Predicting Oxygen Uptake From Accelerometer Output In Adults With Down Syndrome: Vector Magnitude Vs. Vertical Axis

Abstract

Adults with Down syndrome (DS) have altered movement patterns. Especially during walking, their altered mediolateral and anteroposterior body motion predicts their elevated energy cost. Triaxial accelerometers provide a metric of three-dimensional acceleration—Vector Magnitude (VM) counts—which may better estimate the rate of oxygen uptake (VO2) during physical activities and sedentary behaviors than the traditionally used Vertical Axis (VA) counts. PURPOSE: To examine if VM counts are more accurate than VA counts in estimating VO2 across different physical activities and sedentary behaviors in adults with DS. METHODS: Sixteen adults with DS (10 men; age 31 ± 15 years) performed 12 tasks: sitting; playing app; drawing; folding clothes; sweeping; fitness circuit; moving box; basketball; standing; and walking at the preferred speed and at 0.8 and 1.4 m.s-1. We measured VO2 with a spirometer (K4b2, Cosmed) and VA and VM with an accelerometer (wGT3X-BT, Actigraph) on the non-dominant hip. We used two separate multi-level regression models to predict VO2 from VA or VM. We evaluated the fit of models with the R2, and accuracy with Bland-Altman plots and absolute percent error which was compared between models across tasks using within-subject (method-by-task) ANOVA and follow-up paired-samples t-tests. RESULTS: Both VM and VA significantly predicted VO2 in separate models (p <0.001; R2 = 0.74 and 0.65, respectively). Across all tasks combined, absolute percent error was lower for the VM than the VA model (23.7 ± 26.2 and 33.6 ± 35.9, respectively). A significant method-by-task interaction in within-subject ANOVA and follow-up t-tests indicated that absolute error was lower for the VM than the VA model for sitting, playing an app, drawing, and standing (p ≤0.004), but did not differ for other tasks. Bland-Altman plots indicated zero mean error for both models; however, the limits of agreement were narrower for the VM than the VA model (-6.44 to 6.44 and -5.57 to 5.57 ml.kg-1.min-1, respectively). CONCLUSION: Both VA and VM counts predict VO2 in adults with DS; however, prediction is more accurate for a VM than a VA model during sedentary behaviors. VM counts should be used in developing accelerometer-based prediction of physical activity and sedentary behavior in adults with DS. Supported by NIH Grant R15HD098660