Impact of accelerometer data processing decisions on the sample size, wear time and physical activity level of a large cohort study

Abstract

Background: Accelerometers objectively assess physical activity (PA) and are currently used in several large-scale epidemiological studies, but there is no consensus for processing the data. This study compared the impact of wear-Time assessment methods and using either vertical (V)-Axis or vector magnitude (VM) cut-points on accelerometer output. Methods: Participants (7,650 women, mean age 71.4 y) were mailed an accelerometer (ActiGraph GT3X+), instructed to wear it for 7 days, record dates and times the monitor was worn on a log, and return the monitor and log via mail. Data were processed using three wear-Time methods (logs, Troiano or Choi algorithms) and V-Axis or VM cut-points. Results: Using algorithms alone resulted in "mail-days" incorrectly identified as "wear-days" (27-79% of subjects had >7-days of valid data). Using only dates from the log and the Choi algorithm yielded: 1) larger samples with valid data than using log dates and times, 2) similar wear-Times as using log dates and times, 3) more wear-Time (V, 48.1 min more; VM, 29.5 min more) than only log dates and Troiano algorithm. Wear-Time algorithm impacted sedentary time (∼30-60 min lower for Troiano vs. Choi) but not moderate-To-vigorous (MV) PA time. Using V-Axis cut-points yielded ∼60 min more sedentary time and ∼10 min less MVPA time than using VM cut-points. Conclusions: Combining log-dates and the Choi algorithm was optimal, minimizing missing data and researcher burden. Estimates of time in physical activity and sedentary behavior are not directly comparable between V-Axis and VM cut-points. These findings will inform consensus development for accelerometer data processing in ongoing epidemiologic studies.