Circadian Rhythm Analysis Using Wearable Device Data: A Novel Penalized Machine Learning Approach

Abstract

AbstractBackgroundWearable devices have been widely used in clinical studies to study daily activity patterns, but the analysis remains the major obstacle for researchers.Study ObjectiveThis study proposed a novel method to characterize sleep-activity rhythms using actigraphy and further used it to describe early childhood daily rhythm formation and examine its association with physical development.MethodsWe developed a machine learning-based Penalized Multi-band Learning (PML) algorithm to sequentially infer dominant periodicities based on Fast Fourier Transform (FFT) and further characterize daily rhythms. We implemented and applied the algorithm to Actiwatch data collected from a 262 healthy infant cohort at 6-, 12-, 18-, and 24-month old, with 159, 101, 111, and 141 subjects participating at each time point respectively. Autocorrelation analysis and Fisher’s test for harmonic analysis with Bonferroni correction were applied to compare with PML. The association between activity rhythm features and early childhood motor development, assessed by Peabody Developmental Motor Scales-Second Edition (PDMS-2), was studied through linear regression.ResultsPML results showed that 1-day periodicity is most dominant at 6 and 12 months, whereas 1-day, 1/3-day, and 1/2-day periodicities are most dominant at 18 and 24 months. These periodicities are all significant in Fisher’s test, with 1/4-day periodicity also significant at 12 months. Autocorrelation effectively detected 1-day periodicity but not others. At 6 months, PDMS-2 is associated with assessment seasons. At 12 months, PDMS-2 is associated with seasons and FFT signals at 1/3-day periodicity (P<.001) and 1/2-day periodicity (P=.04). In particular, subcategories of stationary, locomotion, and gross motor are associated with FFT signals at 1/3-day periodicity (P<.001).ConclusionsThe proposed PML algorithm can effectively conduct circadian rhythm analysis using time-series wearable device data. Application of the method effectively characterized sleep-wake rhythm development and identified the association between daily rhythm formation and motor development during early childhood.