PhD Dissertation Defense: Lara Weed

Title: Operationalizing Human Circadian and Sleep Regulation for Wearable Sensing and Real-World Applications

Abstract: Human circadian and sleep regulation influence physiology and behavior across daily life, yet translating decades of knowledge developed under controlled laboratory conditions into real-world measurement and application remains challenging. Outside the laboratory, behavior and environmental conditions can vary over time and mask the observable expression of circadian timing, while wearable measurements provide only indirect or incomplete representations of the underlying physiology. These challenges are compounded by substantial variation across individuals in physiology, geographic location, environmental exposure, and behavioral constraints, all of which can influence circadian timing and how it is observed. This dissertation develops engineering methods that make these sources of variability and uncertainty explicit and applies them to circadian questions spanning wearable sensing, travel, time policy, population health, and human performance.

First, I advance the computational infrastructure for circadian and sleep regulation modeling. CircadianPOMDP reformulates mechanistic models within a probabilistic framework that supports population-scale simulation of uncertain physiological states and heterogeneous model parameters. Across 1,000 virtual participants, simulations show that the expression of physiological variability depends on environmental context and that targeted light exposure can substantially accelerate predicted adaptation following travel, with personalization providing the greatest benefit when adaptation is otherwise slow or variable. I further develop a time system and light exposure framework that explicitly separates physical time, civil time, solar conditions, behavioral schedules, light exposure, and biological state, enabling consistent simulation across changing locations, seasons, time policies, and travel.

Second, I establish practical requirements for applying circadian models and activity pattern metrics to ambulatory wearable data. Probabilistic circadian phase estimation from wrist-recorded light shows that longer recordings reduce sensitivity to uncertain initialization under regular schedules, while persistent errors remain in shift-working populations despite additional observations. Analyses of wrist actigraphy further demonstrate that activity pattern metrics can be systematically distorted by the timing, duration, and distribution of missing observations, with time-of-day imputation improving robustness under defined missingness conditions.

Third, I apply physiologically grounded circadian science approaches to questions of health and performance. For public health, I developed simulations of United States time policy showing that biannual clock changes burden the circadian system more than either permanent policy, while permanent Standard Time produces the lowest burden overall. Under idealized exposure assumptions, these differences correspond to estimated reductions in stroke and obesity prevalence under permanent Standard Time. For human performance, I show that in young, healthy, naturally cycling women, nighttime neuromuscular performance differences are selective rather than generalized. High-velocity hamstring torque and double-leg drop jump height are reduced at night, while quadriceps peak torque, dynamic balance, and most landing mechanics show no detectable overall day–night differences.

Across these studies, the expression of physiological variability depends on environmental and behavioral context, the temporal reference used for observation influences the biological structure that can be detected, and the consequences of biological timing are selective across populations and outcomes. By making latent state, environmental input, temporal representation, and measurement uncertainty explicit, this work expands the range of circadian questions that can be investigated outside the laboratory and provides a foundation for future systems that estimate biological state from daily-life measurements, predict responses to changing conditions, and guide personalized interventions.