Methods to Use Big Wearable Heart Rate Data for Estimation of Physical Activity in Population Level

摘要

Technologies for wearable health monitoring are becoming increasingly popular and affordable. As a result, large-scale health databases from a large number of individuals are becoming available. However, analysis of these databases requires special methodology to transform available parameters into more generic ones and to manage such non-balanced data characteristics as biases and sampling issues. In this paper, we introduce a methodology for studying physical activity from big wearable heart rate (HR) data on about 5 000 working-age individuals, each measured only for a few days. Physical activity was assessed by oxygen consumption (V02) calculated from measured HR data using a neural network model. Minute-to-minute V02 data was used to quantify various physical activities in a measurement day, as defined according to the health promoting physical activity minutes of the American College of Sports Medicine. We set a posteriori inclusion criteria for the data on the subjects' personal background parameters and the quality of their HR data. The effect of different subjects being measured in different months and weekdays was removed by using a linear model. The linear model sought to estimate the physical activity minutes based on a subject's background parameters. The results show that big data collected in real-life settings and originally for non-research purposes can with appropriate data management and analysis methodology provide unique knowledge of lifestyles and behavior.