WACSAW: An adaptive, statistical method to classify movement into sleep and wakefulness states

Area of Science:

• Biomedical Engineering
• Sleep Science
• Data Science

Background:

• Wearable actimeters are crucial for studying sleep in natural settings.
• Existing algorithms struggle with accuracy, particularly during quiet wakefulness.
• Advancements in hardware enable higher-frequency data collection for sophisticated analysis.

Purpose of the Study:

• To develop and validate a novel statistical algorithm for improved sleep/wake classification using wearable actimetry data.
• To address limitations of current algorithms in accurately identifying behavioral states, especially during periods of low activity.

Main Methods:

• Developed the Wasserstein Algorithm for Classifying Sleep and Wakefulness (WACSAW), utilizing optimal transport techniques.
• WACSAW analyzes movement variability and clusters activity distributions using k-nearest neighbors.
• Segments are classified as sleep or wake based on proximity to an idealized sleep distribution.

Main Results:

• WACSAW achieved over 95% overall accuracy in classifying sleep and wake states, validated against participant logs.
• The algorithm demonstrated approximately 10% superior performance compared to a clinically validated actimetry system.
• The method provides an individually-tuned statistical approach to actimetry.

Conclusions:

• WACSAW offers a novel, accurate, and individually-tuned statistical method for sleep/wake classification from actimetry data.
• The algorithm improves upon existing systems, particularly in challenging conditions like quiet wakefulness.
• The methodology yields auxiliary information potentially relevant to sleep outcomes.