Artificial Intelligence Diagnosis of Obstructive Sleep Apnoea using Overnight Pulse Oximetry: A Systematic Review and Bayesian Meta-Analysis (Preprint)
Bibliographic record
Abstract
BACKGROUND: Obstructive sleep apnoea (OSA) affects 38% of the population, yet over 90% of cases remain undiagnosed. The current gold standard for diagnosis, polysomnography (PSG), requires specialised equipment, and trained personnel, making it inaccessible in primary care and acute settings. With AI advancements, oximetry-based AI models have emerged as a potential alternative for OSA diagnosis. OBJECTIVE: This meta-analysis aims to evaluate the diagnostic accuracy of AI models trained on pulse oximetry readings in diagnosing OSA. METHODS: A systematic search was conducted across Medline/PubMed, Embase, Scopus, Web of Science, and IEEE Xplore databases from inception to 3 January 2026. Studies that evaluated the diagnostic accuracy of AI models trained on SpO₂ recordings, compared to the apnoea-hypopnea index (AHI) as the reference standard were included and screened by two blinded independent reviewers. Studies that did not evaluate AI on AHI-defined OSA and non-English texts were excluded. Models were evaluated using Bayesian bivariate meta-analysis and meta-regression. Publication bias was examined using a selection model approach, while risk of bias and evidence quality were assessed with QUADAS-2 and GRADE. RESULTS: From 13,986 screened articles, 25 studies met the inclusion criteria, encompassing 23,171 participants with a mean age of 40 to 63 years, and a mean BMI of 25 to 37. AI-oximetry models demonstrated a pooled sensitivity of 91.1% (95% CrI: 89.7-92.4%) and specificity of 88.4% (95% CrI: 85.3-90.8%), with a diagnostic odds ratio (DOR) of 77.7 (95% CrI: 60.2-99.6). Neural network classifiers achieved the highest sensitivity (92.7%) and specificity (91.3%). Deep learning feature extraction models were significantly higher in sensitivity by 3.7% than domain expert-based approaches. Sensitivity decreased slightly with higher AHI cut-offs, while specificity increased by 16.6% from an AHI cut-off of ≥5 to ≥30. Sensitivity analyses showed that even with up to 40% probability of unpublished study, changes in accuracy were modest (AUC: 0.902 to 0.877). QUADAS-2 and GRADE assessments found low-moderate risk of bias with high overall quality of evidence. CONCLUSIONS: AI-oximetry models showed high diagnostic accuracy for OSA across models and AHI cut-offs, performing better than or comparably to traditional overnight oximetry and HSATs. This review provides the first pooled quantitative synthesis of AI models trained solely on oximetry data, with additional evaluations of publication bias and methodological limitations. Prior reviews were largely narrative or used alternative AI inputs other than oximetry. This study advances the field by offering a clearer and more reliable evidence base on pooled AI-oximetry performance. These findings support the potential of oximetry-based AI as a convenient and scalable tool for OSA screening and diagnosis, with potential real-world applications in both primary care and inpatient settings for early identification of high-risk patients. Prospective external validation in diverse populations and low-prevalence settings is still needed before widespread real-world use. CLINICALTRIAL: This review was registered on PROSPERO (CRD42025648556) and did not receive any source of funding.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Direct model labels (unvalidated)
Per-model category and study-design labels from the labeling rounds. They are machine output, unvalidated, and the disagreement between models ships as data. No study design here is MEDLINE-validated yet.
| Model arm | Categories | Study design | Confidence |
|---|---|---|---|
| gemma | no category Domain: not available · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Meta-analysis | low |
| gpt | no category Domain: not available · Genre: Review About the Canadian research system: no · About a Canadian topic: no | Meta-analysis | low |
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.025 | 0.072 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.015 | 0.031 |
| Bibliometrics | 0.009 | 0.007 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.004 | 0.002 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.003 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedLabeled directly by 2 models reading the full record.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".