Background: Computer analysis of the sleep electroencephalogram (EEG) waveforms is widely employed, but there have been no systematic studies of its reliability.
Methods: The most commonly used computer methods are power spectral analysis with the fast-Fourier transform (FFT) and period amplitude analysis (PAA) with zero cross or zero first derivative half-wave measurement. We applied all three computer methods to the digitized EEG of 16 normal subjects who underwent 5 consecutive nights of baseline (placebo) recording. We evaluated the internight reliability of three non-rapid eye movement (NREM) frequency bands of special importance to sleep research: delta (0.3-3 Hz), sigma (12-15 Hz), and beta (15-23 Hz).
Results: Both FFT and the two methods of PAA gave excellent internight reliability for delta and sigma. Even a single night of recording correlated highly (r >.9) with the 5-night mean. Beta reliability was lower but still highly significant for both the PAA and the FFT measures.
Conclusions: Computer-analyzed sleep EEG data are highly reliable. Period amplitude methods demonstrate that wave incidence and period as well as amplitude are reliable, indicating that the reliability of composite measures (FFT power, PAA integrated amplitude) is not solely based on individual differences in EEG amplitude. The high internight stability of NREM delta indicates that it possesses traitlike characteristics and is relatively independent of day-to-day variations in state.