Sleep trackers rarely miss sleep apnoea, but they often cry wolf

Obstructive sleep apnoea is a common disorder linked with significant illness and death, yet diagnosing it has traditionally required a full overnight sleep study, known as polysomnography. In recent years a wave of devices that estimate apnoea from blood oxygen readings has arrived, from dedicated medical kit to consumer wearables people already own. The obvious question is whether any of them can be trusted against the gold standard. If they can, screening becomes far easier; if they cannot, they risk false reassurance or a flood of unnecessary referrals.

The authors swept the published literature for studies that tested oximetry-based devices against polysomnography for diagnosing obstructive sleep apnoea in adults, then pooled the results. Eighteen studies made the cut. The devices were split into two subgroups for analysis: consumer wearables on one side and medical devices on the other. For each group the team calculated how often the devices correctly identified people with apnoea, how often they correctly cleared people without it, and overall accuracy.

Overall sensitivity was 97%, meaning the devices picked up nearly every true case, but specificity was only 63%, meaning more than a third of people without apnoea were wrongly flagged. Consumer wearables managed 93% sensitivity with the same 63% specificity, while medical devices matched the overall figures at 97% and 63%. Accuracy measured by area under the curve and post-test probability both came out at 90%. In short, the devices are excellent at ruling apnoea out in high-risk adults and mediocre at ruling it in.

This pooled analysis gives the clearest picture yet of where oximetry devices sit in the diagnostic pathway: useful as a first screen, not as a replacement for the sleep laboratory. The authors note that the low specificity raises concerns about unnecessary healthcare costs if positive readings are taken at face value. It also narrows the gap between consumer wearables and medical devices, since both subgroups showed broadly similar performance. Open questions remain about which specific devices perform best and how they fare outside high-risk groups.

If you snore heavily, wake unrefreshed or have been told you stop breathing at night, this is a reason to take a wearable's oxygen data seriously as a first signal. A consistently clean reading carries real weight, because these devices missed very few true cases in the pooled data. A flagged reading, though, is wrong often enough that it should be read as a prompt to pursue a formal sleep study rather than a verdict. The sensible frame is screening tool, not diagnosis.