The Differences in the Evaluation of Accelerometer-Based Physical Activity with GGIR and ActiLife

Background: GGIR is an open-source code usable with the software RStudio. The difference in the evaluation process between GGIR and other evaluation methods is the use of the Euclidian norm minus one (ENMO). The benefit of ENMO and GGIR is that they can be used no matter the accelerometer brand which makes processed accelerometer data more comparable. This study compares the evaluation of the accelerometry data of the PACINPAT study sample with the GGIR package and ActiLife.
Methods: A randomized controlled sample aged 18-65 years diagnosed with major depressive disorder were examined 3 times over 1 year regarding their physical and mental health. To assess their everyday physical activity objectively they were given an accelerometer to wear for 7 continuous days. The accelerometer was set to record in 60Hz frequency. GGIR summarized the data into 5 seconds epochs, ActiLife worked with 60 seconds epoch. The data from the sensors was processed with ActiLife and the GGIR package. The metric used in the GGIR package was ENMO, ActiLife evaluated with counts per minute (cpm). Both tools are working with cut-points to classify the recorded data into the categories light (GGIR: 47mg, ActiLife: 611cpm), moderate (GGIR: 70mg, ActiLife: 2690cpm) and vigorous (GGIR: 260mg, ActiLife: 6167cpm) physical activity, and inactive (GGIR: <47mg, ActiLife: <611). The intensity gradient was calculated with the GGIR package.
Results: There was a positive correlation between the two evaluation tools ranging from weak to strong according to Pearson’s correlation coefficient. The absolute values derived from the two evaluation tools using different cut-points are never within 10% of each other.
Conclusion: The choice of epoch length and cut-points are responsible for the high discrepancy between the processing methods. It is therefore not possible to compare the two tools to one another if the chosen method for processing is cut-points. This calls for a cut-point-free method to evaluate and compare accelerometry data. The intensity gradient is a promising candidate to fill this role as it is not susceptible to the same research bias as cut-points are. The GGIR package supports the calculations of the intensity gradient.