Evaluation Of Garmin And Fitbit Algorithms To Automatically Detect Children's Semi-structured Activity Type And Duration

Consumer wearables with automatic activity detection algorithms have the potential to improve the assessment of children’s physical activity energy expenditure. PURPOSE: This study will evaluate the ability of Garmin Move IQ and Fitbit SmartTrack automatic activity detection algorithms to detect children’s activity type and duration compared to direct observation (DO). METHODS: Children (n = 17, 11.0 ± 1.6 years, 58.8% female) completed a ~ 60-minute, semi-structured activity protocol while wearing a Garmin Vivoactive 4S and a Fitbit Sense concurrently on their nondominant wrist. The watches were counterbalanced with each protocol, and children’s activities were measured via DO. The protocol was counterbalanced and consisted of 3 activities (walking, running [run 10 meters following the sound of an audio recording], and outdoor biking), each separated by seated rest and lasting ≥10 minutes. Activity type and duration data from Garmin and Fitbit were exported via Fitabase and translated into 1-second epochs. Epoch-by-epoch (i.e., accuracy) and measurement agreement (i.e., absolute error) analyses were conducted to evaluate the performance of Garmin Move IQ and Fitbit SmartTrack for detecting children’s activity type and duration compared to DO. RESULTS: Garmin Move IQ correctly predicted biking 71.9% (95%CI = 54.9%, 88.8%), running 5.8% (95%CI = -5.6%, 17.2%), and walking 47.6% (95%CI = 25.6%, 69.5%) of the time, while Fitbit SmartTrack correctly predicted biking 63.9% (95%CI = 43.5%, 84.3%), running 82.1% (95%CI = 67.3%, 97.0%), and walking 49.1% (95%CI = 28.9%, 69.4%) of the time. For duration, Fitbit SmartTrack produced lower mean absolute errors (biking = 3.2 ± 4.6 minutes, running = 3.7 ± 7.8, walking = 10.5 ± 13.1) than Garmin Move IQ (biking = 5.3 ± 4.6 minutes, running = 9.5 ± 2.4, walking = 16.6 ± 10.3). CONCLUSIONS: Preliminary analyses suggest Move IQ and SmartTrack algorithms have low overall performance for detecting walking, biking, and running activities and duration among children. Supported by: Research reported in this abstract was supported in part by the NIDDK of the National Institutes of Health under Award Numbers R01DK129215 (Weaver) and F31DK136205 (White). Co-author Finnegan was supported by the NIGMS Award Numbers T32GM081740 and T32GM145226. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. James White was also supported in part by a SPARC Graduate Research Grant from the Office of the Vice President for Research at the University of South Carolina.