Comparison of Accelerometry Methods for Estimating Physical Activity

    loading  Checking for direct PDF access through Ovid



This study aimed to compare physical activity estimates across different accelerometer wear locations, wear time protocols, and data processing techniques.


A convenience sample of middle-age to older women wore a GT3X+ accelerometer at the wrist and hip for 7 d. Physical activity estimates were calculated using three data processing techniques: single-axis cut points, raw vector magnitude thresholds, and machine learning algorithms applied to the raw data from the three axes. Daily estimates were compared for the 321 women using generalized estimating equations.


A total of 1420 d were analyzed. Compliance rates for the hip versus wrist location only varied by 2.7%. All differences between techniques, wear locations, and wear time protocols were statistically different (P < 0.05). Mean minutes per day in physical activity varied from 22 to 67 depending on location and method. On the hip, the 1952-count cut point found at least 150 min·wk−1 of physical activity in 22% of participants, raw vector magnitude found 32%, and the machine-learned algorithm found 74% of participants with 150 min of walking/running per week. The wrist algorithms found 59% and 60% of participants with 150 min of physical activity per week using the raw vector magnitude and machine-learned techniques, respectively. When the wrist device was worn overnight, up to 4% more participants met guidelines.


Estimates varied by 52% across techniques and by as much as 41% across wear locations. Findings suggest that researchers should be cautious when comparing physical activity estimates from different studies. Efforts to standardize accelerometry-based estimates of physical activity are needed. A first step might be to report on multiple procedures until a consensus is achieved.

Related Topics

    loading  Loading Related Articles