Artificial Intelligence Diagnosis of Obstructive Sleep Apnea Using Overnight Pulse Oximetry: A Systematic Review and...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Sleep Medicine Diagnostics

Background:

• Obstructive sleep apnea (OSA) is highly prevalent (38%) but significantly underdiagnosed (>90%).
• Current gold-standard diagnosis (polysomnography) is resource-intensive and inaccessible in primary/acute care.
• Advancements in artificial intelligence (AI) offer potential for novel diagnostic approaches using pulse oximetry.

Purpose of the Study:

• To conduct a meta-analysis evaluating the diagnostic accuracy of AI models trained on pulse oximetry data for OSA.
• To compare the performance of different AI model types and assess the impact of apnea-hypopnea index (AHI) cutoffs.

Main Methods:

• Systematic literature search across major databases (Medline/PubMed, Embase, Scopus, Web of Science, IEEE Xplore) up to January 2026.
• Inclusion of 25 studies (23,171 participants) evaluating AI-oximetry models against AHI.
• Bayesian bivariate meta-analysis, meta-regression, publication bias assessment (selection model), and risk of bias/quality evaluation (QUADAS-2, GRADE).

Main Results:

• AI-oximetry models achieved pooled sensitivity of 91.1% and specificity of 88.4%.
• Neural network classifiers showed highest accuracy (sensitivity 92.7%, specificity 91.3%).
• Deep learning models outperformed domain expert approaches; accuracy varied with AHI cutoffs but remained robust.

Conclusions:

• AI-oximetry models demonstrate high diagnostic accuracy for OSA, comparable or superior to traditional methods.
• This meta-analysis provides the first quantitative synthesis of AI oximetry performance, supporting its potential for scalable OSA screening.
• Further external validation is recommended before widespread clinical implementation in diverse settings.