B0412
Title: Modeling accelerometer-based physical activity
Authors: Loki Natarajan - University of California San Diego (United States) [presenting]
Abstract: The use of wearable sensors to monitor physical activity is ubiquitous. These devices provide activity measurements at the minute-level (or even 30 Hz-level). Extracting useful information and modelling these densely sampled data can be challenging. We discuss functional data and other methods for analyzing accelerometer-derived physical activity measurements. We implement functional principal component analysis to study temporal patterns and derive the main modes of variation in accelerometer-based physical activity data and functional regression to estimate associations between functional physical activity profiles and health outcomes in a longitudinal setting. We discuss methodological issues pertinent to nested functional data (e.g., multiple visits or days of accelerometer wear within a participant). We illustrate methods via simulations and through application to a longitudinal study of physical activity and glucoregulatory markers.
