Minimizing Interrater Variability in Staging Sleep by Use of Computer-Derived Features
Magdy Younes1,2,3, Patrick J Hanly2
1YRT Ltd, Winnipeg, MB, Canada.
Summary
Digitally derived data significantly improved polysomnogram (PSG) sleep staging accuracy by reducing inter-scorer variability. This tool helps scorers by providing objective information on sleep depth and specific EEG patterns, minimizing guesswork in sleep analysis.
Area of Science:
- Sleep Medicine
- Neurophysiology
- Biomedical Engineering
Background:
- Inter-scorer variability in sleep staging of polysomnograms (PSGs) is a significant challenge.
- This variability often stems from difficulties in interpreting specific electroencephalogram (EEG) patterns within epochs, such as wakefulness, spindles, K complexes, and delta wave duration.
Purpose of the Study:
- To investigate whether providing digitally derived information to PSG scorers can reduce inter-scorer variability.
- To assess the impact of objective digital data on the consistency of sleep staging.
Main Methods:
- Fifty-six PSGs were manually scored by two technologists (M1, M2).
- An automatic system was used, with scorers editing the output (Edited-Auto).
- Manual scores were modified using digitally obtained data on sleep depth, delta duration, spindles, and K complexes.
Main Results:
- Percent agreement between scorers increased from 78.9% to 96.5% after modification.
- Errors in scoring averaged 7.1% and 6.9% for the two scorers.
- Excellent agreement was observed between modified scores and initial manual scores.
Conclusions:
- Digitally obtained information on sleep depth, delta duration, spindles, and K complexes can substantially reduce interrater variability in sleep staging.
- This approach minimizes subjective interpretation and guesswork in scoring ambiguous epochs.
