A novel method to increase specificity of sleep-wake classifiers based on wrist-worn actigraphy

Area of Science:

• Sleep science
• Chronobiology
• Biomedical engineering

Background:

• Accurate assessment of sleep-wake patterns is vital for understanding health.
• Current wrist-worn actigraphy algorithms often overestimate sleep due to low specificity.

Purpose of the Study:

• To develop a novel sleep-wake classifier with enhanced specificity.
• To improve the accuracy of actigraphy in distinguishing between sleep and wake states.

Main Methods:

• Artificially balanced a training dataset with equal sleep and wake epochs.
• Optimized classification parameters for sensitivity and specificity using actimeter data from 12 subjects.
• Validated the classifier on night sleep data from 19 healthy subjects.

Main Results:

• Achieved 80.4% specificity and 88.6% sensitivity on a balanced dataset (3079.9 h).
• Validated classifier showed 89.4% sensitivity and 64.6% specificity for night sleep.
• Accurately estimated total sleep time (12.16 min difference) and sleep efficiency (2.83% difference).

Conclusions:

• Developed a device-independent sleep-wake classifier reducing bias towards sleep detection.
• Provides a foundation for more accurate sleep-wake assessments in real-world settings.
• Potential application in monitoring patients with disrupted sleep-wake patterns.