Abstract
Objectives:
Non-wear time algorithms have not been validated in pregnant women with overweight/obesity (PW-OW/OB), potentially leading to misclassification of sedentary/activity data, and inaccurate estimates of how physical activity is associated with pregnancy outcomes. We examined: (1) validity/reliability of non-wear time algorithms in PW-OW/OB by comparing wear time from five algorithms to a self-report criterion and (2) whether these algorithms over- or underestimated sedentary behaviors.
Study Design:
PW-OW/OB (N=19) from the [anonymous] randomized controlled trial wore an ActiGraph GT3x+ for 7 consecutive days between 8–12 weeks gestation.
Methods:
Non-wear algorithms (i.e., consecutive strings of zero acceleration in 60-second epochs) were tested at 60, 90, 120, 150, and 180-minutes. The monitor registered sedentary minutes as activity counts 0–99. Women completed daily self-report logs to report wear time.
Results:
Intraclass correlation coefficients for each algorithm were 0.96–0.97; Bland-Altman plots revealed no bias; mean absolute percent errors were <10%. Compared to self-report (M=829.5, SD=62.1), equivalency testing revealed algorithm wear times (minutes/day) were equivalent: 60- (M=816.4, SD=58.4), 90- (M=827.5, SD=61.4), 120- (M=830.8, SD=65.2), 150- (M=833.8, SD=64.6) and 180-minute (M=837.4, SD=65.4). Repeated measures ANOVA showed 60- and 90-minute algorithms may underestimate sedentary minutes compared to 150- and 180-minute algorithms.
Conclusions:
The 60, 90, 120, 150, and 180-minute algorithms are valid and reliable for estimating wear time in PW-OW/OB. However, implementing algorithms with a higher threshold for consecutive zero counts (i.e., ≥150-minutes) can avoid the risk of misclassifying sedentary data.
Keywords: Activity monitor, accelerometer, non-wear algorithm, physical activity, pregnancy
Introduction
Research measuring physical activity using device-based measurement in pregnant women with overweight or obesity (PW-OW/OB) is scant. Further, the literature in this area is limited by methodological challenges such as recall and social desirability bias from self-reported physical activity, differences in placement of devices, and variations in protocol methods to analyze device-based data.1,2 Moreover, most PW-OW/OB seem to be inactive or engage in low levels of activity3,4 making it more difficult to accurately assess whether women are meeting recommended physical activity guidelines (i.e., >150 minutes, moderate intensity/week).5 In recent years, researchers have increasingly advocated to use device-based assessments such as accelerometers to estimate whether pregnant women are achieving recommended levels of physical activity and how achieving these levels are associated with pregnancy outcomes.6
There are, however, no standard guidelines for processing accelerometer data in PW-OW/OB. This contributes to the inconsistency in the current definition for non-wear time activity with the ActiGraph GT3x+ accelerometer. Non-wear time algorithms (i.e., automated approaches to determining whether an accelerometer is worn) have not been validated in PW-OW/OB, which may lead to misclassification of sedentary data and inaccurate estimates of daily physical activity levels. When the ActiGraph GT3x+ is not being worn, it will typically result in a string of 60-second epoch zero activity counts. Non-wear time algorithms indicate the number of minutes a device allows for a string of consecutive zero activity counts before registering the monitor as not worn. The general consensus in the literature is that non-wear time using different accelerometers should be set to at least 90-minutes ranging up to 180-minutes.7 For example, if the non-wear time algorithm is set at 90-minutes, once a string of zero activity counts hits 90 consecutive minutes, the monitor will register that time as “non-wear.”
