Universal approach to actigraphic sleep/wake scoring, verified against 5 classic algorithms on 3 datasets

Area of Science:

• Biomedical Engineering
• Sleep Science
• Signal Processing

Background:

• Actigraphy is a common, non-invasive tool for monitoring sleep-wake patterns using wrist-worn sensors.
• Traditional actigraphy analysis involves smoothing and thresholding activity counts with study-specific parameters.
• Existing methods require empirical tuning, limiting generalizability across different hardware and datasets.

Purpose of the Study:

• To introduce a universal low-pass filter for actigraphy data processing.
• To enhance the accuracy and consistency of sleep/wake detection in actigraphy.
• To compare the performance of the proposed filter against established actigraphy algorithms.

Main Methods:

• Development and application of a universal low-pass filter for actigraphy data.
• Validation using 1635 coregistrations of actigraphy and polysomnography (PSG) data from three diverse datasets.
• Cross-validation comparing the universal filter's performance with five classic sleep/wake detection algorithms (Cole-Kripke, Sazonov, Scripps, UCSD, Webster).

Main Results:

• The universal low-pass filter demonstrated significantly higher concordance with PSG for sleep/wake scoring compared to all classic algorithms.
• Filter parameter optimization across different data subsets converged to similar values, supporting the concept of a universal filter.
• The proposed method achieved superior performance on key metrics relevant to sleep analysis.

Conclusions:

• A universal low-pass filter offers a more accurate and broadly applicable method for analyzing actigraphy data.
• This approach reduces the need for study-specific parameter tuning, improving the reliability of sleep/wake detection.
• The universal filter represents a significant advancement in leveraging actigraphy for sleep research and clinical applications.