Performance of Somno-Art Software compared to polysomnography interscorer variability: A multi-center study.

Area of Science:

• Sleep Medicine
• Computational Neuroscience
• Biomedical Engineering

Background:

• Visual scoring of polysomnography (PSG) is the gold standard for sleep analysis but suffers from inter- and intra-scorer variability.
• Previous validation of Somno-Art Software showed agreement with expert scorers.
• Evaluating algorithm performance against a broader pool of scorers is crucial for clinical adoption.

Purpose of the Study:

• To evaluate the performance of the Somno-Art Software sleep scoring algorithm against multiple visual scoring centers.
• To assess the algorithm's reliability and agreement with human scorers across diverse patient populations.

Main Methods:

• Sixty PSG and actimetry recordings from healthy subjects and patients with OSA, insomnia, or MDD were scored by 5 centers and Somno-Art Software.
• Intra-class correlation coefficients (ICC) and Wilcoxon Signed-Rank Tests were used for statistical analysis.
• Epoch-by-epoch agreement was analyzed between scorers and the algorithm.

Main Results:

• Somno-Art Software's estimations for sleep efficiency, wake, N1+N2, and REM sleep fell within the inter-scorer range.
• The algorithm underestimated N3 sleep in OSA patients and overestimated sleep latency in insomniacs.
• Good to excellent ICC scores were observed for most parameters, with moderate scores for N3 sleep.

Conclusions:

• Somno-Art Software exhibits good inter-scorer reliability, performing within the range of visual PSG scorers.
• The algorithm shows potential for clinical use, though specific adjustments may be needed for certain parameters or patient groups.
• Further validation across larger and more diverse datasets is warranted.