However, these studies were conducted in children, adults, older adults, and postpartum women.7–10 Of particular interest, none of the studies specifically focused on PW-OW/OB. This is an issue given that accelerometers can result in inaccuracies as a result of excess abdominal mass due to tilt angle11 and that the increase in abdominal mass specific to pregnancy may have similar effects on accuracy. Therefore, standard protocols for processing accelerometer data may not be applicable to pregnant women. Despite the lack of evidence, many researchers using accelerometers to measure physical activity in pregnancy are defaulting to the standard algorithm of at least 90-minutes.12 This is especially problematic among PW-OW/OB given their high rates of sedentary behavior and low levels of physical activity compared to the general population.3,4 Low levels of physical activity, sedentary behavior, and sleeping or napping will also all result in a string of zero activity counts, making it difficult to distinguish between these activities. Inaccurately identifying these strings of zero activity counts can lead to over or underestimating wear time, which in turn, can result in misclassifying physical activity, sedentary and sleep behaviors and thus contribute to lack of understanding how these activities are associated with pregnancy outcomes such as gestational weight gain, gestational diabetes mellitus, hypertensive disorders, and preterm birth.5
Thus, the purpose of this study was to examine the validity and reliability of non-wear time algorithms developed for the general population for estimating non-wear time in PW-OW/OB by comparing wear time derived from five accelerometer non-wear time algorithms (60-, 90-, 120-, 150-, and 180-minutes) to a self-report criterion. A secondary purpose was to examine the extent to which these non-wear time algorithms over- or underestimated sedentary behaviors. Given the high levels of sedentary behavior among PW-OW/OB,3, 4 it was hypothesized that algorithms with a higher threshold for consecutive zero counts (i.e., ≥ 120-minutes) are more valid and reliable than lower thresholds thus resulting in higher wear time that more closely matches self-reported wear time. It was also hypothesized that these higher thresholds will allow for the monitor to register more sedentary time rather than non-wear time.
Methods
Women (N = 31) were PW-OW/OB participating in a GWG regulation intervention.13 Women were randomized to an intervention or control group from ~8–36 weeks gestation. Women were recruited via on-site clinic, community-based, and web-based strategies. Inclusion criteria were: pregnant women with overweight or obesity (body mass index [BMI] > 25.0 kg/m2), singleton pregnancy 8–12 weeks gestation, 18–40 years old, English-speaking and with physician consent to participate. A more detailed explanation of the intervention has been described elsewhere.13 Women were excluded from this current study if they were non-compliant with wearing the ActiGraph GT3x+ (n = 1) or completing self-report logs (n = 10), or had an extreme value of wear time from the ActiGraph GT3x+ monitor or the self-report log (i.e., > 3 standard deviations from the mean; n = 1). Thus, a final sample of N = 19 (collapsed across intervention group) was used for the study analyses.
This study was approved by the Institutional Review Board from a University in the Northeast for all research activities. Women met at the University Clinical Research Center at 8–12 weeks gestation where study procedures were explained and written informed consent obtained. Women were given an ActiGraph GT3x+ monitor to wear for seven consecutive days. They were instructed to wear the monitor, secured by a nylon belt, at the waist with the monitor sitting on top of their hip (i.e., anterior iliac crest) on their dominant-hand side, in line with their knee (i.e., midline of thigh). Past evidence shows that pregnant women have increased muscle activity at the hip, knee, and ankle joints14 suggesting placement of accelerometers at these joints may influence monitor data. However, past research has established validity for placement of accelerometers on top of the hip in line with the knee in pregnant women.15 This is consistent with evidence in individuals with overweight/obesity showing that placement of an accelerometer on top of the hip in line with the midline of the thigh can reduce inaccuracies in the monitors resulting from large tilt angles.11 They were asked to place the monitor on their hip when they woke up in the morning, wear it during waking hours, and take the monitor off at bedtime. Women were also asked to remove the monitor during water activities (e.g., shower, swimming, etc.). Each day the monitor was worn, women completed a self-report log sheet reporting the start date and time the monitor was put on and off throughout the day.
