Model-Based Human Circadian Phase Estimation Using a Particle Filter

We present a method for tracking an individual's circadian phase that integrates dynamic models of circadian physiology with physiological measurements in a Bayesian statistical framework. A model of the circadian pacemaker's response to light exposure is transformed into a nonlinear state-space model with a circadian phase state. The probability distribution of the circadian phase is estimated by a particle filter that predicts changes over time based on the model, and performs updates with information gained from physiological measurements. Simulations demonstrate how probability distributions allow flexible initialization of model states and enable statistical quantification of entrainment and divergence properties of the circadian pacemaker. The combined use of sleep-wake scheduling data and physiological measurements is demonstrated in a case study highlighting advantages for addressing the challenge of noninvasive ambulatory monitoring of circadian physiology.