Separating Bedtime Rest from Activity Using Waist or Wrist-Worn Accelerometers in Youth

Recent interest in sedentary behavior and technological advances expanded use of watch-size accelerometers for continuous monitoring of physical activity (PA) over extended periods (e.g., 24 h/day for 1 week) in studies conducted in natural living environment. This approach necessitates the development of new methods separating bedtime rest and activity periods from the accelerometer recordings. The goal of this study was to develop a decision tree with acceptable accuracy for separating bedtime rest from activity in youth using accelerometer placed on waist or wrist. Minute-by-minute accelerometry data were collected from 81 youth (10–18 years old, 47 females) during a monitored 24-h stay in a whole-room indirect calorimeter equipped with a force platform covering the floor to detect movement. Receiver Operating Characteristic (ROC) curve analysis was used to determine the accelerometer cut points for rest and activity. To examine the classification differences, the accelerometer bedtime rest and activity classified by the algorithm in the development group (n = 41) were compared with actual bedtime rest and activity classification obtained from the room calorimeter-measured metabolic rate and movement data. The selected optimal bedtime rest cut points were 20 and 250 counts/min for the waist- and the wrist-worn accelerometer, respectively. The selected optimal activity cut points were 500 and 3,000 counts/min for waist and wrist-worn accelerometers, respectively. Bedtime rest and activity were correctly classified by the algorithm in the validation group (n = 40) by both waist- (sensitivity: 0.983, specificity: 0.946, area under ROC curve: 0. 872) and wrist-worn (0.999, 0.980 and 0.943) accelerometers. The decision tree classified bedtime rest correctly with higher accuracy than commonly used automated algorithm for both waist- and wrist-warn accelerometer (all p