Commercial Devices Provide Estimates of Energy Balance with Varying Degrees of Validity in Free-Living Adults.

Background The challenges of accurate estimation of energy intake (EI) are well-documented, with self-reported values 12%-20% below expected values. New approaches rely on gold-standard assessments of the other components of energy balance, energy expenditure (EE) and energy storage (ES), to estimate EI. Objectives The purpose of this study was to evaluate the validity, repeatability, and measurement error of consumer devices when estimating energy balance in a free-living population. Methods Twenty-four healthy adults (14 women, 10 men; mean ± SD age: 30.7 ± 8.2 y) completed two 14-d assessment periods, including assessments of EE and ES using gold-standard [doubly labeled water (DLW) and DXA] and commercial devices [Fitbit Alta HR activity monitor (Alta) and Fitbit Aria wireless body composition scale (Aria)], and of EI by dietician-administered recalls. Accuracy and validity were assessed using Spearman correlation, interclass correlation, mean absolute percentage error, and equivalency testing. We also applied linear measurement error modeling including error in gold-standard devices and within-subject repeated-measures design to calibrate consumer devices and quantify error. Results There was moderate to strong agreement for EE between the Fitbit Alta and DLW at each time point (rs = 0.82 and 0.66 for Times 1 and 2, respectively). There was weak agreement for ES between the Fitbit Aria and DXA (rs = 0.15 and 0.49 for Times 1 and 2, respectively). Correlations between methods to assess EI ranged from weak to strong, with agreement between the DXA/DLW-calculated EI and dietary recalls being the highest (rs = 0.63 for Time 1 and 0.73 for Time 2). Only EE from the Fitbit Alta at Time 1 was equivalent to the DLW value using equivalency testing. Conclusions Commercial devices provide estimates of energy balance in free-living adults with varying degrees of validity compared to gold-standard techniques. EE estimates were the most robust overall, whereas ES estimates were generally poor.