Wearable Breathing Rate Misses by 1 Breath on Average
Wearable devices were 0.68 breaths a minute off on average across 84 studies. Individual readings could miss by about six.
Evidence rating for the proposition “Consumer-grade wearable devices measure respiratory rate with accuracy comparable to clinical reference methods”: Uncertain
On average, close. Any single reading, less so. A review of 84 studies found wearable respiratory rate ran 0.68 breaths per minute away from a hospital reference on average. The pooled limits of agreement were -5.65 to +6.85, so a single reading can miss by five or six breaths. Accuracy fell after exercise in a 20-person test. The evidence comes mostly from clinical or research devices, not consumer watches.
How close is a wearable's breathing rate to a hospital reading?
On average, within about one breath per minute. A 2022 review found 84 studies of wearable and contactless devices measuring heart rate, respiratory rate and oxygen saturation in hospital settings. Fifty-six studied wearables.
For wearables, the pooled respiratory rate bias was 0.68 breaths per minute (95% CI 0.05 to 1.32). The pooled 95% limits of agreement ran from -5.65 to 6.85. That range describes how far individual readings strayed.
Contactless devices had a pooled bias of 0.30 (-0.26 to 0.87) with limits of -3.94 to 4.29. Wearable studies were more often commercial devices tested in acute clinical settings. Contactless studies were more experimental and carried a higher risk of patient selection and rater bias.
1 study
- In a review of 84 studies (56 wearable, 29 contactless), pooled respiratory rate bias for wearables was 0.68 breaths per minute (95% CI 0.05 to 1.32), with pooled 95% limits of agreement from -5.65 to 6.85.
Does it hold up in sick patients?
In one wearable, yes, with a caveat. A photoplethysmography-based monitor was compared with medical devices in three groups. The first was 35 healthy volunteers against a capnograph. The second was 18 ventilated patients against a ventilator monitor. The third was 92 COVID-19 patients with active lung disease against a capnograph.
Correlations were high in all three (r = 0.991, 0.884 and 0.888). Bias was under 0.1 breaths per minute. The 95% limits of agreement were plus or minus 2.3 in healthy people and plus or minus 3.9 in the COVID-19 patients.
So the average held while the spread widened in the sickest group. The abstract does not name the device.
1 study
- A wearable photoplethysmography-based monitor was compared with medical-grade references in 35 healthy volunteers (r = 0.991), 18 ventilated patients (r = 0.884) and 92 COVID-19 patients (r = 0.888). Bias was under 0.1 breaths per minute, and 95% limits of agreement were plus or minus 2.3, 1.7 to -1.6, and plus or minus 3.9.
Does accuracy drop after exercise or during daily movement?
Yes, in the two studies that tested it. Twenty healthy people wore two stretch sensors while a spirometer measured breathing. At rest, error was 0.06 to 2.39 breaths per minute. After exercise it was 1.57 to 3.72. The authors say detection after exercise needed improvement.
A chest patch combining ECG and an accelerometer was tested in 15 elderly people. Mean absolute error was 1.02 breaths per minute during paced breathing and 1.67 during spontaneous breathing. During daily activities it was 2.03.
2 studies
- In 20 healthy people, two wearable stretch sensors had a respiratory rate error of 0.06 to 2.39 breaths per minute at rest and 1.57 to 3.72 after exercise, against a spirometer.
- In 15 elderly people, a chest patch combining ECG and accelerometer data had a mean absolute error of 1.02 breaths per minute during paced breathing, 1.67 during spontaneous breathing and 2.03 during daily activities.
Does the signal a device uses matter?
It does. Many devices estimate breathing rate from the pulse waveform or the ECG, not from airflow. A 2016 study tested 314 algorithms on healthy participants against a nasal-oral pressure sensor.
The best ECG-based algorithm had 95% limits of agreement of -4.7 to 4.7 breaths per minute and a bias of 0.0. The best photoplethysmogram-based algorithm had limits of -5.1 to 7.2 and a bias of 1.0. Algorithms did better with ECG than with the pulse signal.
Under ideal conditions in healthy people, then, even the best algorithms could miss by about five breaths on a single reading.
Is any of this about consumer watches and rings?
Very little. The studies above used clinical or research devices and the abstracts name none of the consumer brands. A review of consumer-grade camera monitors found 26 studies, and only three measured respiratory rate. Most had 30 or fewer participants. The authors said more studies are needed for respiratory rate.
A t-shirt-embedded sensor was compared with overnight polysomnography in 32 people with sleep-disordered breathing. Its computed respiratory parameters had about 95% accuracy. That is a different device category again.
What does this not settle?
It does not tell a reader how a specific watch or ring performs. It does not cover breathing rate during sleep on consumer devices, or in people with lung disease wearing a consumer device.
The exercise studies were small. The reference in each study was a clinical or laboratory device, and conditions were controlled.
The exercise and daily-movement findings come from studies of 15 and 20 people using clinical or research sensors. None of the sources reviewed here validates a named consumer smartwatch or ring against a reference respiratory rate.
Common questions
How accurate is respiratory rate on a wearable?
In a review of 84 studies, wearable respiratory rate had a pooled bias of 0.68 breaths per minute against a reference. The 95% limits of agreement ran from -5.65 to 6.85, so single readings can miss by five or six breaths.
Does breathing rate accuracy drop after exercise?
In one 20-person test of two stretch sensors, error rose from 0.06 to 2.39 breaths per minute at rest to 1.57 to 3.72 after exercise.
Does it work in people who are ill?
One photoplethysmography wearable stayed close to references in 92 COVID-19 patients (r = 0.888, bias under 0.1 breaths per minute), though limits of agreement were wider at plus or minus 3.9.
Is pulse or ECG better for estimating breathing rate?
In a test of 314 algorithms in healthy people, ECG-based algorithms did better. The best had limits of agreement of -4.7 to 4.7 breaths per minute against -5.1 to 7.2 for the best pulse-based one.
Have consumer watches been validated for breathing rate?
Not in the sources reviewed here. A review of consumer camera monitors found only three of 26 studies measured respiratory rate, and called for more studies.
Sources
- Novel wearable and contactless heart rate, respiratory rate, and oxygen saturation monitoring devices: a systematic review and meta-analysis.
- Clinical validation of a wearable respiratory rate device: A brief report. [author affiliations include Biobeat]
- Prospective observational study of 2 wearable strain sensors for measuring the respiratory rate.
- Ambulatory respiratory rate detection using ECG and a triaxial accelerometer.
- An assessment of algorithms to estimate respiratory rate from the electrocardiogram and photoplethysmogram.
- Effectiveness of consumer-grade contactless vital signs monitors: a systematic review and meta-analysis. [funding listed from a ResMed-endowed research chair]
- Morphic Sensors for Respiratory Parameters Estimation: Validation against Overnight Polysomnography. [conflict statement: three authors hold shares in the company with rights to sell the sensor tested]