E-16 Thematic Poster - Predicting Energy Expenditure: JUNE 3, 2011 8:00 AM - 10:00 AM: ROOM: 404
The Fitbit activity monitor is a consumer device for predicting energy expenditure (EE) and tracking activity patterns. To improve the EE estimation accuracy, Fitbit provides a web-based software program that allows a user to classify periods as distinct activities (based on the compendium of physical activities). However, the effect of this manual activity classification on EE estimation accuracy is not known.
PURPOSE: To compare the EE prediction accuracy of the Fitbit before and after classifying activities.
METHODS: Fifteen adults (7 male), 71.91(17.3) kg, 24.17(3.9) kg/m2, completed a four hour stay in a room calorimeter. Participants wore Fitbit on the right hip, and performed a series of randomly assigned activities/postures including supine, seated (quietly and using a computer), standing, walking, stepping, cycling, sweeping, and self-selected activities. We used the web-based software to classify each activity, and compared the estimated EE to the measured EE before and after activity classification.
RESULTS: Without activity classification, Fitbit significantly underestimated EE (368(18) vs. 499(24)kcal, mean(SE)). Classifying activities resulted in improved estimates of EE (516(13) vs. 499(24)kcal, mean(SE)). Root mean square error for non-classified EE was 136.7kcals (27.4%) and was reduced to 64.25kcals (12.9%) with activity classification. The non-classification estimates always underestimated EE, while the classified values were underestimated about half of the time, and were more accurate in all but two subjects.
CONCLUSIONS: Fitbit is most accurate when the time is taken to classify the activities that were performed while wearing the device, though this may not be practical for the average consumer.