0187 Hypnograms for 365 Nights Predicts Subjective Sleep Quality in Healthy Adults: Results from the Ultra Long-Term Sleep (ULTS) Study

Abstract

Abstract Introduction Sleep quality is fundamental to our somatic and mental health. However, the relationship between subjective sleep quality and sleep architecture remains poorly understood. New wearable or minimally invasive technologies facilitate the recording of electroencephalography (EEG) with lower spatial resolution than standard EEG but much greater longitudinal dispersion. This enables investigation of day-to-day variation in sleep measured directly with EEG. This study will compare EEG-derived sleep parameters with covariates such as sustained attention and subjective sleep quality. Methods Twenty-five healthy adults were implanted with a two-channel subcutaneous EEG (sqEEG) lead. Twenty subjects completed the 1-year protocol (average 32±13 years of age). Their sqEEG signals were recorded each night for 1 year alongside a morning 3-minute Psychomotor Vigilance Task (PVT) and self-reported sleep quality, which included Karolinska Sleepiness Scale (KSS). A deep learning model, U-Sleep, was fine-tuned on sqEEG with synchronized gold standard polysomnography used as ground truth. Hypnograms and sleep parameters were thus automatically calculated. Results Subjective sleep quality measured by KSS revealed a moderate negative correlation with rapid-eye-movement (REM) duration (r=-0.31, 95% CI=(-0.31, -0.31)), and total sleep time (TST) (r=-0.31, 95% CI=(-0.31, -0.31)). There was a moderate correlation between KSS and mean PVT reaction time (r=0.21, 95% CI=(0.21, 0.22)). There was a low negative correlation between PVT and TST (r=-0.1). Preliminary results indicate a moderate correlation between sleep parameters and subjective sleep quality. The correlations with PVT were lower, which suggests that 3-minute PVT is not sensitive to TST in normal sleep. However, the correlation between PVT and KSS suggests that PVT does predict subjective sleep quality, but to a smaller degree than standard sleep parameters. Conclusion Measuring day-to-day variation in high-quality EEG-based sleep recordings has the potential of creating a new branch in sleep medicine. Patients can be evaluated not only by findings in a single recording but the stability and variation of all findings can be analyzed. Preliminary results suggest that subjective sleep quality can be predicted directly from sqEEG and potentially be explained by behavioral factors in a subsequent cause-effect analysis. Support (if any) The project is supported by Innovation Fund Denmark, UNEEG medical, and T&W Engineering.