Distinguishing Patients with Central from Obstructive Sleep Apnea Using Overnight Breath Sound Recordings

Abstract

Introduction: Despite the increasing need for portable home sleep apnea (SA) monitoring, the limited number of channels renders most monitors incapable of distinguishing central from obstructive SA (CSA and OSA respectively). BresoDx® is a device that detects apneas and hypopneas via breath sound recordings during sleep. Tha apnea-hyponea index determined by BresoDx has 95% agreement with polysomnography (PSG). The aim of this study was to determine the accuracy of BresoDx in distinguishing patients with predominantly central sleep apnea (CSA) from those with predominantly obstructive sleep apnea (OSA). Methodology: Patients undergoing overnight PSG had a simultaneous BresoDx recording. Acoustic features of the BresoDx recording for respiratory events were extracted including turbulence, frequency content, and snoring distribution. Acoustic features and demographics were used to train an artificial intelligence algorithm using PSG scoring as the reference. Data from all subjects except one were used for training the algorithm and the left-out subject used for validation. This process was repeated iteratively so that each subject was used for validation once. A subject was classified as having CSA if > 50% of events were central and OSA otherwise. Results: Comparison of BresoDx with PSG in 52 subjects is shown in Table 1. BresoDx had 85.7% sensitivity, 88.9% specificity, and 88.5% accuracy in detecting CSA vs PSG. Discussion and Conclusion Analysis of BresoDx recordings and demographic data was very good at distinguishing those with mainly CSA from those with mainly OSA.