Hurricane Sally Update
The ActiGraph office is open, and we have resumed normal operations as of Monday, September 21st. Shipping delays are still possible as our community recovers from the storm. If you need immediate assistance, please contact us by email at firstname.lastname@example.org and we will respond as quickly as possible. Thank you for your continued support.
A Universal Method For Accurate Classification Of Physical Activity And Sedentary Behavior With Triaxial Accelerometry
- Presented on May 29, 2014
Background: A wide spectrum of accelerometers is used for objective assessment of physical activity (PA) and sedentary behavior (SB). However, differences in proprietary analysis algorithms of different accelerometers compromise direct comparison of PA and SB results.
Purpose: To introduce a universally applicable preprocessing and analysis method for raw acceleration data.
Methods: Three waist-worn accelerometers (Actigraph GTX3, GulfCoast X6-1A and Hookie AM13) with different dynamic ranges and sampling frequencies were compared. From the raw triaxial acceleration signal, a resultant was calculated. 21 healthy adults (age range 23 – 57 yrs) underwent a supervised array of 5 types of SB (MET < 1.5) and 5 activities from slow walking to running. Ability of 23 different time and frequency domain traits in separating different PA intensities from each other was evaluated. For speciﬁc analysis of SB (Hookie only), 30 healthy adults (26 – 62 yrs) underwent a supervised array of lying, sitting, and standing periods performed in a random order and duration. Receiver operator characteristics (ROC) analysis was used.
Results: : Mean amplitude deviation (MAD) of the resultant was consistently the best-performing trait in separating PA intensities from each other. Applying the universal cut-off limits from ROC analyses to the MAD data, intensity levels could be separated equally well with three accelerometers. In the pooled analysis, SB was separated from slow walking with 98.7% sensitivity and 99.7% speciﬁcity, slow walking from normal walking with 100% and 100%, brisk walking from normal walking with 98.9% and 96.9%, and jogging and running from brisk walking with 98.3% and 98.8%, respectively. Lying was separated from sitting with 99.5% sensitivity and 99.9% speciﬁcity, while standing was separated from sitting with 93.8% and 92.6%, respectively.
Conclusions: A simply calculable MAD together with universal cut-off limits provides a robust analysis for accurate intensity-speciﬁc evaluation of bipedal PA basically for all triaxial accelerometers irrespective of technical speciﬁcations. Further, the use of triaxial raw data permits a reliable classiﬁcation of SB. A broader application of the present approach is expected to render the results from different accelerometry studies directly comparable with each other.
ACSM 2014 Annual Meeting