Hypnodensity sleep metrics from fingertip photoplethysmography are better associated with daytime sleepiness and fatigue than traditional metrics from polysomnography.

BACKGROUND: Traditional sleep metrics from polysomnography are often limited by the hypnogram, in which each 30-s epoch is classified into one of the five sleep stages. This oversimplifies sleep stage continuity and fragmented sleep, typical in sleep-disordered breathing (SDB). Deep learning offers more dynamic sleep modelling with a higher temporal resolution by providing probabilities for each sleep stage, known as hypnodensities. RESEARCH QUESTION AND METHODS: We investigated whether high-frequency (1/s) hypnodensities captured from nocturnal fingertip photoplethysmography (PPG) recordings better correlate with patient-reported outcomes in SDB than traditional metrics. RESULTS: Among 2,280 patients (65% male, mean age 62 years, mean apnoea-hypopnoea index (AHI) 23.5/h), we found that several PPG-derived hypnodensity metrics correlate with self-reported daytime fatigue and sleepiness. The level of correlation was better or similar compared to traditional hypnogram-based sleep metrics derived from manually scored polysomnography. Specifically, PPG-based hypnodensities for N1 and N3 sleep showed a stronger association (p < 0.005) with fatigue and sleepiness than manually scored N1 and N3 sleep. Multivariable logistic regression (adjusted for age, sex, body mass index, AHI, and total sleep time below 90% blood oxygen saturation) revealed significant associations (p < 0.01) between multiple PPG-based hypnodensity metrics and daytime sleepiness. CONCLUSION: These findings suggest that PPG-based hypnodensities improve the characterisation of sleep quality and better predict patient-reported outcomes in SDB patients. Future uses may include tracking sleep quality with fingertip pulse oximeters outside of sleep centres, potentially over multiple nights due to their simplicity.