Abstract
Laboratory studies show that performing ~2 min of activity every 20–30 min, known as activity breaks, is associated with improved cardiometabolic health outcomes. However, objectively identifying these activity breaks in a free-living setting remains challenging. This study aimed to develop and validate an algorithm to detect activity breaks using accelerometer data. Thirty-one healthy adults (mean (SD) age 33 (12) y, 71% female) wore three accelerometers (ActiGraphs on the wrist and hip; ActivPAL on the thigh) while completing eight activity breaks, approximately 30 min apart. Participants self-recorded activity break start and stop times. Lasso regression with an ‘activity break’ definition was used to develop an algorithm that used all three accelerometers and for each accelerometer individually, using two-thirds of participant data for development and the remaining third for testing. External validation was undertaken using an independent dataset. For the three-device algorithm, the mean difference between predicted and reported activity breaks was 0.0 (95% CI: −1.7, 1.7), while predicted break duration was, on average, 0.9 min longer than the reported duration (95% CI: 0.2, 1.6). These findings suggest the algorithm accurately estimates the frequency and duration of activity breaks and may be useful for quantifying this behaviour in free-living studies.