]. However, according to a 24-h laboratory-based study, this criterion clearly overestimated non-wear time and underestimated SED in overweight-to-obese youth and adults [
17]. Thus, the authors recommended a 90-min rather than a 60-min criterion. Although these findings were limited in terms of applicability to a free-living setting [
23], several other studies in adults have recommended longer durations of the non-wear criterion being applied (90–180 min of consecutive zero counts with and without allowance of interruptions) [
6,
24‐
26]. It should be noted that while these studies are based on different monitors (Actigraph 7164 and GT3X+, Actical, StepWatch), there seems to be a movement towards applying longer duration non-wear criteria to avoid misclassification of SED as non-wear time.
Based on 369–517 children aged 8–13 years Janssen et al. [
8] compared non-wear criteria of 10, 20 and 60 min of consecutive zero counts against logs. They suggested a 20-min criterion being used in future studies, although the 20 and 60-min criteria performed similarly. This conclusion supports Esliger et al’s [
27] suggestion that 20 min of consecutive zero counts is appropriate in children, as motionless bouts ≥20 min are biologically implausible. However, it has been suggested that this low threshold causes an unrealistic high number of non-wear periods, leading Chinapaw et al. [
28] to suggest a 60-min criterion being used. Interestingly, the International Children’s Accelerometry Database (ICAD) [
29], a large database pooling many child cohort studies using objective measurements of PA, apply a 60-min criterion allowing for 2 min of non-zero counts, which might be expected to result in a valid wear time somewhere in between of the previous recommendations. These conflicting recommendations and practices lead to confusion and reduced comparability among studies in pediatric epidemiology, and hinder knowledge development concerning the health promoting effects of physical activity.
The aim of the present study was to compare non-wear periods and wear time among 10 different accelerometer non-wear time criteria and a log during a 7-day measurement period in a large sample of 11-year-old children.
Discussion
We compared the number of waking non-wear periods and wear time among 10 different accelerometer non-wear criteria and logs of non-wear during a 7-day measurement period in 11-year-old children. According to the logs, the number of waking non-wear periods ranged from 0 to 3 periods per day; on most days children did not remove the accelerometer (77.1% of days). The choice of criterion caused up to 10% variation in population level estimates of SED and PA level after controlling for wear time. Yet, the practical significance of using different non-wear criteria for association analyses – where the rank of the children are important – are minor for criteria between 20 and 60-min, as correlations among estimates obtained from these criteria were high (ρ ≥ 0.97). The greatest differences among the criteria were found for estimates of SED in bouts of long durations. Determining wear and non-wear time will always be a classification-misclassification compromise, as SED and non-wear will be impossible to separate accurately using the current approach.
Previous studies have recommended defining child non-wear time as 20 to 60 min of consecutive zero counts (not allowing any time above zero counts) [
8,
27,
28]. The recommendations by Esliger et al. [
27] and Chinapaw et al. [
28] were based on plausible values for length and number of non-wear periods, respectively. The current study together with the studies by Chinapaw et al. [
28] and Toftager et al. [
5] are rather consistent showing up to 10, 7, and 4 non-wear periods per day for 20, 30 and 60-min non-wear criteria, respectively. In comparison, we found 0–3 non-wear periods per day reported by the log. As concluded previously [
28], we find it implausible that children remove the accelerometer more than 3–4 times per day on average. This conclusion is consistent with the current findings showing the best agreement between logbook and accelerometer estimates for the 45 (0–4 non-wear periods per day) and 60-min (0–3 non-wear periods per day) criteria. Yet, with increasing age, overall SED increase [
34], the time spent in longer bouts of SED increase [
8,
14,
15], and the number of breaks in SED time decrease [
14,
16]. This development increase the risk to misclassify SED as non-wear time with increased age. This hypothesis is supported by a somewhat increased number of non-wear bouts across criteria in the study by Toftager et al. [
5] in 11-to-14-year-olds and Chinapaw et al. [
28] in 9-to-13-year-olds compared to the present study (11-year-olds), increased differences between SED estimates using more prolonged non-wear duration with older age [
8,
9], and findings showing a greater overestimation of non-wear periods in adults compared to youth [
17]. This supports the use of a 60-min criterion over a 45-min criterion in children older than 11-years old. Although the evidence for a relation of misclassification to other sociodemographic characteristics are conflicting [
35,
36], such a relationship could lead to selective attrition due to violation of wear-time requirements, and result in skewed PA and SED estimates [
5]. In any case, supporting a previous study [
5], we do not recommend introducing group-specific non-wear criteria, as we believe such a practice easily could lead to more, rather than less, bias, in addition to increasing the complexity of data reduction procedures.
Criteria allowing some epochs above zero counts allows for touching or moving the accelerometer by accident and spurious spikes of accelerometer counts during non-wear without turning non-wear time into SED time [
4]. Yet, allowing for interruptions might decrease classification accuracy [
24] as well as making results vulnerable to variation in wear time if analyzed with different epoch lengths [
37].
