Processing accelerometer data of children and adolescents - influence of wear time algorithm, epoch length and cut-off points on age specific subgroups

INTRODUCTION: In order to determine physical activity (PA) in free-living conditions the most common used objective method is accelerometry. Large-scale epidemiologic studies frequently use “counts” derived from the ActiGraph device to classify PA. The PA outcome is influenced by different parameters set during data collection and data processing. The aim of this study was to analyse the effect of epoch length (EL), wear time (WT) algorithm and cut-off points (CP) on PA outcome for age specific subgroups within the sample of the Motorik-Modul-Study (MoMo) (Woll et al., 2017).
METHODS: 1,848 children and adolescents between the age of 6 and 17 (N(male) = 901, N(female)=947) wore either ActiGraph GT3x+ or ActiGraph wGT3X-BT as part of the examination throughout the nationwide MoMo survey “Wave 2” (2014-2017). The participants wore the device during all waking hours on 8 consecutive days. The device was placed laterally at the hip and data was sampled with a frequency of 30Hz. Afterwards data was downloaded and converted into one second epochs using ActiLife. Further data processing was done in MATLAB. Data was reintegrated into 5, 15 and 60 second epochs and Troiano (Troiano et al., 2008) and Choi (Choi et al., 2011) WT algorithms were applied. ... mehrPA outcome was determined using different CPs commonly used for the specific age groups (Evenson et al., 2008, Freedson et al., 2005, Hänggi et al., 2013, Romanzini et al., 2014).
RESULTS: In contrast to the results found in existing literature the MoMo data shows no effect of EL on activity count outcome. Data shows that EL does affect total WT when using the Troiano algorithm but does not affect total WT when using the Choi algorithm. Applying 8hWT-criterion leads to 13% more valid data sets than 10h-WT-criterion. EL and CP have a significant effect on PA classification leading to differences in the prevalence of adherence to the PA recommendations by the WHO of up to 50% for certain age groups.
CONCLUSION: The discrepancy between results from this study and existing literature (Logan et al., 2016) on the effect of EL on activity count outcome could be due to the calculation of the Vector Magnitude (VM) when integrating data. Choi WT algorithm should be preferably used over Troiano algorithm when data is reintegrated and compared to different ELs. The choice of EL, WT and CP influences PA outcome significantly and have to be carefully investigated before reporting population-based results