Wearable sleep apnea research: what PPG studies show about breathing in sleep

Obstructive sleep apnea is common and often missed. A 2019 analysis estimated that close to a billion adults aged 30 to 69 worldwide have some degree of it, and several hundred million have moderate to severe forms 1. Most do not know. The standard test, an overnight sleep study, is accurate but limited by cost and access. That gap is why wearable researchers have spent a decade asking whether a wrist or finger sensor worn at home can help.

On our map of wearable PPG research, this topic spans two clusters: breathing rate, with three studies, and sleep and sleep-disordered breathing, with five. This review reads both together, adds the regulatory and newer validation work, and sets out how a team could design a sleep study with raw PPG. For a general introduction, see this overview of PPG and sleep apnea.

How PPG picks up breathing

PPG is a pulse signal, but breathing leaves fingerprints on it. Each breath changes the pressure inside the chest, the return of blood to the heart and the activity of the autonomic nervous system. Those changes show up in three ways 2:

• Baseline wander. The slow level of the signal drifts up and down with each breath.
• Amplitude modulation. The height of each pulse rises and falls.
• Frequency modulation. The time between beats shortens on the in-breath and lengthens on the out-breath, an effect called respiratory sinus arrhythmia.

How strongly each one shows up depends on the person, their age, posture and breathing pattern, which is why the best methods combine all three 2. One study fused several of these modulations to estimate breathing rate breath by breath rather than as a slow average 3.

Breathing rate from PPG: how good is it?

The largest benchmark compared 314 algorithms for estimating breathing rate from ECG and PPG against a reference 4. The headline findings still shape the field. Breathing rate can be estimated from PPG. ECG-based estimates performed better than PPG-based ones. The best PPG methods had limits of agreement of roughly 5 to 7 breaths per minute either side of the reference, and performance varied widely between algorithms.

For wearables, that means two things. Breathing rate is most reliable during rest and sleep, when motion is low. And the algorithm matters as much as the sensor, so a team that keeps the raw waveform can apply and compare methods rather than accept one fixed output.

Breathing rate also has a role in sleep research beyond apnea. Because motion is low during sleep, overnight breathing rate is one of the steadier signals a wearable can produce, and researchers use it as a trend across nights, for example during illness, training blocks or changes in sleep habits. A single night's value matters less than how it moves relative to a person's own baseline.

What sleep apnea screening studies found

The five studies in the map's sleep cluster show the range of approaches.

Estimating the apnea index from the wrist. One study trained a model on 250 clinical sleep recordings and tested it on 188 others, using wrist-worn reflective PPG alone. The estimated apnea-hypopnea index correlated moderately with the sleep study value, and the model separated severity classes with areas under the curve around 0.85 5.

Classifying events. Using PPG together with blood oxygen, another study detected apnea and hypopnea events in 96 overnight recordings, correctly flagging 75 percent of one-minute segments and separating central from obstructive events with high accuracy 6.

Respiratory arousals. In a small study of 10 patients, PPG features identified arousals linked to breathing events, with good discrimination but modest sensitivity 7.

Sleep staging. A deep learning model staged sleep from finger PPG in 894 people referred for suspected sleep apnea. Accuracy was about 80 percent for three stages (wake, non-REM, REM) and fell as more stages were added 8. Staging matters because breathing events are counted per hour of sleep, not per hour in bed.

Blood pressure surges during events. Using pulse transit time, one study in 32 patients compared pressure changes during obstructive events in REM and non-REM sleep 9. This links the sleep cluster to the blood pressure research review.

Newer work has moved to consumer form factors. A single-center study of a PPG smartwatch recruited 119 patients and compared the watch with a sleep study or home test. It reported high accuracy for separating moderate to severe cases from the rest, but also excluded a share of participants for poor PPG quality and called large-scale validation imperative 10.

There is also a related, older technology. Peripheral arterial tonometry measures changes in finger arterial tone and is used in some home sleep tests. A meta-analysis of 14 studies with 909 patients found strong correlation between its apnea index and sleep studies 11. It shows that a peripheral pulse signal can carry sleep apnea information when the device and analysis are validated for that purpose.

For a deeper technical look at the methods, see this review of PPG sleep apnea detection.

The first regulated wrist feature

In 2024 the FDA granted De Novo authorization to Samsung's Sleep Apnea Feature 12. Its public decision summary is useful reading for anyone working in this area, because it shows what regulators asked for.

The feature uses PPG-derived blood oxygen from the wrist plus motion data. It looks for repeated relative drops in oxygen across two qualifying nights within ten days, and then tells adults aged 22 and older whether they show signs of moderate to severe obstructive sleep apnea. It is an over-the-counter tool. Its labeling says it cannot detect central sleep apnea, should not replace diagnosis by a clinician, and is not intended to help clinicians diagnose sleep disorders.

In the clinical study, 620 people enrolled and the feature was compared with physician-scored polysomnography. Sensitivity was 82.7 percent and specificity was 87.7 percent. The summary notes that specificity did not meet the predefined acceptance criterion, and that about 17 percent of nights did not produce enough usable data 12. Even a well-resourced, regulated program had a meaningful share of unusable nights, and its result is a prompt to seek care.

For context on how blood oxygen from the wrist behaves overnight, see whether SpO2 is accurate at night.

Why sleep studies remain the reference

The American Academy of Sleep Medicine guideline is clear that polysomnography or a home sleep apnea test is needed to diagnose obstructive sleep apnea, and that questionnaires or prediction tools should not be used to diagnose it on their own 13. Wearable screening research is measured against those tests, not the other way round.

The reasons are practical. A sleep study measures airflow, breathing effort, oxygen and brain activity directly. A wearable infers breathing indirectly from pulse and oxygen changes, and it cannot always tell an apnea from a movement, a position change or a poor signal.

How to design a sleep study with wearable PPG

Pick a reference. Plan at least some nights with simultaneous polysomnography or a validated home sleep test, scored by people blind to the wearable data.

Keep raw red and infrared. Blood oxygen methods depend on red and infrared light, and breathing methods depend on beat-level timing and amplitude. As an example, Sensor Bio V2 records raw green, red and infrared PPG at 25, 50 or 100 Hz with motion data, and partner teams can export the waveforms for their own analysis. V2 is a general wellness device and is not FDA cleared, and it does not screen for or diagnose sleep apnea.

Report data loss. Track and publish the share of nights or hours with unusable signal, and whether it differs by skin tone, wrist size or sleep position.

Count per hour of sleep. Use sleep staging or a clear rule for sleep time, since events per hour in bed and per hour asleep differ.

Plan multiple nights. Night-to-night variability is high. The regulated feature above needed two qualifying nights for a reason.

Separate breathing rate from events. Breathing rate tracking and apnea screening are different questions with different references and error profiles. For how SpO2 is used more broadly, see SpO2 monitoring.

How these signals are measured: the science of PPG. For every research area, return to the wearable PPG research map.

For research teams

Put the raw signal in your study or product

Designing an overnight study? We can help you plan raw red and infrared PPG capture and export.

Talk to Sensor Bio

Limits and open questions

• Most screening studies were done in people already referred to sleep clinics, where sleep apnea is more common than in the general population. Performance in everyday users may differ.
• Many studies used finger sensors during a sleep study. Wrist results need their own validation.
• Central and obstructive events, and hypopneas, are harder to separate from wrist signals than from a full sleep study.
• Unusable nights are common and not always reported.
• Few studies report results across skin tones in enough detail.