All ActiGraph GT3x+ data were downloaded and processed using the ActiLife (version 6, ActiGraph, LLC., Pensacola, FL) software under the following five different non-wear zero count algorithms that automatically defined non-wear time: 60-, 90-, 120-, 150-, 180-minutes. Once consecutive zero counts met the threshold, the monitor registered as non-wear time. For example, after the activity data was downloaded under the 60-minute algorithm, any portion that consisted of consecutive zero counts for 60-minutes or longer was labeled as non-wear time. ActiLife then produced reports calculating the amount of wear time minutes, sedentary minutes (activity count cut-points: 0–99), and moderate-to-vigorous physical activity (activity count cut-points: ≥1,952) for each day for each woman under each algorithm. All monitor data was used including the first and last hours of wearing the device given the purpose of this study was to examine the discrepancy between the monitor and self-report. Self-reported wear time was calculated based on the self-reported logs and was calculated as minutes between the time the women reported placing the monitor on and off each day. Although self-report logs for reporting activity monitor wearing practices are not without limitations, our confidence in accuracy of these logs were increased by using cross-checking of the data with the ActiGraph GT3x+ as done in past research.7 To further ensure accuracy of the self-reported wear time, logs were visually inspected for missing data points and corrected based on when the monitor was worn. Of the 133 possible data points (i.e., 19 women times 7 days), there were 30 missing data points from the self-report logs (i.e., missing time of when the monitor was put on/taken off). All of these missing data points were hand corrected using the ActiGraph inclinometer (i.e., algorithm that detects sitting, standing, laying) resulting in 0% missing data from the self-report logs, further increasing the confidence in accuracy.
Statistical analyses were conducted using SPSS v25. Means and standard deviations were used to describe mean difference between self-reported wear time and wear time from the five different algorithms. Mean differences were calculated as [self-reported wear time] – [algorithm wear time]. To assess the reliability of the self-report and wear time algorithms for estimation of wear time, we calculated intraclass correlation coefficients (ICC) for absolute agreement between the methods; estimates <0.50, 0.50–0.75, 0.75–0.90, and >0.90 were indicative of poor, moderate, good, and excellent reliability, respectively.16 Bland-Altman plots were constructed to visually observe over- or underestimation of the wear time algorithms and to explore potential systematic bias in the estimates. Limits of agreement were calculated as the mean difference between each algorithm wear time and self-reported wear time ± 95% confidence interval. Over- or underestimation was considered significant if the 95% confidence intervals did not include zero. Mean absolute percent errors (MAPE) for estimation of wear time between the wear time algorithms and self-report were also assessed to obtain an additional measure of validity. MAPE was calculated as the mean of: [(daily algorithm wear time–daily self-reported wear time /daily self-reported wear time)*100]. Researchers have suggested a MAPE value of 10–15% is a reasonable amount of error when examining the validity of consumer-based physical activity monitors.17 We also performed equivalence testing by comparing the 90% confidence interval of the mean wear time from each algorithm with a 10% equivalence zone defined as the wear time range equal to the criterion measure (self-report) mean ± 10%.18 Finally, mean sedentary minutes between the five algorithms were compared using a repeated measures analysis of variance (RM ANOVA). Data were separated by weekday and weekend days and analyzed separately; analyses revealed there were no differences, thus results are presented together.
Results
Mean age of study participants was 30.9 (SD = 3.2) years, mean gestational age was 10.1 (SD = 1.7) weeks, mean pre-pregnancy BMI was 32.5 (SD = 7.7; overweight = 52.6%, obese = 47.4%) and mean moderate-to-vigorous physical activity per day was 31.1 minutes (SD = 14.5). The sample was Non-Hispanic, White (100%), married (89.5%), employed full-time (89.5%), had a family income ≥ $40,000 (73.7%), and the highest level of education was a graduate/professional degree (52.6%).
Means, standard deviations, intraclass correlation coefficients, and MAPEs are presented in Table 1. According to the daily self-report logs, women wore the ActiGraph GT3x+ for an average of 829.5 (SD = 62.1) minutes/day. The average wear time from the algorithms ranged from 816.4 (60-minute) to 837.4 (180-minute) minutes/day. Descriptive means revealed that compared to the self-reported wear time, wear time from the 60-, 90-, 120-, 150-, and 180-minute algorithms differed by 13.0, 2.0, −1.3, −4.3, and −8.0 minutes/day, respectively.
Table 1.
