Evaluating Prediction Models of Sleep Apnea From Smartphone-Recorded Sleep Breathing Sounds

• DOI: 10.1001/jamaoto.2022.0244
• Publication Date: 2022-04-14T15:00:51Z
• Authors: Sung-Woo Cho; Sung Jae Jung; Jin Ho Shin; Tae-Bin Won; Chae-Seo Rhee; Jeong-Whun Kim
• Keywords: Sleep; Feature (linguistics)

Breathing sounds during sleep are an important characteristic feature of obstructive apnea (OSA) and have been regarded as a potential biomarker. can be easily recorded using microphone, which is found in most smartphone devices. Therefore, it may easy to implement evaluation tool for prescreening purposes.To evaluate OSA prediction models smartphone-recorded identify optimal settings with regard noise processing sound selection.A cross-sectional study was performed among patients who visited the center Seoul National University Bundang Hospital snoring or from August 2015 2019. Audio recordings were routine, full-night, in-laboratory polysomnography. Using random forest algorithm, binary classifications separately conducted 3 different threshold criteria according hypopnea index (AHI) 5, 15, 30 events/h. Four regression created reduction selection input predict actual AHI: (1) without selection, (2) (3) neither nor (4) reduction. Clinical polysomnographic parameters that associated errors assessed. Data analyzed September 2019 2020.Accuracy models.A total 423 (mean [SD] age, 48.1 [12.8] years; 356 [84.1%] male) analyzed. split into training (n = 256 [60.5%]) test data sets 167 [39.5%]). Accuracies 88.2%, 82.3%, 81.7%, areas under curve 0.90, 0.89, 0.90 AHI events/h, respectively. In analysis, had not denoised only selected attributes resulted highest correlation coefficient (r 0.78; 95% CI, 0.69-0.88). The (β 0.33; 0.24-0.42) efficiency -0.20; -0.35 -0.05) estimation error.In this study, breathing used create reasonably accurate models. Future research should focus on real-life various