Assessing bouts of activity using modeled clinically validated physical activity on commodity hardware

Human activity can be measured through identification of bouts of activity. The Freedson cut point method used by ActiGraph has become one very common and well accepted standard for estimating times of continuous moderate to vigorous physical activity (MVPA). However, such methods do not directly apply to other data sources such as the Fitbit Flex, a wrist worn wireless pedometer. In previous research by the authors, a model was presented to improve the estimates of physical activity (PA) level in the Fitbit devices. This paper considers the estimates of activity bouts, building on the modeled PA level from the Fitbit Flex as compared to the results from the ActiGraph GT3X. The purpose of this paper is to compare the “gold standard” ActiGraph to modeled Fitbit Freedson methods and to establish normative values of expected errors in bout detection between the two devices and methods, both of which are proxy methods aimed at measuring actual physical activity levels. Here we compare bout identification using three measures, the ActiGraph Freedson method, Fitbit Intensity Score, and the modeled Fitbit Freedson using three different outcomes. First, we compare a baseline of per subject per day number and duration of bouts from an ActiGraph GT3X to the results found from using the same methods on the Intensity Score reported by Fitbit and the modeled Fitbit Freedson method. Next, we compare the difference in duration of bouts identified in each data source matched according to similar start and end times. Finally, we compare the bouts found from the three methods to bouts identified in a self report diary.