Skip to main content
ARS Home » Research » Publications at this Location » Publication #183819

Title: Estimates of adherence and error analysis of physical activity data collected via accelerometry in a large study of free-living adults.

Author
item PAUL, DAVID - JHU
item Kramer, Matthew
item Stote, Kim
item Spears, Karen
item Moshfegh, Alanna
item Baer, David
item Rumpler, William

Submitted to: BMC Medical Research Methodology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/9/2008
Publication Date: 6/9/2008
Citation: Paul, D., Kramer, M.H., Stote, K.S., Spears, K.E., Moshfegh, A.J., Baer, D.J., Rumpler, W.V. 2008. Estimates of adherence and error analysis of physical activity data collected via accelerometry in a large study of free-living adults. BMC Medical Research Methodology. 8:38.

Interpretive Summary: Activity monitors are small, electronic devices used to quantify the amount and intensity of physical activity. Since this is a new technology, not much is known about how to process and interpret the data. Although activity monitors are well-tolerated by subjects, they are commonly removed by subjects during the day and lead to blocks of missing data. This investigation indicates that the missing data due to monitor removal can lead to biased measurements of physical activity. Fortunately, some of these errors can be minimized by imputation. The study indicates that activity monitor data must be treated as a 24 hr day, regardless of how long the monitor is actually worn by the subject. This study will be useful to scientists that utilize activity monitors in studies interested in assessing habitual physical activity.

Technical Abstract: Activity monitors (AM) are small, electronic devices used to quantify the amount and intensity of physical activity (PA). One limitation of this method is the periodic loss of data due to AM removal during waking and sleeping hours, potentially producing a bias in the estimation of PA. To estimate the effect of AM removal, the results of thirty-three subjects that wore an AM for at least 20 hrs/day for seven consecutive days were analyzed. The loss of a single hour of data biased estimates of PA (p < 0.05). Simulating AM removal due to sleep also affected estimates of PA (p < 0.05), but the error could be reduced by using imputation. Despite a 5% increase in the CV, the loss of an entire day did not significantly affect the group estimate of PA and imputation did improve the estimation. Studies using AM's should treat the data as a 24 hr day, regardless of the length of wear, and use imputation for missing data. Investigators should report missing data and describe their procedures for processing them.