A hybrid approach for continuous detection of sleep-wakefulness fluctuations: validation in patients with...

Fluctuations in sleep/wake state have been proposed as an important mechanism contributing to the development of oscillatory breathing patterns, including Cheyne-Stokes respiration in patients with heart failure. In order to properly assess the interactions between changes in state and changes in ventilatory parameters, a methodology capable of continuously and reliably detecting state transitions is needed. Traditional fixed-epoch analysis of polysomnographic recordings is not suitable for this purpose. Moreover, visual identification of changes in the dominant electroencephalogram activity at the transition from wakefulness to sleep and vice versa is often very subjective. We have therefore developed a hybrid approach--including both visual scoring and computer-based procedures--for continuous analysis of state transitions from polysomnographic recordings, specifically tailored for fluctuations between wakefulness and non-rapid eye movement-1 and -2 sleep. The overall analysis process comprises three major phases: (1) manual identification of relevant electroencephalogram/electrooculogram features and events, including a sample of unequivocal alpha and theta-delta activity; (2) automatic statistical discrimination of dominant electroencephalogram activity; and (3) state classification (wakefulness, non-rapid eye movement-1 and -2). The latter is carried out by merging information from visual scoring with the output of the discriminator. Validation has been carried out in 16 patients with heart failure during daytime Cheyne-Stokes respiration, using a training and testing set of electroencephalogram polysomnograms. The statistical discriminator correctly classified 99.1 ± 1.4% and 99.2 ± 1.1% of unequivocal alpha and theta-delta activity. This approach has therefore the potential to be used to reliably measure the incidence and location of sleep-wake transitions during abnormal breathing patterns, as well as their temporal relationship with major ventilatory events.