Is the pain of activity log-books worth the gain in precision when distinguishing wear and non-wear time for tri-axial accelerometers?

OBJECTIVE: To compare three methods for assessing wear time from accelerometer data: automated, log-books and a combination of the two.
DESIGN: Cross-sectional study.
METHODS: Forty-five office workers wore an Actigraph GT3X accelerometer and kept a detailed activity log-book for 7 days. The automated method used six algorithms to determine non-wear time (20, 60, or 90 min of consecutive zero counts with and without 2-min interruptions); the log-book method used participant recorded on/off times; the combined method used the 60-min automated filter (with ≤2 min interruptions) plus detailed log-book data. Outcomes were number of participants with valid data, number of valid days, estimates of wear time and time spent in sedentary, light, moderate and vigorous activity. Percentage misclassification, sensitivity, specificity, and area under the receiver-operating curve were compared for each method, with the combined method as the reference.
RESULTS: Using the combined method, 34 participants met criteria for valid wear time (≥10 h/day, ≥4 days). Mean wear times ranged from 891 to 925 min/day and mean sedentary time s from 438 to 490 min/day. Percentage misclassification was higher and area under the receiver-operating curve was lower for the log-book method than for the automated methods. Percentage misclassification was lowest and area under the receiver-operating curve highest for the 20-min filter without interruptions, but this method had fewer valid days and participants than the 60 and 90-min filters without interruptions.
CONCLUSIONS: Automated filters are as accurate as a combination of automated filters and log-books for filtering wear time from accelerometer data. Automated filters based on 90-min of consecutive zero counts without interruptions are recommended for future studies.