Advancing accelerometry-based physical activity monitors: quantifying measurement errors and improving energy expenditure prediction

Abstract

As the rate of obesity increases in the western world, the interest in understanding the process of maintaining healthy body weight has become an increasingly important public health priority. Because physical activity is the most variable component of energy expenditure both intra- and inter-individual, it has become a key factor in both individual weight loss prescriptions and public health recommendations. In spite of its widely recognized importance, the ability to accurately quantify patterns of physical activity has been limited by measurement technology that is often unable to render accurate predictions of energy expenditure over the course of days or weeks. One popular measurement tool for quantifying physical activity intensity is the accelerometer. Though the physical basis of the measurement would suggest that accelerometers can predict energy expenditure with a high degree of accuracy, to date, this promise has not been realized. This thesis addresses several critical gaps in our understanding of energy expenditure predictions using accelerometers. Three commercially available, single site accelerometers were coupled with seven regression equations from the literature to predict energy expenditure in a heterogeneous sample of healthy adult volunteers. We explored errors in energy expenditure prediction by both examining the accelerometer hardware as well as by proposing an analytical approach to energy expenditure prediction incorporating high frequency (32 Hz) data collection and artificial neural network modeling. Results of these experiments highlighted limitations in the single site accelerometers both in clinical data and in data collected using mechanically generated accelerations. Additionally we demonstrated improvements in energy expenditure prediction relative to single site accelerometers as well as a multi-site accelerometer array by using an artificial neural network approach for predicting energy expenditure. Results from these studies will be used to better understand the capabilities of accelerometers to assess physical activity in the field environment, which serves to improve our understanding of energy balance and obesity.