Free-Living Validation and Harmonization of 10 Wearable Step Count Monitors

Abstract

ABSTRACT This study aimed 1) to determine the step count accuracy of numerous wrist-, hip-, and thigh-worn consumer and research monitors (and their corresponding algorithms) compared with the StepWatch (SW) across all waking hours under free-living conditions and 2) to develop correction methods to calibrate all monitors to the SW. Forty-eight participants 28 ± 12 yr old (mean ± SD) wore monitors across two waking days. Different wrist (Apple Watch Series 2, Fitbit Alta, Garmin vivofit 3, and ActiGraph GT9X), hip (Yamax Digiwalker SW-200, Omron HJ-325, GT9X, and Fitbit Zip), and thigh (activPAL) monitors were worn across 2 d, with the exception of the SW, which was worn on both days. Monitor estimates were compared with SW to compute percent of SW steps, absolute percent error, mean difference, root-mean-square error, and Pearson correlations. Monitor-specific correction factor linear regression models were fit to estimate SW steps and evaluated using leave-one-subject-out cross validation. All monitors were significantly different from the SW (P < 0.05). Consumer wrist and hip monitors underestimated SW steps (72%–91% of SW steps per day), whereas step estimates from research monitors ranged widely (67%–189%). Mean absolute percent error across all devices were greater than 10%. After a correction method was applied, all monitor estimates were not significantly different from SW steps. Overall, some consumer monitors produced step estimates that are closer to the validated SW than research-grade monitors (and their corresponding algorithms) and could be used to measure steps for healthy adults under free-living conditions. The specific correction methods may facilitate comparisons across studies and support research efforts using consumer and research monitors for large-scale population surveillance and epidemiological studies.