(1) Background. To facilitate accurate actigraphy data analysis, inactive periods have to be distinguished from periods during which the device is not being worn. The current analysis investigates the degree to which off-wrist and inactive periods can be automatically identified. (2) Methods. In
[...] Read more.
(1) Background. To facilitate accurate actigraphy data analysis, inactive periods have to be distinguished from periods during which the device is not being worn. The current analysis investigates the degree to which off-wrist and inactive periods can be automatically identified. (2) Methods. In total, 125 actigraphy records were manually scored for ‘off-wrist’ and ‘inactivity’ (99 collected with the Motionlogger AMI, 26 (sampling frequency of 60 (
n = 20) and 120 (
n = 6) s) with the Philips Actiwatch 2.) Data were plotted with cumulative frequency percentage and analyzed with receiver operating characteristic curves. To confirm findings, the thresholds determined in a subset of the Motionlogger dataset (
n = 74) were tested in the remaining dataset (
n = 25). (3) Results. Inactivity data lasted shorter than off-wrist periods, with 95% of inactive events being shorter than 11 min (Motionlogger), 20 min (Actiwatch 2; 60 s epochs) or 30 min (Actiwatch 2; 120 s epochs), correctly identifying 35, 92 or 66% of the off-wrist periods. The optimal accurate detection of both inactive and off-wrist periods for the Motionlogger was 3 min (Youden’s Index (
J) = 0.37), while it was 18 (
J = 0.89) and 16 min (
J = 0.81) for the Actiwatch 2 (60 and 120 s epochs, respectively). The thresholds as determined in the subset of the Motionlogger dataset showed similar results in the remaining dataset. (4) Conclusion. Off-wrist periods can be automatically identified from inactivity data based on a temporal threshold. Depending on the goal of the analysis, a threshold can be chosen to favor inactivity data’s inclusion or accurate off-wrist detection.
Full article