Accuracy of Energy Expenditure Predictions and Activity Identification in Consumer-Based Activity Monitors

INTRODUCTION: Consumer-based physical activity (PA) monitors are increasingly common, and must be validated against criterion measures to determine which models are accurate. Such studies will aid PA intervention research and individual consumers purchasing these devices. METHODS: Thirty participants (mean ±[plus or minus] SD; age, 25.5 ± 3.7 years; BMI, 24.9 ± 2.6 kg/m2[meters squared]) completed a structured PA routine including 11 activities ranging from sedentary behaviors to vigorous intensities. During the routine participants wore an Oxycon portable calorimeter (criterion measure of energy expenditure (EE)), a Basis Peak and Garmin Vivofit on the non-dominant wrist, and three Withings Pulse devices (right hip, shirt collar, dominant wrist). Two repeated measures ANOVAs were used to examine differences between the Oxycon and predicted EE from each monitor, and also examine differences between three Withings placements. Intraclass correlation coefficients (ICC) was calculated to determine reliability of EE predictions between Withings placement sites. Paired samples T-tests were used to determine mean differences between directly observed minutes of structured walking, running, and cycling compared to Basis Band predictions. RESULTS: The Basis Peak was the only device not significantly different from measured gross EE for the entire PA routine (P>[is greater than]0.05), however it had large individual error (95% prediction interval, -290.4 to +233.1 kilocalories (kcals)). All devices were significantly different from measured EE for at least eight individual activities (P<[is less than]0.05); Basis (mean error range: 0.4-24.9 kcals, 3.1%-92.8%), Garmin (0.65-32.5 kcals, 4.3%-78.4%), Withings wrist (0.8-34.7 kcals, 5.4%-69.8%), Withings collar (0.6-29.0 kcals, 4.6%-69.9%), and Withings hip (0.9-29.0 kcals, 6.5%- 69.9%). Withings ICC ranged from 0.085-0.558. The Basis Peak correctly identified ≥[is greater than or equal to] 92% of directly observed minutes during treadmill walking, over-ground walking, and over-ground running (P>0.05). However, only 40.4% of overground cycling minutes were correctly identified and no stationary cycling minutes were identified (P<0.001). CONCLUSION: The Basis Peak had the most accurate EE predictions, on average, for the entire PA routine. The Withings Pulse hip and shirt collar predictions are most similar, but inaccurate compared to the criterion. The Basis Peak activity identification function accurately predicts minutes spent walking and running, but not cycling.