Abstract: Alterations of sleep duration and architecture have been associated with increased morbidity and mortality, and specifically linked to chronic cardiovascular disease and psychiatric disorders, such as type 2 diabetes or depression. Measurement of sleep quality to assist in the diagnosis or treatment of these diseases is not routinely performed due to the complexity and cost of conventional methods. The objective of this study is to cross-validate the accuracy of an automated algorithm that stages sleep from the EEG signal acquired with sensors that can be self-applied by patients.
Methods: This retrospective study design included polymsomnographic records from 19 presumably healthy individuals and 68 patients suspected of having sleep disordered breathing (SDB). Epoch-by-epoch comparisons were made between manual vs. automated sleeps staging (from the left and right electrooculogram) with the impact of SDB severity considered.
Results: Both scoring methods reported decreased Stage N3 and REM and increased wake and N1 as SDB severity increased. Inter-class correlations and Kappa coefficients were strong across all stages except N1. Agreements across all epochs for subjects with normal and patients with mild SDB were: wake = 80%, N1 = 25%, N2 = 78%, N3 = 84% and REM = 75%. Agreement decreased in patients with moderate and severe SDB amounting to: wake = 71%, N1 = 30%, N2 = 71%, N3 = 65%, and REM = 67%. Differences in detection of sleep onset were within three-minutes in 48 % of the subjects and 10-min in 73 % of the cases and were not impacted by SDB severity. Automated staging slightly underestimated total sleep time but this difference had a limited impact on the respiratory disturbance indexes.
Conclusions: This cross-validation study demonstrated that measurement of sleep architecture obtained from a single-channel of forehead EEG can be equivalent to between-rater agreement using conventional manual scoring. The accuracies obtained with automated sleep staging were inversely proportional to SDB severity at a rate similar to manual scorers. These results suggest that the automated sleep staging used in this study may prove useful in evaluating sleep quality in patients with chronic diseases.