Extent and Impact of Accelerometer Time Drift on Linkage with GPS Data for Physical Activity Research
Topics: Medical and Health Geography
,
,
Keywords: Accelerometer time lag, GPS and accelerometer linkage, time spent outdoor, ActiGraph wGT3X-BT
Session Type: Virtual Paper
Day: Sunday
Session Start / End Time: 4/11/2021 09:35 AM (Pacific Time (US & Canada)) - 4/11/2021 10:50 AM (Pacific Time (US & Canada))
Room: Virtual 8
Authors:
Wei Liu, Michigan State University
Kimberly A. Clevenger, National Cancer Institute, National Institutes of Health
Amber L. Person, Michigan State University
Karin Allor Pfeiffer, Michigan State University
,
,
,
,
,
,
Abstract
Purpose: Linkage of GPS and accelerometer data provides rich contextual information about physical activity (PA) in health research. A recent study indicated that recorded accelerometer time may drift over the wear period. This study aimed to 1) confirm the findings from Steel et al. (2019); 2) identify if time drift is consistent between devices and/or data collection waves; and 3) conduct a sensitivity analysis to assess time drift’s impact on PA research. Methods: Eleven ActiGraph wGT3X-BT accelerometers were shaken every six hours either automatically or manually over two waves for 10 days each. Accelerometer time was compared to UTC to identify drifts. A researcher wore all eleven accelerometers and a GPS tracker simultaneously for seven days, visiting a park and recorded the actual start and end time of PA every day. The sensitivity analysis tested three scenarios on time spent doing PA in the park after linking GPS and adjusted accelerometer data: 1) adjusting each accelerometer’s time using observed drifts, 2) adjusting by overall mean drift, and 3) no adjustment (status quo). Results & conclusion: Our results were consistent with the existing study. There was an average of 6.06 seconds of lag in accelerometer time over 10 days. Time drifts varied largely among accelerometers (sd = 5.13 on day 10) but no significant difference between waves (p = 0.19). Our sensitivity analysis indicated that research requiring very high temporal resolution needs to assess the time drifts of individual accelerometers before data collection and linkage to GPS data.