Descriptives of wear time, mean absolute percent error, and intraclass correlation coefficients.
| Sedentary time (minutes ± SD) | Wear time (minutes ± SD) | Wear time difference from criterion (minutes ± SD) | ICC against self-report | Mean absolute percent error for wear time (% ± SD) | |
|---|---|---|---|---|---|
| Self-Report (criterion) | -- | 829.5 ± 62.1 | -- | -- | -- |
| 60-Minute | 547.9 ± 56.9 | 816.4 ± 58.4 | 13.0 ± 15.7 | 0.97* | 5.4 ± 4.4 |
| 90-Minute | 558.8 ± 55.3 | 827.5 ± 61.4 | 2.0 ± 20.0 | 0.97* | 5.2 ± 4.3 |
| 120-Minute | 562.1 ± 58.2 | 830.8 ± 65.2 | −1.3 ± 24.5 | 0.96* | 5.4 ± 5.0 |
| 150-Minute | 565.1 ± 59.3 | 833.8 ± 64.6 | −4.3 ± 24.1 | 0.96* | 5.3 ± 5.1 |
| 180-Minute | 568.8 ± 61.2 | 837.4 ± 65.4 | −8.0 ± 24.6 | 0.96* | 5.1 ± 5.1 |
Note. SD = standard deviation; ICC = intraclass correlation coefficient;
p<0.01
Intraclass correlation coefficients values for each algorithm ranged from 0.96–0.97, indicating excellent agreement of wear time between each wear time algorithm and self-report. Bland-Altman plots (Figure 1) revealed no bias in estimation of wear time suggesting that each wear time algorithm was in agreement with the self-report criterion. MAPE values for wear time were similar between each wear time algorithm compared to the self-report criterion and were all <10% resulting in a reasonable amount of error.17 Equivalency testing (Figure 2) revealed that each wear time algorithm 90% confidence interval fit within the ± 10% equivalence zone suggesting that wear time from each algorithm was equivalent to self-reported wear time.
Figure 1.
Bland-Altman plot between self-report (criterion) and ActiGraph algorithms for estimates of wear time (minutes).
Note. Solid black line = mean of the difference between the methods; red dashed lines = upper and lower 95% confidence intervals of the mean difference. Diff = difference; alg = algorithm; SR = self-report.
Figure 2.
Equivalency testing for agreement in wear time between self-report and non-wear time algorithms.
Note. Gray area indicates proposed equivalency zone (±10% of the average self-report wear time); solid black lines indicate 90% confidence intervals for estimated wear time from non-wear time algorithms.
The average sedentary time produced by the algorithms ranged from 547.9 (60-minute) to 568.8 (180-minute) minutes/day. The RM ANOVA trended towards significance, Wilks’ Lambda = 0.57, F(4, 15) = 2.9, p = 0.06, and revealed that the 60-minute algorithm underestimated sedentary minutes compared to the 150- (mean difference = −17.2 minutes/day) and 180-minute algorithm (mean difference = −20.9 minutes/day; p’s < 0.05). Also, the 90-minute algorithm underestimated sedentary minutes compared to the 180-minute algorithm (mean difference = −9.9 minutes/day).
Discussion
The purposes of this study were to examine the validity and reliability of non-wear time algorithms for processing wear time of accelerometers in PW-OW/OB by examining five non-wear algorithms for the ActiGraph GT3x+ and to examine whether these algorithms over- or underestimated sedentary behavior. In summary, the 60-, 90-, 120-, 150-, and 180-minute algorithms resulted in excellent ICCs, low MAPEs, and produced wear time estimates that were equivalent to wear time from self-report logs. However, results also show that the 60- and 90-minute algorithms trended towards underestimating sedentary behavior compared to the 150- and 180-minute algorithms. Overall, these findings suggest that the 60-, 90-, 120-, 150-, and 180-minute algorithms are valid and reliable for estimating wear time when processing ActiGraph data in PW-OW/OB. However, implementing higher thresholds for non-wear time algorithms (i.e., ≥150-minutes) can avoid the risk of underestimating sedentary behavior. This will also allow for a better understanding of physical activity and sedentary behaviors in pregnancy in an effort to identify how these behaviors are associated with pregnancy outcomes.
