A multi-sensor machine learning approach to precision sleep tracking for nightshift workers - Project Summary/Abstract All essential 24-hr operations (e.g., first responders, hospital services) rely on nightshift workers who forgo nocturnal sleep for work. Shift workers comprise 20% of the workforce, with 15 million regularly working nights. Most do not adapt to the inverted sleep-wake schedule, especially as they are provided little to no support in managing this challenge. Because nightshifts are critical to safety-sensitive operations, impaired sleep health and performance increases risks for catastrophic accidents, making the well-being of nightshift workers an issue of public health and safety. Given this landscape, there is an urgent need for precision sleep medicine solutions with algorithms designed specifically for nightshift workers to account for their unique challenges. Fortunately, the growth of digital health technology has finally reached a point where precision sleep medicine for nightshift workers is now within reach. A critical obstacle to providing tailored sleep interventions for nightshift workers has been the reliance on actigraphy methods that were developed in the early 1990s. These legacy methods utilize algorithms trained on nighttime sleep, and do not perform adequately in detecting daytime sleep in nightshift workers. As such, these legacy algorithms have reduced validity for nightshift workers whose sleep is displaced to the daytime due to the nightshift schedule. To address these challenges, this proposal seeks to validate a novel sleep tracking approach that leverages machine learning (ML) on data from a multi-sensor system. This system will comprise more powerful accelerometers in a consumer-based wearable device (Apple Watch) within an ecosystem of other sensors that provide contextual information about the individuals ambient environment and behaviors. We proposed that combined with machine learning, a multi-sensor approach will be more powerful than legacy actigraphy algorithms because it can distinguish between three key psychophysiological states: 1) awake and alert, 2) awake but attempting sleep, and 3) asleep. The objective of this proposal is to validate a multi-sensor ML approach to track sleep in nightshift workers, and to identify facilitators and barriers for real-world implementation. We propose a type-I hybrid effectiveness-implementation trial. We will compare a multi-sensor ML approach to legacy actigraphy algorithms validated against gold-standard in-lab polysomnography. Our central hypothesis is that a multi- sensor ML approach will outperform legacy algorithms. This will be followed by at-home implementation of the multi-sensor tracking for four weeks, after which we will evaluate user experience and identify facilitators and barriers of use. Upon completion of the study, the researchers anticipate having established a validated, feasible, and open-source multi-sensor ML system for tracking sleep in nightshift workers. This innovative approach holds the potential to significantly advance precision sleep medicine for this population, ultimately impacting outcomes related to occupational health, performance, safety, and public health.
