Responsiveness Of Hip And Wrist Accelerometer-based Physical Activity Estimates In Youth

Measurement of change in physical activity (PA) between time points is important for informing PA interventions and outcomes. The ability to detect change over time, or responsiveness, has received minimal attention in wearable activity monitors, especially compared with a criterion measure. PURPOSE: To determine the responsiveness of accelerometer-based methods located at the hip and wrist to predict PA intensity in youth, compared to a criterion measure. METHODS: Forty-eight participants (11.8 + 3.3 yrs, 23 males) performed two 2-hr free-living visits. Participants wore a portable indirect calorimeter (Cosmed K4b2 or K5) and ActiGraph GT9X devices (right and left wrists, right hip). For each visit, time spent in sedentary behavior (SB, <1.5 METs), light PA (LPA, 1.5-2.99 METs), and moderate to vigorous PA (MVPA, >3.0 METs) was calculated using indirect calorimetry and estimated using three hip-based regression methods and seven wrist-based cut-point methods. Change in time spent in each intensity level between visits was used as the measure of responsiveness. One-way repeated-measures ANOVAs and mean absolute error were used to compare responsiveness reported by each method to indirect calorimetry. RESULTS: There were no statistically significant differences between indirect calorimetry and the hip and wrist methods for mean responsiveness in SB (F = 0.17, p = 0.99), LPA (F = 1.00, p = 0.43), and MVPA (F = 1.16, p = 0.31). Although there were no group-level differences, individual errors ranged from 0.0-67.5% across the different intensity categories. CONCLUSIONS: The responsiveness of the accelerometer-based PA estimation algorithms in this analysis was similar to the change demonstrated by indirect calorimetry at a group level but demonstrated meaningful individual variability, indicating a potential washout effect. At a group level, the estimates appropriately responded to changes in youth PA levels, but further investigation is needed to assess individual-level applicability. Supported by: This research was supported by the NIH grant R01HD083431.