A taxonomic analysis of sleep stages

To study the structure of human sleep at the level of sleep stages. We applied taxonomic statistics to detect significant configurations (types) of different physiologic variables and their relationship to sleep stages. DESIGN AND STATISTICS: Polygraphic sleep recordings from 32 subjects (normal sleepers as well as patients with insomnia, sleep apnea, or narcolepsy; n = 8 per group) were visually scored and submitted to a configural frequency analysis. The configural frequency analysis was computed with 3 continuous input variables: an electroencephalogram parameter, which represents the point of gravity of the EEG frequency distribution; the alpha slow-wave index, and the Rest Index, based on the presence or absence of phasic electromyographic activity. These variables were dichotomized for further analysis. The combination of 2 levels ( + or -) and 3 variables resulted in 2(3) patterns ( +++ to - - - ). The configural frequency analysis is a nonparametric X2-type multivariate statistic that identifies significant patterns or types.

Each sleep stage contained 3 or 4 types. For non-rapid eye movement sleep stages 2, 3, and 4, types overlapped, whereas there was no overlap of types between stages 1 and 2. Types of rapid eye movement sleep did not overlap with those from stages 2, 3, and 4 but did overlap with wake and stage 1 types. The majority of observed types were significant in all 4 groups of subjects.

Sleep stages appear to be less homogenous than rule-based sleep scoring would suggest. Types were either restricted to one stage or overlapped with neighboring stages.