Machine Listening for Obstructive Sleep Apnea Diagnosis: A Bayesian Meta-Analysis.

BACKGROUND Among 1 billion patients worldwide with obstructive sleep apnea (OSA), 90% remain undiagnosed. Their main barrier is the overnight polysomnogram, which requires specialized equipment, skilled technicians and inpatient beds available only in tertiary sleep centers. Recent advances in artificial intelligence (AI) have enabled OSA detection using breath sound recordings. RESEARCH QUESTION What is the diagnostic accuracy and how can we optimize machine listening for OSA? STUDY DESIGN AND METHODS PubMed, Embase, Scopus, Web of Science and IEEE Xplore were systematically searched. Two blinded reviewers selected studies comparing the patient-level diagnostic performance of AI approaches using overnight audio recordings, versus conventional diagnosis (apnea-hypopnea index [AHI]) using a train-test split or k-fold cross-validation. Bayesian bivariate meta-analysis and meta-regression were performed. Publication bias was assessed using a selection model. Risk of bias and evidence quality were assessed using QUADAS-2 and GRADE. RESULTS From 6,254 records, we included 16 studies (41 models) trained/tested on 4,864/2,370 participants. No study had a high risk of bias. Machine listening achieved a pooled sensitivity (95% credible interval) of 90.3% (86.9-93.1%), specificity of 86.7% (83.1-89.7%), diagnostic odds ratio of 60.8 (39.4-99.9), positive and negative likelihood ratios of 6.78 (5.34-8.85) and 0.113 (0.079-0.152). At AHI cut-offs of ≥5, ≥15, ≥30: sensitivities were 94.3% (90.3-96.8%), 86.3% (80.1-90.9%), 86.3% (79.2-91.1%); specificities were 78.5% (68.0-86.9%), 87.3% (81.8-91.3%), 89.5% (84.8-93.3%). Meta-regression identified higher sensitivity for: higher audio sampling frequencies; non-contact microphones; higher OSA prevalence; train-test split model evaluation. Accuracy was equal regardless of: home smartphone versus in-laboratory professional microphone recordings; deep learning versus traditional machine learning; varying age and sex. Publication bias was not evident. The evidence was of high quality. INTERPRETATION Machine listening achieved excellent diagnostic accuracy, superior to STOP-Bang and comparable to common home sleep tests. Digital medicine should be further explored and externally validated for accessible and equitable OSA diagnosis.