Validation of a wireless accelerometer network for energy expenditure measurement

Abstract

The purpose of this study was to validate a wireless network of accelerometers and compare it to a hip-mounted accelerometer for predicting energy expenditure in a semi-structured environment. Adults (n = 25) aged 18-30 engaged in 14 sedentary, ambulatory, exercise, and lifestyle activities over a 60-min protocol while wearing a portable metabolic analyser, hip-mounted accelerometer, and wireless network of three accelerometers worn on the right wrist, thigh, and ankle. Participants chose the order and duration of activities. Artificial neural networks were created separately for the wireless network and hip accelerometer for energy expenditure prediction. The wireless network had higher correlations (r = 0.79 vs. r = 0.72, P < 0.01) but similar root mean square error (2.16 vs. 2.09 METs, P > 0.05) to the hip accelerometer. Measured (from metabolic analyser) and predicted energy expenditure from the hip accelerometer were significantly different for the 3 of the 14 activities (lying down, sweeping, and cycle fast); conversely, measured and predicted energy expenditure from the wireless network were not significantly different for any activity. In conclusion, the wireless network yielded a small improvement over the hip accelerometer, providing evidence that the wireless network can produce accurate estimates of energy expenditure in adults participating in a range of activities.

Keywords: Accelerometry; ActiGraph; artificial neural network; machine learning; physical activity.