Pulmonology · Medical Literature
Accuracy of Machine Learning Algorithms Based on Electroencephalogram in Sleep Apnea Detection: Systematic Review and Meta-Analysis.
Abstract
Sleep apnea (SA) is a serious sleep disorder, and its diagnostic gold standard, polysomnography, is costly and time-consuming. Electroencephalogram (EEG) signals, due to their direct correlation with neural activity and ease of extraction, represent a promising tool. Despite increasing research on machine learning (ML) and deep learning for EEG-based SA detection, model performance has not been consistently evaluated. This systematic review evaluated the accuracy of ML in detecting SA from EEG data and provided an evidence base for further clinical application and future research. Following the PRISMA-DTA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 expanded checklists, we systematically searched PubMed, Embase, Web of Science, Cochrane Library (CENTRAL), Scopus, IEEE Xplore, and ClinicalTrials.gov databases from inception to April 2026. Studies evaluating the value of ML algorithms for detecting SA based only on EEG data were included. The Quality Assessment of Diagnostic Accuracy Studies-2 and Prediction Model Risk of Bias Assessment Tool for Artificial Intelligence tools were used to assess the risk of bias in each study. Statistical analysis was performed using the mada and metafor packages in R (version 4.6.0; R Foundation for Statistical Computing) and the Meta-DiSc (version 1.4; Hospital Ramón y Cajal) software. We used GRADE (Grading of Recommendations Assessment, Development and Evaluation) to evaluate the certainty of evidence. A total of 27 retrospective studies were included. Segment-level analyses showed high diagnostic performance, with a pooled sensitivity of 0.90 (95% CI 0.85-0.94; 95% prediction interval 0.43-0.99) and specificity of 0.92 (95% CI 0.87-0.95; 95% prediction interval 0.46-0.99). The pooled area under the summary receiver operating characteristic curve was 0.95 (95% CI 0.92-0.99). Meta-regression identified EEG channel configuration, region, and validation strategy as significant sources of heterogeneity (P=.004, P=.003, and P=.046, respectively). Multichannel EEG, deep learning approaches, and hold-out validation strategies generally demonstrated better diagnostic performance. Only 2 studies evaluated patient-level diagnostic performance, which was summarized qualitatively. To our knowledge, this is the first systematic review and meta-analysis specifically focused on the diagnostic accuracy of EEG-based ML models in the detection of SA. This meta-analysis indicates that ML models based on EEG demonstrate good diagnostic accuracy in detecting SA at the segment level and show promise as tools for SA screening and clinical decision support. However, most current studies are retrospective segment-level analyses, which may overestimate the practical value of this technology in real-world clinical settings. To reliably integrate EEG-based ML models into clinical diagnostic workflows, further prospective studies incorporating full-night monitoring and patient-level validation are needed.
Clinical relevance
Sleep apnea affects a substantial global population and remains underdiagnosed due to the high cost, complexity, and limited accessibility of polysomnography, the current diagnostic gold standard. This systematic review and meta-analysis addresses the critical gap of whether EEG-based machine learning models can serve as accurate, scalable alternatives for SA screening, directly relevant to pulmonologists and sleep medicine specialists seeking practical diagnostic adjuncts. With 27 studies now synthesized under a rigorous PRISMA-DTA framework through April 2026, this represents the first meta-analysis specifically focused on EEG-derived ML diagnostic accuracy for SA, providing timely evidence as AI-assisted diagnostics enter clinical consideration.
Key findings
1. Pooled segment-level sensitivity of ML-based EEG models was 0.90 (95% CI 0.85–0.94), with a wide 95% prediction interval of 0.43–0.99, indicating high average but variable cross-study performance. 2. Pooled segment-level specificity was 0.92 (95% CI 0.87–0.95; 95% prediction interval 0.46–0.99), consistent with high discriminative ability at the signal-segment level. 3. The pooled area under the summary receiver operating characteristic (sAUROC) curve was 0.95 (95% CI 0.92–0.99), reflecting excellent overall diagnostic accuracy at the segment level. 4. Meta-regression identified three statistically significant sources of heterogeneity: EEG channel configuration (P=.004), recording region (P=.003), and validation strategy (P=.046), underscoring that methodological and technical design choices critically modulate model performance. 5. Multichannel EEG configurations, deep learning architectures, and hold-out validation strategies were associated with generally superior diagnostic performance compared to single-channel, traditional ML, and cross-validation approaches, respectively. 6. Patient-level diagnostic performance—the clinically actionable metric—was reported in only 2 of 27 included studies and could not be meta-analyzed, representing a critical evidence gap; all 27 studies were retrospective, introducing significant risk of overestimating real-world utility.
Practice implications
Clinicians and health systems evaluating AI-assisted sleep apnea screening tools should recognize that current EEG-based ML models demonstrate promising segment-level accuracy (sAUROC 0.95) but lack adequate patient-level validation, meaning they are not yet ready to replace or be formally integrated into standard diagnostic workflows as a PSG substitute. When appraising or procuring such tools, preference should be given to systems utilizing multichannel EEG with deep learning architectures and prospectively validated hold-out testing, as these features are associated with superior and more generalizable performance. Future clinical adoption should be contingent on prospective studies that incorporate full-night EEG monitoring and report patient-level sensitivity, specificity, and real-world diagnostic yield to confirm that segment-level accuracy translates into actionable clinical benefit.