Evaluation Of Wrist-based Accelerometer Intensity Classifications During Activities Of Daily Living In A Whole-room Calorimeter

Abstract

Objective measurements of physical activity (PA) from wrist-worn accelerometers are important for understanding the relationship between PA and disease risk. While locating activity monitors on the wrist improves participant compliance relative to hip-worn protocols, valid methods for classifying PA intensity from the wrist are still in their infancy. PURPOSE Evaluate intensity classification algorithms for wrist-worn accelerometers against a criterion measure during activities of daily living (ADL) and a treadmill walk. METHODS Informed consent was obtained from 7 males (mean ± SD; age 28 ± 3 y; body mass 86.0 ± 15.2 kg; height 178.5 ± 9.6 cm) who completed 1-hr of ADLs consisting of four 10-min tasks separated by 5 min seated rest, and a 30 min treadmill walk at 1.3 m·s-1. For all activities, an ActiGraph GT9X (AG) was worn on the non-dominant wrist and oxygen consumption (VO2, ml/kg) was measured by a whole-room calorimeter (RC). The Hildebrand 2014 and Montoye 2020 wrist classification algorithms were applied to raw AG data. The number of minutes in each intensity category for AG data and RC data were calculated. Metabolic equivalent (MET) categories were defined as: sedentary (sed): < 1.5 METs; light (LPA): 1.5-2.9 METs; and moderate to vigorous (MVPA): > 3.0 METs). Bias (AG estimated - RC measured), mean absolute error (MAE), and their 95% confidence intervals were used to assess method accuracy and precision. RESULTS Across both methods and activity intensities, bias estimates ranged from -33.3 to 27.4 min for the period, and -4.1 to 3.9 min for the treadmill walk (Table 1). Mean absolute error was highest for LPA (33.3 and 21.1 min) and MVPA (27.4 and 12.0 min) during ADLs. Both algorithms overestimated time in MVPA and underestimated time in LPA during ADLs. CONCLUSION These bias and MAE estimates suggest that additional studies are needed to develop more accurate and precise wrist-based accelerometer algorithms focused on ADLs and free-living behaviors.