Combined influence of epoch length, cut-point and bout duration on accelerometry-derived physical activity

Abstract

BackgroundIt is difficult to compare accelerometer-derived estimates of moderate-to-vigorous physical activity (MVPA) between studies due to differences in data processing procedures. We aimed to evaluate the effects of accelerometer processing options on total and bout-accumulated time spent in MVPA in adults.Methods267 participants from the ProActive Trial provided 1236 days of valid physical activity (PA) data, collected using a 5-s epoch with ActiGraph GT1M accelerometers. We integrated data over 5-s to 60-s epoch lengths (EL) and applied two-level mixed effects regression models to MVPA time, defined using 1500 to 2500 counts/minute (cpm) cut-points (CP) and bout durations (BD) from 1 to 15 min.ResultsTotal MVPA time was lower on longer EL and higher CP (47 vs 26 min/day and 26 vs 5 min/day on 1500 vs 2500 cpm on 5-s and 60-s epoch, respectively); this could be approximated as MVPA = exp[2.197 + 0.279*log(CP) + 6.120*log(EL) - 0.869*log(CP)*log(EL)] with an 800 min/day wear-time. In contrast, EL was positively associated with time spent in bout-accumulated MVPA; the approximating equation being MVPA = exp[54.679 - 6.268*log(CP) + 6.387*log(EL) - 10.000*log(BD) - 0.162*log(EL)*log(BD) - 0.626*log(CP)*log(EL) + 1.033*log(CP)*log(BD)]. BD and CP were inversely associated with MVPA, with higher values attenuating the influence of EL.ConclusionsEL, CP and BD interact to influence estimates of accelerometer-determined MVPA. In general, higher CP and longer BD result in lower MVPA but the direction of association for EL depends on BD. Reporting scaling coefficients for these key parameters across their frequently used ranges would facilitate comparisons of population-level accelerometry estimates of MVPA.

Substituting Sedentary Time With Light and Moderate to Vigorous Physical Activity is Associated With Better Cardiometabolic Health

“…This threshold corresponds with the more often used 600 min/day threshold for accelerometerbased monitors that are worn during waking hours only [18,19]. Furthermore, to achieve reliable estimates of total physical activity, only subjects who met the compliance criterion of at least 3 weekdays and both weekend days were admitted, resulting in the exclusion of 207 (32%) of the original 652 participants [20].…”

A machine learning approach to measure and monitor physical activity in children

“…Children completed seven different physical activities performed in a randomised order, which took place in the school playground or classroom with 5 minutes seated rest between each activity. To capture both the sporadic nature of children's activity [38] and locomotive movement best suited to accelerometers [39], the activities incorporated both intermittent and continuous (i.e., walking and jogging) movements representative of culturally-relevant-free-play situations. Sedentary activities were watching a DVD and drawing, which were consistent with those used previously [26].…”

A Machine Learning Approach to Measure and Monitor Physical Activity in Children to Help Fight Overweight and Obesity

“…Children completed seven different physical activities performed in a randomized order, which took place in the school playground or classroom as appropriate with 5 minutes seated rest between each activity. To capture both the sporadic nature of children's activity (Orme et al, 2014) and locomotive movement best suited to accelerometers (Welk, 2005), the activities incorporated both intermittent and continuous (i.e., walking and jogging) movements representative of culturally-relevant-free-play situations. Children who were performing sedentary activities were watching a DVD and drawing, which were consistent with those used previously (Evenson, Cattellier, Gill, Ondrak, & McMurray, 2008).…”