Galaxy Watch 4 Oxygen Was About 2 Points Off in Sleep Tests
A Samsung wrist sensor came within 2.3 percentage points of the lab finger sensor overnight. Error grew as sleep apnea got worse.
This article covers how a wrist-worn oxygen sensor compared with a finger pulse oximeter in sleep testing, and what is known about the accuracy limits of finger oximeters. It does not cover other watch brands' sleep apnea features or advice about oxygen readings.
Fairly close on average, worse when breathing is disrupted. In 97 adults with sleep problems, the Galaxy Watch 4 had a root mean squared error of 2.3% against the sleep-lab finger sensor and a bias of -0.2%. Error rose from 1.65% in people without apnea to 2.93% in severe apnea. The watch also rejected 26.5% of its data. The finger sensor is not a perfect benchmark either. Its errors grow with low blood flow and darker skin.
How close is a wrist oxygen reading to the finger sensor?
Within about two percentage points, in one sleep study. Ninety-seven adults with sleep disturbances wore a Samsung Galaxy Watch 4 during a polysomnography night. A finger transmittance oximeter served as the reference.
The watch had an overall root mean squared error of 2.3% and a bias of -0.2%. The Bland-Altman plot showed good agreement. The authors state the device complies with FDA and ISO standards.
The watch discarded 26.5% of its data. The causes were fluctuating contact pressure and readings below 70% saturation.
1 study
- In 97 adults with sleep disturbances, the Samsung Galaxy Watch 4 had a root mean squared error of 2.3% and a bias of -0.2% against a transmittance pulse oximeter during polysomnography. The data rejection rate was 26.5%.
Does the error change when sleep apnea is severe?
Yes. The error rose with apnea severity. It was 1.65% in the 18 people with normal breathing. It was 1.76% in 21 with mild apnea, 1.93% in 23 with moderate apnea and 2.93% in 35 with severe apnea.
The same paper tested a desaturation count from the watch. An oxygen desaturation index of 5 per hour or more predicted an apnea-hypopnea index of 15 or more with 89.7% sensitivity and 64.1% specificity. Accuracy was 79.4%.
The pattern matches the sleep apnea detection results in whether a watch's oxygen sensor can detect sleep apnea. Sensitivity is high and specificity lower.
Is the finger sensor a perfect benchmark?
No. Finger oximeters have their own limits. A review of 22 studies of oximeter accuracy in poorly perfused adults found 75% of models accurate in those conditions. Newer models with more complex algorithms did better than older ones. Earlobe placement seemed more accurate than the fingertip. Only one study controlled for skin pigmentation.
Motion is another problem. A laboratory test of 20 oximeters during hand motion and low oxygen scored each on how often its reading stayed within 7% of a control. The Masimo SET scored 94%, the Nellcor N-395 69% and an older Criticare 5040 28%.
Circulation is covered in how poor circulation affects blood oxygen sensor accuracy.
2 studies
- In 22 studies of adults with poor peripheral perfusion, 75% of oximeter models were deemed accurate. Newer models using complex algorithms were more likely to be accurate, earlobe placement seemed more accurate than fingertip, and only one study controlled for skin pigmentation.
- In a laboratory test of 20 pulse oximeters during motion and low oxygen, the share of time within 7% of control was 94% for the Masimo SET, 69% for the Nellcor N-395 and 28% for the Criticare 5040.
Do darker skin tones change the reading?
On finger devices, yes. In 146 healthy people, missed hypoxemia under low perfusion was 1.1% in light skin, 8.2% in medium and 21.1% in dark skin. The devices were the Nellcor N-595 and Masimo Radical 7.
For wrist devices the direct evidence here is a simulation. A Monte Carlo model of reflectance oximeters gave errors against transmittance calibration. Root mean square error was at least 0.40% for light skin, 0.91% for moderate and 8.41% for dark skin. The authors say a one-algorithm-fits-all calibration may be insufficient and that validation is needed.
No clinical study here tests a wrist wearable by skin tone. See whether a wearable's blood oxygen reading works the same for everyone and how wearables measure blood oxygen.
What does this not settle?
The wrist result comes from one watch model in one sleep clinic population. It does not describe other watches or rings. It does not cover daytime readings, exercise or altitude.
The skin tone figures come from finger devices and from a simulation. Neither is a clinical test of a wrist wearable.
The wrist evidence is a single study of one watch model in 97 adults with sleep disturbances tested overnight against a reference oximeter. It does not establish accuracy in other brands, in daytime use, or in people with different skin tones.
Common questions
How accurate is a smartwatch blood oxygen sensor against a finger oximeter?
In 97 adults during sleep testing, the Galaxy Watch 4 had a root mean squared error of 2.3% and a bias of -0.2% against the reference finger oximeter. It rejected 26.5% of its data.
Does sleep apnea make the wrist reading less accurate?
Yes in that study. Error was 1.65% in people without apnea and 2.93% in those with severe apnea.
Is a finger pulse oximeter always accurate?
No. A review of 22 studies found 75% of models accurate in poorly perfused adults, with newer models doing better. Motion and skin pigmentation also add error.
Do darker skin tones affect finger oximeter readings?
In a 146-person study, missed hypoxemia under low perfusion was 1.1% in light skin, 8.2% in medium and 21.1% in dark skin.
Has a wrist wearable been tested by skin tone?
Not in the studies reviewed here. A simulation of reflectance oximeters found error of at least 8.41% for dark skin against 0.40% for light skin, and called for validation.
Sources
- Performance evaluation of a wrist-worn reflectance pulse oximeter during sleep.
- Accuracy of pulse oximeters in measuring oxygen saturation in patients with poor peripheral perfusion: a systematic review.
- "Motion-resistant" pulse oximetry: a comparison of new and old models.
- Low Perfusion and Missed Diagnosis of Hypoxemia by Pulse Oximetry in Darkly Pigmented Skin: A Prospective Study.
- Monte Carlo Simulation of the Effect of Melanin Concentration on Light-Tissue Interactions in Reflectance Pulse Oximetry.