Ordinal Sleep Depth: A Data-Driven Continuous Measurement of Sleep Depth

Area of Science:

• Neuroscience
• Sleep Medicine
• Artificial Intelligence in Healthcare

Background:

• Conventional sleep staging uses discrete stages, potentially missing the continuous nature of sleep depth.
• A need exists for a more nuanced, data-driven measure of sleep depth.

Purpose of the Study:

• To develop and validate a continuous measure of sleep depth, ordinal sleep depth (OSD), using deep learning.
• To assess OSD's correlation with arousal probability and its association with age, sex, sleep-disordered breathing (SDB), and cognitive impairment.

Main Methods:

• A convolutional neural network was trained on 21,787 polysomnography recordings to estimate continuous sleep depth.
• Ordinal regression was incorporated to account for the ordered nature of non-REM sleep stages.
• OSD was compared against the odds ratio product (ORP) and correlated with clinical variables.

Main Results:

• OSD demonstrated a strong linear correlation with arousal probability (r=0.994), outperforming ORP (r=0.923).
• OSD reflected age-related decreases in sleep depth and showed females have deeper sleep than males.
• OSD more accurately captured sleep depth reductions in patients with SDB and cognitive impairment.

Conclusions:

• Ordinal sleep depth (OSD) is a validated, data-driven measure of sleep depth with strong correlation to arousal probability.
• OSD effectively captures variations in sleep depth related to age, sex, SDB, and cognitive impairment.
• OSD offers a nuanced understanding of sleep architecture, highlighting sleep depth as a critical dimension.