Backgrounds: Consumer-level sleep analysis technologies have been developed for screening obstructive sleep apnea (OSA). However, no study has confirmed the predictability of OSA based on in-home recording data performed simultaneously with level 2 polysomnography (PSG). This study aims to predict OSA based on data collected using a smartphone and level 2 PSG at home.

Methods: This study included participants who underwent unattended level 2 full-night home PSG. Breathing sounds were recorded during sleep using two smartphones, one with an iOS and the other with an Android operating system, concurrently with home PSG. In-home recordings were validated to predict OSA using a pre-developed prediction model, which was trained using acoustic signals recorded during level 1 in-laboratory PSG. Binary classifications were conducted with a threshold criterion at an apnea hypopnea index (AHI) of 15, and the performances of the sensitivity, specificity, and area under the curve (AUC) of the receiver operating characteristic (ROC) curve were computed.

Results: A total of 117 participants were included. The sensitivity and specificity for the iOS smartphone were 0.91 and 0.95, respectively, while for the Android smartphone, they were 0.84 and 0.95, respectively. The AUC was 0.93 and 0.89 for iOS and Android, respectively.

Conclusion: Our study demonstrated that it is possible to predict OSA with some accuracy from breathing sounds obtained during sleep at home. This is the first study to verify the performance of the predictive model by labeling breathing sounds with level 2 home PSG performed simultaneously with sound recordings at home.