In contrast to the hypothesis, wear times from each of the ActiGraph non-wear algorithms produced excellent ICC and low MAPE values and were equivalent to the self-reported wear time indicating good validity and agreement/reliability. These findings suggest that there is flexibility in choosing the most appropriate non-wear time algorithm for PW-OW/OB. However, lower thresholds (i.e., 60- and 90-minutes) may classify sedentary data incorrectly as non-wear time given that sedentary data can mimic non-wear time by also producing a consecutive string of zero activity counts as shown in this study. That is, we found the 60- and 90-minute algorithms underestimated sedentary time compared to the 150- and 180-minute algorithms. PW-OW/OB engage in high sedentary behavior3,4 and may be sedentary for an extended period of time and thus non-wear time algorithms with lower thresholds may risk misclassifying sedentary data in this particular population and thus influencing implications of activity and sedentary behaviors. Overall, if there is a lack of information about past activity levels of the target population or there is concern with the risk of misclassifying and excluding essential data (i.e., sedentary data), implementing an algorithm with a higher threshold of consecutive zero counts (i.e., at least 150-minutes) can avoid the risk of underestimating the true amount of wear time.
This study is the first to our knowledge to examine non-wear time algorithms in PW-OW/OB and establish reliability and validity of non-wear time algorithms for processing wear time of ActiGraph accelerometers in PW-OW/OB. A strength of the study was the frequency of assessments. Women wore the ActiGraph GT3x+ for seven consecutive days, which allowed for a better understanding of average physical activity and sedentary behaviors per day. An additional strength was the use of a detailed self-report log that was hand corrected using the ActiLife program, which increased confidence of accuracy and allowed for a strong criterion. However, self-report logs are not without their limitations. For example, self-report logs for physical activity are associated with recall and reporting errors, social desirability, dependency on written language, and other external factors (e.g., complexity of questionnaire).19 Other limitations include the generalizability of the study findings. Although these findings can be generalized to PW-OW/OB, the sample was only representative of women who were White, educated, employed and married. Future studies should replicate these study findings in more diverse samples of PW-OW/OB. Also, despite the inclusion criteria included PW-OW/OB of all activity levels, the current sample was considered on average to be at least lightly active (i.e., only a few women reported to be sedentary). These findings should be replicated in a less active or sedentary population of PW-OW/OB given that this may provide further insight as to whether lower non-wear time algorithms may misclassify sedentary data.
Conclusion
This study was the first to our knowledge to establish validity and reliability of non-wear time algorithms for the ActiGraph in PW-OW/OB and identify that lower thresholds can misclassify sedentary behavior. Incorrectly estimating the true amount of wear time increases the risk of excluding important consecutive strings of zero counts (e.g., sedentary data). The 60-, 90-, 120-, 150-, or 180-minute non-wear time algorithms are valid and reliable for assessing wear time in PW-OW/OB. However, researchers should use non-wear time algorithms with a higher threshold for consecutive zero counts to avoid misclassifying sedentary data. Future studies using these recommendations may help to better identify associations between accelerometer-derived physical activity and maternal-fetal outcomes.
Practical Implications.
Non-wear time ActiGraph algorithms of 60-, 90-, 120-, 150-, or 180-minute are valid and reliable for estimating wear time in pregnant women with overweight or obesity.
Non-wear time ActiGraph algorithms of 60- and 90-minutes may underestimate sedentary minutes compared to the 150- and 180-minute algorithms in pregnant women with overweight or obesity.
It is recommended to use at least a threshold of 150-minutes to avoid the risk of misclassifying sedentary data in pregnant women with overweight or obesity.
