Original researchIs the pain of activity log-books worth the gain in precision when distinguishing wear and non-wear time for tri-axial accelerometers?
Introduction
With recent advances in technology, use of accelerometers for the measurement of both physical activity and sedentary behaviour is becoming increasingly popular. There are however, many methodological challenges in processing of accelerometer data and clear evidence-based guidelines for the handling of accelerometer data are lacking.1
One of these challenges is distinguishing sedentary behaviour from non-wear time, as both non-movement and non-wear can result in accelerometer output of zero counts. Most large studies employ automated algorithms to identify non-wear time (e.g. filtering ≥60 or ≥90 min of zero counts, with varying allowances for ‘interruptions’).2, 3 For uni-axial accelerometers, these algorithms have been shown to either systematically over or under-estimate non-wear time.4, 5 Overestimation of non-wear time occurs when participants wear the accelerometer but are very sedentary, resulting in erroneous exclusion of days as invalid (i.e. if wear time is ≤10 h/day) and biases in both time spent in sedentary and higher intensity physical activity.3, 4, 5
One solution is to rely on activity log-books, in which participants are asked to record the times the accelerometer is put on and taken off.6 This clarifies when participants wear the monitor (but are very still) and when they do not wear it. For some participants, the logbook may serve as a reminder to wear the monitor; however, for other participants completing a log may be an extra burden and increase noncompliance. It also creates a burden for researchers, as the log data have to be manually entered and combined with the accelerometer data.
As tri-axial accelerometers pick up movement in three dimensions, the resulting vector magnitude counts (i.e. weighted sum of counts from the three axes) are higher than uni-axial accelerometer counts. Consequently, different algorithms may be needed to differentiate between wear and non-wear time. To our knowledge, only one study has evaluated methods for filtering wear time when using tri-axial accelerometers,7 however, in this study the tri-axial accelerometer was worn on the wrist and results may be different for monitors worn in other locations.8 The aim of this paper was therefore to compare three methods for assessing non-wear time when using the Actigraph GT3X+ accelerometer (Actigraph LLC, Pensacola, FL). The first method used automated filters to detect non-wear time using six different algorithms. The second used on/off time recorded in log-books to determine non-wear time, and the third used a combination of a commonly used automated filter (i.e. 60-min of consecutive zero counts without allowing for interruptions),2, 5 with verification of wear/non-wear time from the full log-book entries (including a description of activities in 15-min intervals). The outcomes were number of participants with valid data and number of valid days, as well as estimates of wear time and time spent in sedentary, light, moderate and vigorous activity, for each of the three methods. The ability of automated and log-book methods to accurately detect wear-time and non-wear time was assessed using the combined method as a reference, because this method included the maximum available information and compensates for disadvantages of the automated and log-book methods.
Section snippets
Methods
Data were from a pilot study designed to examine patterns of occupational and leisure time physical activity and sedentary behaviour among office-based government employees. After an introduction session, with an explanation of the study requirements, employees were invited to participate and to provide written consent. The study protocol was approved by the Medical Research Ethics Committee at the University of Queensland, Brisbane, Australia (National Statement on Ethical Conduct in Human
Results
Forty-five employees agreed to participate and completed the protocol, but one accelerometer failed to record data for one participant. Using the combined method, 34 (76%) of the 45 participants had sufficient valid wear time. Age ranged from 24 to 61 years and the majority were women, Australian born and met the physical activity guidelines. No statistically significant differences were found between participants with and without sufficient valid wear time (Table 1).
Summary data on wear time,
Discussion
The aim of this paper was to compare three methods for assessing non-wear time when using the Actigraph GT3X+ accelerometer. One method used a range of automated algorithms, while the second used log-book recorded on/off times to determine non-wear time. The third method used a combination of one of the automated algorithms with verification from detailed log-book entries. The findings showed only small differences in average wear time and times spent in sedentary, light, moderate and vigorous
Conclusion
In conclusion, automated filters are as accurate as a combination of automated filters and activity log-book for differentiating between accelerometer non-wear and sedentary time. Taking into account the accuracy of identifying wear time and potential loss of valid data due to misclassification, the automated filter based on 90 min of consecutive zeroes, without allowing for interruptions, is recommended for future studies.
