Comparability of the ActivPAL CREA and GHLA Algorithms for Determining Posture and Physical Activity in Free-Living Adults

Background : The thigh-worn activPAL has been a primary device used for classifying posture (lying, sitting, standing, stepping, and cycling) in laboratory and free-living environments. Occasionally, activPAL releases software-based algorithm updates, but it is not clear if these algorithms produce equivalent outcomes. Purpose: To compare activity and posture outcomes from activPAL’s CREA and GHLA algorithms. Methods : Adult participants in the LIFT trial ( n = 350) provided 6,038 valid days of data from wearing the activPAL3 micro on the right thigh for 7 days on four occasions. Then, CREA and GHLA algorithm daily data were downloaded in the PALbatch software and were compared using correlations, mean absolute differences, mean absolute percent differences, and equivalence testing (within 3%). Results : Point estimate for most activity metrics were similar, though not identical, between the CREA and GHLA algorithms. Correlations between algorithms were excellent ( r = .99–1.00) for all metrics including posture outcomes, sitting, and stepping bouts and for sitting time ( r = .97). Mean absolute differences for all postural outcomes were ≤7.1 min (and mean absolute percent difference ≤3.5%), except for sitting (14 min; mean absolute percent difference 3.2%). Of the 29 compared outcome variables (postural outcomes, sitting bouts, and stepping bouts), 24 were equivalent to within 3%, indicating excellent agreement between the CREA and GHLA algorithms. Conclusions : Most outcome metrics for the thigh-worn activPAL accelerometer calculated using the embedded CREA and GHLA algorithms were highly comparable (within a strict 3% equivalence range) and would permit comparisons across studies regardless of which algorithm was used.