P093 An algorithm for detecting local sleep during wakefulness in OSA

Abstract Aim Local sleep during wakefulness is characterised by regional increased theta EEG activity and is associated with impaired cognitive performance in sleep deprived healthy subjects. Local sleep during wake may underlie neurobehavioral impairment in obstructive sleep apnea (OSA). To date, existing algorithms to detect local sleep have only been validated in healthy populations and not in people with OSA. To address this, we developed a local sleep detection algorithm using key theta wave characteristics: negative peak amplitude, cycle duration, and spectral power. Methods High-density EEG (256 channels) was recorded during resting wakefulness at six time points across 24 hours (Karolinska Drowsiness Test at baseline, three post simulated driving sessions and pre and post sleep). EEG signals were band-pass filtered in the theta range (4–8 Hz), zero crossings were detected, and theta waves were identified when half wave duration was 0.0625-0.125 sec. We then verified that the negative peaks of the identified theta waves co-occurred with the maximal spectral theta power. A within-subject histogram of all verified negative peak amplitudes across recordings, for each EEG channel, was fitted with a Gaussian distribution to determine a statistical threshold (p=.01; z ≈ -2.326). Local sleep events were identified when theta wave amplitudes exceeded this threshold. Results Comparison between automated and manually identified theta waves in 12 recordings showed good algorithm accuracy (F1 ≈ 0.81, recall = 0.89, precision = 0.74). Conclusion These results support the algorithm’s utility to detect local sleep during wake in OSA and offers an approach to explain inter-individual differences in performance.