Four State Sleep Staging From a Multilayered Algorithm Using Electrocardiographic and Actigraphic Data.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Sleep Medicine

Background:

• Polysomnography is the gold standard for sleep studies but requires specialized equipment and technicians.
• Simpler methods for sleep evaluation are needed to expand the accessibility of sleep recordings.

Purpose of the Study:

• To develop and validate a scalable algorithm for sleep staging using engineered features from actigraphy, oxygenation, and electrocardiography sensors.
• To model polysomnography results with a simpler data acquisition approach.

Main Methods:

• Utilized a large, publicly available sleep dataset (1769 nights) from the Multi-Ethnic Study of Atherosclerosis.
• Engineered features from actigraphic, oxygenation, and electrocardiographic data.
• Applied a layered approach using recurrent neural networks and Extreme Gradient Boosting (XGBoost) for algorithm development and validation on a separate dataset (177 nights).

Main Results:

• The algorithm achieved an accuracy of 0.833 and a kappa of 0.736 in classifying four sleep states: wake, light sleep, deep sleep, and REM sleep.
• Heart rate variability was identified as the most significant feature for sleep staging.
• The developed algorithm outperformed previously described approaches in accuracy and kappa for sleep staging.

Conclusions:

• A multilayered algorithm shows significant potential for accurate sleep staging.
• The findings support the development of simple, wearable devices for effective sleep staging.
• The study provides accessible code for the developed sleep staging algorithm.