As it is impossible to accurately disentangle SED bouts from non-wear bouts, the applied non-wear criterion will be a trade-off, hopefully leading to the best possible classification-misclassification compromise. We found that showering was the main reason for not wearing the accelerometer for periods less than 30 min of duration (86.5%), whereas swimming in particular, but also forgetting to wear the accelerometer, and other reasons, were important reasons for periods over 30 min. Showering is not SED nor MVPA and would probably not have any meaningful influence on the data at a group or individual level. Thus, failing to remove these non-wear periods are arguable not critical. In contrast, failing to remove non-wear time spent swimming will clearly underestimate PA and overestimate SED, while we might hypothesize other activities might vary in intensity and thereby counterbalance each other. These findings suggest application of longer (more liberal) criteria over shorter (stricter) ones, supporting the 45 or 60-min criterion.
Consistent with previous studies [
4‐
9], we found that the choice of non-wear time criterion affect population level estimates of PA and SED. Interestingly, the non-wear time criterion affects estimates of activity level to a much greater extent than other wear requirements (i.e., valid hours per day and valid days per week) [
5]. Still, the non-wear criterion applied is the algorithm least frequently reported in the pediatric literature (missing in 48.6% of studies) [
2]. We found a difference of up to 17% for SED time, whereas estimates for minutes of LPA, MPA, VPA or MVPA were not influenced by the non-wear criteria. Yet, overall PA (cpm) and proportions of intensity-specific PA varied up to 10% between criteria. Because wear time is most often controlled either as a covariate or by using percentages of valid wear time in analyses, variability is often introduced in estimates for all variables. Still, the high correlations among variables, especially for the 20-to-60 min non-wear time criteria, indicate that the non-wear criterion used probably have a minor impact in correlational designs, as previously shown in a large sample from the European Youth Heart Study [
9]. Nevertheless, we strongly recommend future studies report non-wear time criteria along with other data reduction algorithms applied.
To the best of our knowledge, no previous studies have determined the influence of different non-wear time criteria on SED bouts. Contrary to findings for total SED and PA, we found great differences in estimates of prolonged SED bouts, probably being the most detrimental to health, if assuming that accumulation of SED in bouts were detrimental beyond total SED time. Giving non-wear criteria high attention therefore seems to be crucial in order to determine associations for SED bouts with health outcomes. Worth noting is that previous studies targeting associations between SED bouts and health in children have used a 20-min criterion [
38,
39], a 60-min criterion [
40], and a 60–2 criterion [
41,
42]. Studies in adults have indicated that a longer non-wear criterion detects more time in prolonged bouts of SED [
43,
44], however, such a pattern is not clear in child studies, possibly owing to other inconsistencies in bout definitions (e.g., whether the study allows for interruptions in SED within a bout) [
38‐
42]. The present findings show that most criteria remove nearly all SED time in bouts over 60 min of duration, indicating that 11-year-old children are unable to sit uninterrupted for 60 min or more. Researchers investigating patterns of SED should be aware of this finding.
Strengths and limitations
We consider the inclusion of 10 different accelerometer non-wear criteria together with a logbook for determining non-wear periods and wear time in a large representative sample of children strengths of the current study. In advance of some previous studies [
5,
27,
28], the use of logs allows for determining agreement for different non-wear criteria, beyond reliance on common sense regarding plausible non-wear characteristics. Still, assessing behavior by self-report has several well-known limitations, possibly leading to bias in addition to general inaccuracy. For example, we might expect children to both forget and consciously underreport non-wear, but in contrast to a previous study in adolescents [
33], we have no clear indication of underreporting. Additionally, because we did not assess specific time points for wear and non-wear in the log, we could not assess classification accuracy for each wear/non-wear period as determined by accelerometry.
We believe inclusion of results on non-wear periods as well as SED bouts for the different non-wear criteria are other strengths of the study. We suggest these measures be reported and considered “diagnostic” tools in future studies, to aid deciding whether the applied non-wear criterion seem sensible in a given study. Importantly, including such characteristics allow for qualified judgements about sensible criteria in large surveillance studies, without requirements for any additional data (e.g., logs). Developing new criterion measures, for example for time in and type of SED behavior [
45], will be a clear advantage for establishing such standards.
While tri-axial accelerometry allows for detection of accelerations across three axes and might capture movement beyond a single axis, the present study was limited to apply the vertical axis only. Yet, a previous study has shown similar wear time based on vector magnitude and the vertical axis, indicating this issue being of minor importance [
36].
Acknowledgements
We thank all children, parents, teachers and principals at the participating schools for their excellent cooperation during the data collection. We also thank Turid Skrede, Mette Stavnsbo, Katrine Nyvoll Aadland, Øystein Lerum, Einar Ylvisåker, and students at The Sogn og Fjordane University College for their assistance during the data collection.