Abstract 030: Estimates of Energy Expenditure from an Accelerometer Worn on the Wrist Versus the Hip in Adults

Abstract

Epidemiological researchers have often measured physical activity by placing an accelerometer on the participants’ hip for seven days. Recently, NHANES decided to place the monitor on the wrist to increase participant wear time. This study compared established calibration methods for accelerometers on the hip versus the wrist. Healthy adults (n=37) wore custom accelerometers (Wockets) on the wrist and hip for twenty activities along with a portable metabolic unit (Oxycon Mobile) to measure energy expenditure. The activities included sedentary behaviors, treadmill walking (including uphill), cycling, work or household chores. Data included only metabolic steady state periods of the activities. Each signal from the triaxial accelerometer was filtered and integrated over 1-second intervals to obtain an area under the curve (AUC) summary, and the three axes were summed to create a single motion measure. Because traditional calibration methods do not include a specific method for assessing sedentary behavior, receiver operating characteristic (ROC) curves were used to determine a sedentary cutpoint and an exercise (moderate and vigorous activity over 3.0 METs) cutpoint for both hip and wrist. The ROC for determining sedentary behavior at the hip had greater sensitivity and specificity (71% and 96%) than the wrist (53% and 76%). The ROC for exercise also had higher sensitivity and specificity on the hip (70% and 83%) than the wrist (30% and 69%). The coefficient of variation (CV) in the signal was used to separate ambulatory activities from non-ambulatory activities. Logistic regression determined the optimal CV from the data on the hip for a two-regression model with energy expenditure, but cycling activities prevented a useful prediction model. The ROC for determining the CV associated with ambulation had a larger AUC at the hip compared to the wrist (0.83 and 0.74), and no CV separation between ambulation and non-ambulatory activities existed for the wrist. Therefore, mixed model analysis was used to predict energy expenditure on all non-sedentary activities. The prediction model on the hip resulted in an average difference of 0.17 (+/− 0.99) METs, while the wrist resulted in an average difference of 0.28 (+/− 1.47) METs. Since the CV was not useful for separating activity type at these locations, an analysis was prepared using the known activity separation of sedentary activities, ambulation, and other. On the hip, the mean difference was 0.05 (+/− .92) METs, whereas the wrist was 0.06 (+/− 1.12) METs. In conclusion, methods for estimating energy expenditure from an accelerometer on the hip do not apply to the wrist, and there is greater variation in the estimates from the wrist compared to the hip. There is improvement in energy expenditure estimates if the activity type is known, but from this study, it appears that activity type cannot be determined by CV from the wrist as as it can from the hip.