Acknowledgements
The authors would like to thank all study participants for their participation. Support of this work was provided by the National Heart, Lung, and Blood Institute (NHLBI) of the National Institutes of Health (NIH) through grant R01HL119245-01.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Declarations of Interest: None
References
- 1.Ainsworth B, Cahalin L, Buman M, et al. The current state of physical activity assessment tools. Prog Cardiovasc Dis. 2015;57:387–95. [DOI] [PubMed] [Google Scholar]
- 2.Mâsse LC, Fuemmeler BF, Anderson CB, et al. Accelerometer data reduction: A comparison of four reduction algorithms on select outcome variables. Med Sci Sports Exerc. 2005;37:S544–54. [DOI] [PubMed] [Google Scholar]
- 3.Lindqvist M, Lindkvist M, Eurenius E, et al. Leisure time physical activity among pregnant women and its associations with maternal characteristics and pregnancy outcomes. Sex Reprod Healthc. 2016;9:14–20. [DOI] [PubMed] [Google Scholar]
- 4.Catov JM, Parker CB, Gibbs BB, et al. Patterns of leisure-time physical activity across pregnancy and adverse pregnancy outcomes. Int J Behav Nutr Phys Act. 2018;15:68. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.ACOG Committee Opinion No. 650: Physical Activity and Exercise During Pregnancy and the Postpartum Period. Obstet Gynecol. 2015;126(6):e135–142. doi: 10.1097/AOG.0000000000001214 [DOI] [PubMed] [Google Scholar]
- 6.Guérin E, Ferraro ZM, Adamo KB, et al. The need to objectively measure physical activity during pregnancy: Considerations for clinical research and public health impact. Matern Child Health J, 2018;22:637–41. [DOI] [PubMed] [Google Scholar]
- 7.Hutto B, Howard VJ, Blair SN, et al. Identifying accelerometer nonwear and wear time in older adults. Int J Behav Nutr Phys Act. 2013;10:1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Heesch KC, Hill RL, Aguilar-Farias N, et al. Validity of objective methods for measuring sedentary behaviour in older adults: A systematic review. Int J Behav Nutr Phys Act. 2018;15:119. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Migueles JH, Cadenas-Sanchez C, Ekelund U, et al. Accelerometer data collection and processing criteria to assess physical activity and other outcomes: A systematic review and practical considerations. Sports Med. 2017;47:1821–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Evenson KR, Terry JW Jr. Assessment of differing definitions of accelerometer nonwear time. Res Q Exerc Sport. 2009;80:355–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Crouter SE, Schneider PL, Bassett DR Jr. Spring-levered versus piezo-electric pedometer accuracy in overweight and obese adults. Med Sci Sports Exerc. 2005;37:1673–9. [DOI] [PubMed] [Google Scholar]
- 12.Hesketh KR, Evenson KR, Stroo M, et al. Physical activity and sedentary behavior during pregnancy and postpartum, measured using hip and wrist-worn accelerometers. Prev Med Rep. 2018;10:337–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Anonymous [2018]. Details omitted for double-blind reviewing. [Google Scholar]
- 14.Gottschall JS, Sheehan RC, Downs DS. Pregnant women exaggerate cautious gait patterns during the transition between level and hill surfaces. J Electromyogr Kines. 2013;23:1237–42. [DOI] [PubMed] [Google Scholar]
- 15.DiNallo JM, Symons Downs D, Le Masurier G. Objectively assessing treadmill walking during the second and third pregnancy trimesters. J Phys Act Health. 2012;9:21–8. [DOI] [PubMed] [Google Scholar]
- 16.Koo TK, Li MY. A guidelines of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med. 2016;15:155–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lee JM, Youngwon K, Welk GJ. Validity of consumer-based physical activity monitors. Med Sci Sports Exerc. 2014;46:1840–8. [DOI] [PubMed] [Google Scholar]
- 18.Dixon PM, Saint-Maurice PF, Kim Y, et al. A primer on the use of equivalence testing for evaluating measurement agreement. Med Sci Sports Exerc. 2018;50:837–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sylvia LG, Bernstein EE, Hubbard JL, Keating L, Anderson EJ. A practical guide to measuring physical activiy. J Acad Nutr Diet. 2015;114:199–208. [DOI] [PMC free article] [PubMed] [Google Scholar]