Practical implications
- •
This study adds to the recommendations for estimating wear-time from tri-axial accelerometer data.
- •
In terms of filtering wear time from accelerometer data, automated filters are as accurate as a combination of automated filters plus activity log books.
- •
Not requiring activity log-books would reduce the burden for both researchers and participants in accelerometer studies of physical activity measurement.
Acknowledgements
The authors would like to thank the participants for the time and effort given to this study, and Nicola W. Burton, Helen E. Brown and Kylah McCarthy for research assistance.
References (17)
- et al.
Utility of the RT3 triaxial accelerometer in free living: an investigation of adherence and data loss
Appl Ergon
(2010) - et al.
Validation and comparison of ActiGraph activity monitors
J Sci Med Sport
(2011) - et al.
Reliability and validity of a modified self-administered version of the active Australia physical activity survey in a sample of mid-age women
Aust NZ J Public Health
(2008) - et al.
Accelerometer use in physical activity: best practices and research recommendations
Med Sci Sports Exerc
(2005) - et al.
Amount of time spent in sedentary behaviors in the United States, 2003–2004
Am J Epidemiol
(2008) - et al.
Accelerometer data reduction: a comparison of four reduction algorithms on select outcome variables
Med Sci Sports Exerc
(2005) - et al.
Validation of accelerometer wear and nonwear time classification algorithm
Med Sci Sports Exerc
(2011) - et al.
Identifying sedentary time using automated estimates of accelerometer wear time
Br J Sports Med
(2012)
Cited by (41)
Validating the Fitbit Charge 4© wearable activity monitor for use in physical activity interventions
2024, Journal of Science and Medicine in SportComprehensive assessment of postoperative mobility during the first days after mini-invasive lung surgery: A prospective observational study
2023, Journal of Clinical AnesthesiaNeuroelectric indices of motor response preparation are selectively associated with physical activity among adults with obesity
2022, International Journal of PsychophysiologyCitation Excerpt :The raw acceleration signal was sampled at 100 Hz and data were converted to vertical axis counts over 60s epochs using ActiLife software (v. 6.13.3, ActiGraph LLC., Pensacola, FL, USA). ActiLife software was then used to exclude non-wear time, defined as 60 min of consecutive 0 counts (Peeters et al., 2013), and compute time spent in MVPA (min/day) based on the ≥2020 counts per minute (CPM) threshold for vertical axis counts/min (Troiano et al., 2008). Attentional inhibition was assessed using a modified Eriksen Flanker task (Eriksen and Eriksen, 1974b).
Towards personalized assessment of fatigue perpetuating factors in patients with chronic fatigue syndrome using ecological momentary assessment: A pilot study
2021, Journal of Psychosomatic ResearchCitation Excerpt :The standard deviation of activity level was calculated to serve as a measure of fluctuation of activities. In accordance with previous studies, time periods of at least 90 min of no activity were registered as non-wear and recorded as missing data [31,32]. The Actilog was used in CFS/ME research previously, to assess activity in CFS/ME patients and other groups and to distinguish between low active and fluctuating active CFS/ME patients.
Compositional data analysis of 24-hour movement behaviors and mental health in workers
2020, Preventive Medicine ReportsCitation Excerpt :An epoch-length was set at 60-seconds and estimated metabolic equivalents (METs) were obtained using developer-provided software. Non-wear time was defined in intervals of 20 consecutive minutes with activity counts under the detection limit (Peeters et al., 2013) and valid days were defined as days when participants wore the device for ≥ ten hours (Tudor-Locke et al., 2012). Non-wearing time was subtracted from 24 h to obtain wearing time.
Obesity as a moderator of the relationship between neighborhood environment and objective measures of physical activity in chilean adults
2019, Journal of Transport and HealthCitation Excerpt :Data were downloaded and analysed at the individual level using the Actilife 6 Software. An automated algorithm was used to identify wearing and non-wearing periods (i.e., 90 min of consecutive inactivity)(Choi et al., 2011; Peeters et al., 2013). A day was considered valid if the participant wore the device for at least 10 h. Accelerometer data were included in the analyses if participants had valid information at least three weekdays and one weekend day(Sasaki et al., 2018).