B1578
Title: Measurement and analysis of sedentary behavior derived from wearable sensors
Authors: Rong Zablocki - University of California San Diego (United States)
Loki Natarajan - University of California San Diego (United States) [presenting]
Abstract: Sedentary behaviour (SB) is a recognized risk factor for many chronic health conditions. Wearable accelerometers offer a unique opportunity to measure SB at fine (e.g. 1-minute) granularity. The ActiGraph and ActivPAL are two accelerometers widely used to measure SB. Actigraphs measure movement, while ActivPALs measure posture. Data streams are used from both devices, and functional principal components analysis (FPCA) is applied to explore the variation of subjects' movement while sitting. A multilevel (to account for days nested within participants) FPCA is implemented on 400 post-menopausal women. Using principal (PC) scores, individuals' SB patterns and impact on metabolic health are described. The analyses show that more than 90\% of the total variation is explained by two subject-level and six day-level PCs, dramatically reducing dimension from the original minute-level scale. The first subject-level PC captures overall movement during any sitting bout, whereas the second PC contrasts movement during short vs medium vs long bouts. The application of machine learning methods is also discussed for posture classification, obviating the need for two devices. It is shown how novel statistical and machine learning methods can be applied to elucidate patterns of SB and their impact on health.
