Curve Registration to Identify Circadian Rhythm Chronotypes in Accelerometer Data

• Activity Number: 181 - SPEED: Statistical Learning and Data Science Speed Session 1, Part 2
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 11:15 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Erin McDonnell* and Julia Wrobel and Jeff Goldsmith and Vadim Zipunnikov
• Companies: Columbia University and Columbia University and Columbia University and Johns Hopkins University
• Keywords: Functional data analysis; Registration; Parametric; Circadian rhythm

Abstract

Accelerometers give rich insights into activity levels over time. Collected nearly continuously over 24 hours, their data can provide novel insights into circadian rhythms. Prior work using such data has revealed population-level activity patterns, but considerable subject-level variability exists in features such as wake/sleep times and activity intensity. These variations yield an opportunity to refine our understanding of chronotypes, or behavioral manifestations behind one’s natural 24-hour rhythm. Registration, or alignment, is a technique in functional data analysis that separates variability in activity intensity from variability in time-dependent markers like wake and sleep times; this is well-suited to issues in data-driven approaches to studying chronotypes using accelerometer data. We develop a parametric registration framework that emphasizes interpretability: specifically, we generate subject-specific piecewise linear warping functions. From resulting parameters, we can identify chronotypes (e.g. “morning larks” and “night owls”). We apply this method to data from the Baltimore Longitudinal Study of Aging and associate chronotypes with demographics and health outcomes.