Variability and predictability in human sleep

Area of Science:

• Neuroscience
• Sleep Science
• Computational Biology

Background:

• Sleep quality and cognitive function depend on the cyclical alternation of REM and NREM sleep stages.
• Understanding individual sleep architecture variability and predictability is crucial for sleep medicine interventions.
• Previous research lacked longitudinal data on sleep architecture robustness over extended periods.

Purpose of the Study:

• To characterize individual variability in sleep architecture over weeks using ultra-long-term recordings.
• To test the predictability of sleep stage transitions in healthy humans.
• To investigate the robustness of NREM-REM sleep patterns across successive nights.

Main Methods:

• Utilized minimally invasive, sub-scalp electroencephalography (EEG) for longitudinal sleep recording over a median of 30 days in eight subjects.
• Decomposed EEG signals into five frequency bands (δ, θ, α, σ, β) using multi-taper time-frequency analysis.
• Employed unsupervised and supervised learning for quantifying spectral variability and predicting sleep stage transitions, respectively.

Main Results:

• Sleep architecture dissimilarity was significantly lower within individuals (D = 2.5 ± 0.7) compared to across individuals (D = 4.1 ± 0.5).
• Identified archetypal sleep patterns unique to each individual's NREM-REM spectral architecture.
• Successfully predicted NREM to REM sleep transitions 2-4 minutes in advance with high accuracy (AUC = 0.88) based on δ and σ band interplay.

Conclusions:

• Healthy human sleep exhibits variability across consecutive nights, yet core sleep oscillation dynamics are consistent within individuals.
• The predictability of sleep stage transitions offers potential for anticipating sleep-stage-specific pathological symptoms.
• This study establishes a foundation for personalized sleep medicine through long-term sleep monitoring and predictive modeling.