Title

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

Document Type

Article

Publication Date

8-1-2022

Publication Title

Sleep medicine

Abstract

The visual scoring of gold standard polysomnography (PSG) is known to present inter- and intra-scorer variability. Previously, Somno-Art Software, a cardiac based sleep scoring algorithm, has been validated in comparison to 2 expert visual PSG scorers. The goal of this research is to evaluate the performances of the algorithm against a pool of scorers. Sixty PSG and actimetry recording nights, representative of clinical practice (healthy subjects and patients suffering from obstructive sleep apnea [OSA], insomnia or major depressive disorder), were scored by 5 different sleep scoring centers and by the Somno-Art Software. Intra-class correlation coefficient (ICC) and Wilcoxon Signed-Rank Test were calculated between each scorer and the average value of the 6 scorers, including Somno-Art Software. In addition, epoch-by-epoch agreement between scorers were analyzed. Somno-Art Software estimation of sleep efficiency, wake, N1+N2, N3 and REM sleep fit within the interscorer range for the full dataset and the subgroups, except for underestimating N3 sleep in OSA patients. Additionally, Somno-Art Software overestimated sleep latency compared to the average scoring for insomniacs (+4.7 ± 1.6min). On the full dataset, Somno-Art Software had good (0.75 < ICC<0.90) or excellent (ICC>0.90) ICC scores for all sleep parameters except N3 sleep (moderate score, 0.50 < ICC<0.75). For the 4-stages epoch-by-epoch agreement, Somno-Art Software was slightly below that of the visual scorers except for the healthy sub-group where an overlap was demonstrated. Somno-Art Software sleep scoring shows a good interscorer reliability in the range of the 5 visual polysomnography scorers.

Medical Subject Headings

Depressive Disorder, Major; Humans; Polysomnography; Reproducibility of Results; Sleep Apnea, Obstructive; Sleep Stages; Software

PubMed ID

35576829

Volume

96

First Page

14

Last Page

19

Find It @ Sladen

Share

COinS