From Pulse to Phenotype: Sleep Apnea Endotyping for Polygraphy via Oximeter-Derived Autonomic Arousal

Understanding the underlying cause of OSA in the individual patient, referred to as pathophysiologic endotyping, is essential for personalized care. The current classification of these traits from routine sleep recordings relies on manually scored arousals from sleep. Automating this process could widen the applicability of endotyping.

Can analyzing autonomic variability, derived from finger oximeter photoplethysmography (PPG), accurately classify sleep apnea pathophysiologic endotypes with results comparable with the established EEG-based method?

Eighty-seven patients referred for suspected OSA underwent ambulatory polysomnography. Pulse wave amplitude, pulse rate, and pulse propagation time were derived from the PPG pulse waveform. A logistic mixed-effect model was developed to predict the presence of EEG-based arousals after respiratory events using these PPG-derived parameters, together with the associated event type and blood oxygen saturation. The automatically predicted PPG-based arousals then were incorporated into an established model (Phenotyping Using Polysomnography) to determine OSA endotypic traits from the airflow and PPG-based signals. The agreement between endotypes with PPG-based and EEG-based arousals was assessed using intraclass correlation coefficient (ICC).

PPG-derive pulse wave responses to respiratory events were more pronounced in the presence of arousal (P < .001 for all). The model using PPG metrics to predict cortical arousal demonstrated moderate performance (sensitivity, 0.71; specificity, 0.59). Endotypic traits derived by PPG-derived arousals showed strong agreement compared with the EEG-derived reference (ICC for loop gain at 1 cycle/min, 0.95 [95% CI, [0.88-0.98]; ICC for ventilation at active muscle activity, 0.96 [95% CI, 0.81-0.99]; ICC for ventilation at passive muscle activity, 0.99 [95% CI, 0.99-0.99]; ICC for ventilation at minimum muscle activity, 0.99 [0.99-0.99]; ICC for arousal threshold, 0.85 [95% CI, 0.48-0.96]; ICC for muscle compensation, 0.80 [95% CI, 0.60-0.91]).

Our results show that using pulse wave features instead of manually scored EEG-based arousals in respiratory modelling allows for accurately determining OSA endotypes. This approach might enable physiologic endotyping in non-EEG-based sleep studies, expanding the accessibility of personalized OSA management.