Validation of an EEG-derived spectral frequency index (SFx) for continuous monitoring of sleep depth in humans

Sleep in humans is classically assessed by recording a multichannel electroencephalogram (EEG) in connection with an electrooculogram (EOG) and an electromyogram (EMG). In general, human sleep is manually staged into 6 categories (from awake through REM sleep to stage 4 reflecting deep sleep) on the basis of a visual inspection of EEG periods (epochs) of 20 - 30 s duration. This cumbersome methodology is still used in practice and for reference purposes. - The conversion of EEG-signals by means of Fast Fourier Transformation provides objective and reproducible information reflecting specific communicative features of the central nervous system. A special part of this information based on a specific algorithm is defined by the so-called spectral frequency index (SFx). This SFx-algorithm contains relationships among some particular EEG frequencies and provides objective percentage values about the state of consciousness of a person. In order to validate this new SFx-method, sleep as a physiological state of continuous alterations of consciousness and vigilance was chosen. A total of 36 nights of sleep from 18 healthy volunteers were staged manually by a scientist unaware of the protocol. The volunteers received either placebo or lormetazepam prior to commencement of the nocturnal recordings. The manually staged data were compared with the data obtained by the SFx-analysis. Both data sets SFx values and manually staged data were made comparable by averaging their values to a basic period length of 2 min duration giving 7960 pairs of data. The SFx data for sleep were found within a range from 35% to 100%. The SFx-medians of the manually staged data from "awake" to stage 4 were found in a decending order ("awake": 83% (lower and upper quartile 78% / 87%);"REM": 68% (63%/74%),"stage 1" :63% (57%/70%),"stage 2" :51% (47%/57%), "stage 3" :44% (42%/46%) and "stage 4" :42% (40%/44%). The rank correlation coefficient between the data pairs was calculated to be 0.79 indicating a substantial matching between the manually staged score and the SFx. We therefore conclude that the SFx is a suitable and objective indicator of sleep depth in humans.