Sleep breathing sounds recorded by smartphones could be used in building prediction models for obstructive sleep apnea (OSA), a new study suggests, although additional verification of real-life recordings via a wide variety of smartphone devices has been proposed.
Previous research has suggested that breath sounds could be considered potential biomarkers for OSA, which can be recorded through the microphones found in most smartphone devices.
Although laboratory polysomnography (PSG) is still considered “the gold standard” in diagnosing OSA, researchers led by Sung-Woo Cho, MD, of Soul National University College of Medicine, suggested that the high cost of this practice access to the general public.
With this study, Cho and colleagues considered the possibility of using smartphone-recorded breathing songs to predict OSA, and whether minimizing functions and sound processing would affect prediction performance.
Registration and full night PSG
Patients who visited the sleep clinic of the ENT Head and Neck Surgery Department of Seoul National University Bundang Hospital for snoring or sleep apnea from September 2015 to September 2019 were included in the study.
Eligible patients participated in an overnight PSG in the lab.
Before that, the team defined apnea as an interruption in airflow for at least 10 seconds, while hypopnea was defined as a decrease in airflow of more than 50% for at least 10 seconds or a moderate decrease in airflow for at least 10 seconds associated with it. with with excitement or oxygen desaturation (<4%).
Throughout the night PSG, smartphones were used to collect audio recordings of breathing sounds, which were recorded for all stages of sleep from start to finish.
Sound analyzes were started with the conversion of audio files to WAV file formats and audio data was discarded during the first 30 minutes of recording. The analysis was stopped after 6 hours.
The researchers used Audacity to filter out sounds and jAudio was used to extract sound characteristics.
Binary classifications were performed for 3 different threshold criteria related to the Apnea-Hyppnea Index (AHI) threshold of 5, 15 or 30 events/h, and 4 regression models were constructed to predict the true AHI, including noise reduction without feature selection , noise reduction with function selection, neither noise reduction nor function selection, and function selection without noise reduction.
Data was analyzed from September 2019 to September 2020.
The findings
A total of 423 patients participated in the study with a mean age of 48.1 years. A majority of patients were male (84.1%).
Researchers noted that the OSA prediction was 88.2% (κ = 0.46) for 5 events/hr, 82.3% (κ = 0.59) for 15 events/hr, and 81.7% (κ = 0.63) for 30 events/hr, while the sensitivity was 90.8. %, 87.3% and 83% respectively.
In addition, the positive predictive value for smartphone-recorded sleep breathing sounds was 95.8%, 89.3%, and 82% for AHI thresholds of 5, 15, and 30 events/hr, respectively.
Accuracy, sensitivity, PPV, F1 score and area under precision call were found to decrease as the OSA threshold increased, while the κ value, specificity and negative predictive value (NPV) increased along with the threshold.
In general, the prediction models resulted in reasonable prediction performance, with noise suppression being considered “not mandatory” for good prediction performance.
“The skewness of patient distribution in model development and the inability to distinguish sleep and wake states need to be addressed to improve the performance of the prediction model,” the team wrote. “Future research should expand to include real-life smartphone recordings at home using various smartphone devices.”
The study, “Sleep Apnea Prediction Models Evaluating Smartphone-Recorded Sleep Breathing Sounds”, was published online in JAMA Otolaryngology Head and Neck Surgery.