Determine the Occurrence of Apneas and Hypopneas Using Drop in Oxygen Saturation and Breathing Sounds

Saha, Shumit 1, 2 ; Kabir, Muammar 2 ; Gavrilovic, Bojan 2 ; Zhu, Kaiyin 2 ; Yadollahi, Azadeh 1, 2

1. Institute of Biomaterials and Biomedical Engineering, University of Toronto; 2. Toronto Rehabilitation Institute-University Health Network

Background: Obstructive sleep apnea (OSA) is a chronic respiratory disorder characterized by the repetitive collapses of the pharyngeal airway during sleep. OSA severity is determined by the apnea-hypopnea index (AHI) which represents the average number of apneas (complete collapse) and hypopneas (partial collapse) per hour of sleep. The gold standard assessment for OSA requires participants to spend the night in a sleep laboratory and undergo polysomnography (PSG), which requires the attachment of up to 20 sensors to the body. Due to the complex nature of PSG, 84% of Canadians who are at risk of OSA are not diagnosed. To overcome this problem, portable sleep monitoring systems have been introduced. Current portable monitoring devices of OSA report only the overall AHI rather than each occurrence of apneas or hypopneas. However, estimating a wrong event can be nullified by not detecting a true event and have no effect on overall AHI estimation. Therefore, this limits the determination of clinical diagnosis such as postural or stage related AHI and related therapeutic outcomes. Thus, we aim to develop an algorithm using breathing sounds and blood oxygen saturation to detect each event and then estimate the overall AHI.

Methods: Subjects with suspected OSA underwent overnight PSG. We developed a small wearable device to record respiratory sounds over the neck simultaneously with the PSG. We used drop in oxygen saturation (SaO2) as the first signature of the event. Later, features such as energy and duration of breaths and snores were used identify an event and estimated the AHI. Thus, we have determined the events associated with the drop in SaO2. Hence, a misclassification of an event when the number of event is really low could largely affect the overall sensitivity and specificity. To overcome this problem, we calculated the sensitivity and specificity of detecting each event in different severities of OSA such as no OSA: AHI ≤5, mild to moderate OSA: 5<AHI <30 and severe OSA: AHI ≥30.

Expected results: Data from 32 subjects, age: 49.7±15.4 years, BMI: 28.0±3.5 kg/m2 were investigated. A high correlation was observed between the estimated AHI and PSG-AHI (r = 0.95). The sensitivity was approximately 80% for the groups with AHI ≥30 which depicts that the events associated with the drop in SaO2 were classified correctly. For AHI ≤5, specificity of respiratory events’ detection was approximately 80% which means the algorithm was able to determine the drop in SaO2 which were not corresponding to a true event in almost 80% accuracy. The sensitivity and specificity for the mild to moderate OSA groups is around 60 and 75%, respectively.

Conclusion: To best of our knowledge, this is the first study which tests the accuracy of their algorithm to classify the occurrence of each respiratory event. Once validated, this algorithm can be used to calculate postural or sleep related OSA along with the AHI, which will help the physicians to determine adequate treatments.