Comparison Between The Fitbit Flex And Gt3x+ Activity Monitor Under Free-living Settings

Accelerometry-based activity monitors appear to be valid for measuring habitual physical activity (PA) and sedentary behavior. However, the accuracy of consumer-based activity monitors for measuring physical activity and sedentary behavior has not been well established in free-living conditions. PURPOSE: To evaluate the accuracy of the Fitbit Flex (FB) activity monitor against the previously validated accelerometer, ActiGraph GT3X+ (AG), in free-living conditions. METHODS: 68 healthy adults (41 ± 14 yr, female = 74%, body mass index = 26 ± 5) wore FB on non-dominant wrist and AG on dominant hip for seven consecutive days except during bathing and water activities. Sleep log and ActiLife software (Choi 2011 criteria) were used to define sleep and non-wear time periods, respectively. After excluding sleep and non-wear time, times spent (min/day) sedentary (SED), light PA (LPA), and moderate to vigorous PA (MVPA) were estimated using Freedson’s (1998) cut points for AG and using brand-specific algorithms for FB. Paired t-tests were used to determine the mean difference between FB and AG across the estimates of SED, LPA, and MVPA. Pearson correlations and BA plots were used to evaluate the agreement between the estimates from FB and those from AG. RESULTS: The estimates of SED were not statistically different (mean difference: -19 min, p = .18), but the estimates of LPA (mean difference: 116 min, p <.05) and MVPA (mean difference: -82 min, p <.05) were significantly different between the AG and the FB. The significant correlations between AG and FB were the strongest for SED (r = .96, p < .01), and the weakest for MVPA (r = .27, p < .05). The Bland-Altman plot mean difference was -19 min (limit of agreement [LOA] -244 to 206), 116 min (LOA 27 to 205), and -82 min (LOA -182 to 19) for SED, LPA, and MVPA, respectively. CONCLUSION: We found the FB provided comparatively accurate estimates of SED under free-living conditions. However, the estimates of PA were not comparable between the FB and the AG. It is important to note that PA estimates from the AG varied depending on the cut points applied to raw data. Future investigations should consider the effect of cut points when comparing the agreement of the PA estimates between FB and AG.