475 Automatic detection of self-similarity and prediction of CPAP failure

Abstract Introduction Sleep disordered breathing is a significant risk factor for cardiometabolic and neurodegenerative diseases. Tolerance efficacy of continuous positive airway pressure (CPAP), the primary form therapy sleep apnea, often poor. High loop gain (HLG) driving mechanism central apnea or periodic breathing. The current study aimed to develop computational approach detect HLG based on self-similarity in respiratory oscillations during solely using patterns, measured via Respiratory Inductance Plethysmography (RIP). To quantify potential utility developed similarity metric, presented algorithm was used predict acute CPAP failure. Methods We an detecting apneas as periods with reduced effort, manifested RIP signal low amplitude. Our calculates patterns between consecutive hypopnea. Working under assumption that high induces self-similar increases failure CPAP, full night similarity, computed diagnostic non-CPAP polysomnography (PSG), which we defined titration index (CAI)>10. Central labels are obtained both from manual scoring by technologists, automated this study. Massachusetts General Hospital (MGH) database used, including 2466 PSG pairs recordings. Results Diagnostic CAI technologist predicted AUC 0.82 ±0.03. Based automatically generated labels, combination resulted 0.85 ±0.02. A subanalysis performed population labeled CAI>5. Full 0.57 ±0.07 0.65 ±0.06 labels. Conclusion This showed can be derived way. proposed feature, surrogate estimate expressed easily accessible effort signals, regardless admixed obstructive features such flow-limitation, aid prediction success. Support (if any):