Accelerometric estimates of physical activity vary unstably with data handling

• Published in: PloS one Vol. 12; no. 11; p. e0187706
• Main Authors: Smith, Maia P, Standl, Marie, Heinrich, Joachim, Schulz, Holger
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 06.11.2017; Public Library of Science (PLoS)
• Subjects: Accelerometers; Accelerometry - methods; Activity patterns; Adolescent; Adolescents; Biology and Life Sciences; Cohort Studies; Computer and Information Sciences; Data processing; Engineering and Technology; Environmental health; Epidemiology; Estimates; Exercise; Female; Health aspects; Hip; Hospitals; Humans; Luminous intensity; Male; Medicine; Medicine and Health Sciences; Pediatrics; People and Places; Physical activity; Physical fitness; Physical Sciences; Public health; Research and Analysis Methods; Smartphones; Stability analysis; Studies; Teenagers; Valproic acid; Youth; Germany
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0187706

Because of unreliable self-report, accelerometry is increasingly used to objectively monitor physical activity (PA). However, results of accelerometric studies vary depending on the chosen cutpoints between activity intensities. Population-specific activity patterns likely affect the size of these differences. To establish their size and stability we apply three sets of cutpoints, including two calibrated to a single reference, to our accelerometric data and compare PA estimates. 1402 German adolescents from the GINIplus and LISAplus cohorts wore triaxial accelerometers (Actigraph GT3x) for one week (mean 6.23 days, 14.7 hours per day) at the hip. After validation of wear, we applied three sets of cutpoints for youth, including the most common standard (Freedson, 2005) and two calibrated to a single reference, (Romanzini uni- and triaxial, from Romanzini, 2014) to these data, estimating daily sedentary, light, moderate, vigorous and moderate-to-vigorous PA (MPA, VPA, MVPA). Stability of differences was assessed by comparing Romanzini's two sets of cutpoints. Relative agreement between cutpoints was closer for activity of lower intensities (largest difference for sedentary behaviour 9%) but increased for higher intensities (largest difference for light activity 40%, MPA 102%, VPA 88%; all p<0.01). Romanzini's uniaxial and triaxial cutpoints agreed no more closely with each other than with Freedson's. Estimated PA differed significantly between different sets of cutpoints, even when those cutpoints agreed perfectly on another dataset (i.e. Romanzini's.) This suggests that the detected differences in estimated PA depend on population-specific activity patterns, which cannot be easily corrected for: converting activity estimates from one set of cutpoints to another may require access to raw data. This limits the utility of accelerometry for comparing populations in place and time. We suggest that accelerometric research adopt a standard for data processing, and apply and present the results of this standard in addition to those from any other method.