To enable inter- and intrastudy comparisons it is important to ascertain comparability among accelerometer models.
Purpose: The purpose of this study was to compare raw and count data between hip-worn ActiGraph GT3X+ and GT9X Link accelerometers.
Methods: Adults (n = 26 (n = 15 women); age, 49.1 ± 20.0 yr) wore GT3X+ and Link accelerometers over the right hip for an 80-min protocol involving 12-21 sedentary, household, and ambulatory/exercise activities lasting 2-15 min each. For each accelerometer, mean and variance of the raw (60 Hz) data for each axis and vector magnitude (VM) were extracted in 30-s epochs. A machine learning model (Montoye 2015) was used to predict energy expenditure in METs from the raw data. Raw data were also processed into activity counts in 30-s epochs for each axis and VM, with Freedson 1998 and 2011 count-based regression models used to predict METs. Time spent in sedentary, light, moderate, and vigorous intensities was derived from predicted METs from each model. Correlations were calculated to compare raw and count data between accelerometers, and percent agreement was used to compare epoch-by-epoch activity intensity.
Results: For raw data, correlations for mean acceleration were 0.96 ± 0.05, 0.89 ± 0.16, 0.71 ± 0.33, and 0.80 ± 0.28, and those for variance were 0.98 ± 0.02, 0.98 ± 0.03, 0.91 ± 0.06, and 1.00 ± 0.00 in the X, Y, and Z axes and VM, respectively. For count data, corresponding correlations were 1.00 ± 0.01, 0.98 ± 0.02, 0.96 ± 0.04, and 1.00 ± 0.00, respectively. Freedson 1998 and 2011 count-based models had significantly higher percent agreement for activity intensity (95.1% ± 5.6% and 95.5% ± 4.0%) compared with the Montoye 2015 raw data model (61.5% ± 27.6%; P < 0.001).
Conclusions: Count data were more highly comparable than raw data between accelerometers. Data filtering and/or more robust raw data models are needed to improve raw data comparability between ActiGraph GT3X+ and Link accelerometers.