Causal Analysis of Physiological Sleep Data Using Granger Causality and Score-Based Structure Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Sleep studies generate complex time-series sensor data crucial for medical understanding.
• Causal inference from polysomnographic data is vital for personalized medicine.

Purpose of the Study:

• To learn the causal structure from polysomnographic data using Granger causality and DYNOTEARS.
• To compare the effectiveness and results of these two causal discovery methods.
• To explore the relationship between causal structure and participant characteristics.

Main Methods:

• Applied Granger causality and DYNOTEARS to polysomnographic data from 600 adult volunteers.
• Learned dynamic Bayesian networks (DBNs) to represent causal relationships.
• Compared graph structures and analyzed associations with participant anthropometrics.

Main Results:

• Both methods identified similarities, including mutual causation between electrooculogram (EOG) and electroencephalogram (EEG) signals, and between sleeping position and blood oxygen saturation (SpO2).
• DYNOTEARS revealed more causal links to sleeping position than Granger causality.
• Participant waist size was associated with causal links between electrocardiogram (ECG) and EOG/EEG signals.

Conclusions:

• Body shape may influence the heart-brain relationship during sleep.
• Granger causality and DYNOTEARS can yield different causal structures from real-world sleep data.
• Causal discovery methods offer insights into physiological interactions during sleep.