Computer models varied widely in detecting obstructive sleep apnea
A review of adult studies found widely varying accuracy for artificial intelligence models assessed against laboratory sleep testing. Reliability across clinical settings remains uncertain.
Based on the published abstract. The full paper may contain additional methods, results and limitations.
The 30-second takeaway
Across adult studies using laboratory sleep testing as the reference, computer models varied widely in detecting obstructive sleep apnea. Some reported high accuracy, but this review does not establish reliable performance across clinics. These results concern agreement with a diagnostic reference; they do not demonstrate improvements in patients' health.
This abstract-based explanation cannot assess individual study populations or testing procedures in detail. The reported accuracy range is not a pooled estimate, and the authors identify performance variability and difficulty understanding model decisions.
What researchers found
Range of reported sleep-apnea diagnostic accuracies
Adult studies assessed models against laboratory sleep testing. Evaluation timeframes were not reported. These percentages describe accuracy across studies, rather than a pooled estimate.
What served as the reference
Eligible studies used laboratory sleep recordings as the diagnostic reference. Models drew on multiple clinical features, so the review examined how accurately their assessments identified sleep apnea against that reference.
What the range leaves unresolved
The review included different model architectures, with the highest reported accuracies involving hybrid and deep learning models. That observation does not establish a dependable winner; the authors call for diverse training data to improve generalizability.
The original publication
Diagnostic accuracy of artificial intelligence for obstructive sleep apnea detection: a systematic review.
Haghighat S, Joghatayi M, Issa J et al.
BMC Med Inform Decis Mak · 2025
- PubMed ID
- 40722158
- Record checked
AI-assisted research and writing. This explains one selected publication; it is not a complete review of everything known. Our approach.
One more question, understood.
Keep track of the research you’ve explored.