CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0519
Title: Identifying circadian chronotypes using accelerometers Authors:  Julia Wrobel - Columbia University (United States) [presenting]
Jeff Goldsmith - Columbia University (United States)
Vadim Zipunnikov - Johns Hopkins Bloomberg School of Public Health (United States)
Abstract: Circadian rhythms are 24-hour biological processes that influence health on both the macroscopic and molecular level. Activity profiles produced by accelerometers can be used to understand circadian rhythm and detect chronotypes, which are subject-level differences in timing of circadian cycles. The focus is on understanding differences in the timing and intensity of activity, using a technique called registration. Registration aligns accelerometer curves by separating them into components of amplitude and phase variability. After alignment, the amplitudes show population level activity patterns that are consistent with well-documented diurnal patterns, and the phases contain information on subject-specific wake and sleep times. Previous work outlined a novel nonparametric method for registering exponential family functional data. We expand that method to understand circadian patterns in accelerometer curves, developing a 4-parameter approach that emphasizes interpretability of phase and amplitude components. After alignment the amplitudes can be described in terms of two parameters that identify the overall activity level, and whether each subject is likely to have a mid-day energy dip. The phases can be described in terms of parameters that identify the shift and duration of a subjects wake period. We validate our method using accelerometer data from the Baltimore Longitudinal Study of Aging. Code is publicly available as part of the registr package.