How Wearable Metrics Compare to Clinical-Grade Measurements

Consumer wearables now report heart rate, blood oxygen (SpO2), and detailed sleep-stage breakdowns as a matter of course, often presented with the same visual confidence as a hospital monitor readout. But a fitness ring or smartwatch is not a medical device, and the two categories aren't held to the same accuracy standard. We went looking for the actual validation research — studies that put a wearable on one wrist and a clinical reference device (ECG, arterial blood gas analysis, or polysomnography) on the same person at the same time — to see where the numbers really land.

Heart rate: generally the strongest metric, with real brand gaps

Heart rate is the wearable metric with the most validation research behind it, and it's also the one that tends to hold up best against a true ECG reference — but "tends to" is doing real work in that sentence. A study using a Polar H10 chest-strap ECG reference monitored five devices (Garmin Fenix 6, Oura Generation 3, Oura Generation 4, Polar Grit X Pro, and WHOOP 4.0) across 536 nights of sleep in 13 healthy adults. For resting heart rate, the two Oura generations showed the closest agreement with the ECG reference (concordance correlation coefficients of 0.97–0.98, mean absolute percentage error under 2%), while WHOOP and the Polar device showed more error, in the 2.7–3.0% MAPE range.^[1] Heart rate variability — a more sensitive, harder-to-measure signal — showed a wider spread across devices, with the Polar Grit X Pro's error climbing to over 16% on average in that same study.

Separately, a 100-patient study of people with COPD or interstitial lung disease found a very strong correlation between Apple Watch Series 6 heart rate readings and a conventional commercial pulse oximeter (r = 0.995).^[2] That's a strong result, but it's worth noting this was daytime, resting, single-timepoint measurement in a clinic — not the harder case of heart rate during exercise or rapid movement, where photoplethysmography (the light-based sensor most wrist wearables use) is more prone to motion artifact.

SpO2: the metric with the biggest gap between "fine most of the time" and "clinically risky"

Blood oxygen tracking is where the clinical stakes are highest and the wearable-to-clinical gap is most consistently documented. In that same 100-patient lung-disease study, Apple Watch SpO2 readings correlated with a commercial oximeter at r = 0.81 — respectable, but noticeably weaker than the heart rate correlation, and correlation isn't the same thing as interchangeable accuracy.^[2] A separate diagnostic-accuracy study compared several finger-worn and wrist-worn pulse oximeters against arterial blood gas analysis (the true clinical gold standard) during controlled hypoxia induction. All the finger-worn devices met the ISO accuracy standard under stable conditions, with sensitivity for detecting hypoxemia ranging from 0.87 to 0.97 depending on the device — but the same study found that motion (tapping, rubbing, turning pages) degraded SpO2 accuracy past the acceptable error threshold for at least one device tested, and that accuracy in general declines during rapid desaturation, which is exactly the scenario where an accurate reading matters most.^[3]

The practical takeaway from this body of research: a wearable SpO2 reading is reasonable as a rough, resting-state indicator, but it is not validated as a substitute for a clinical pulse oximeter or blood gas test when oxygen levels are actually changing quickly or when the reading is being used to make a medical decision.

Sleep tracking: the hardest problem, and the most brand-dependent

Sleep staging — classifying a given moment as awake, light, deep, or REM sleep — is the area where consumer wearables diverge most from the clinical reference, polysomnography (PSG), the multi-sensor, technician-scored sleep-lab study. A 2023 multicenter study tested 11 consumer sleep trackers (five wrist/ring wearables, three bedside "nearable" devices, and three phone-based "airable" apps) against 543 hours of polysomnography recordings from 75 participants. Performance varied enormously: macro F1 scores for epoch-by-epoch sleep-stage agreement ranged from roughly 0.26 at the low end to about 0.69 at the high end, depending on the specific device and sleep stage.^[4] No single device was best across every stage — one wearable was strongest at identifying wake and REM, a different one was strongest at identifying deep sleep, and devices using accelerometry alone (motion only, no heart-rate sensor) generally scored lowest, around 65% accuracy for even the coarser sleep-versus-wake distinction.

It's also worth noting that polysomnography itself isn't a perfect, unimpeachable standard — two human technicians independently scoring the same overnight PSG recording typically agree with each other only about 83% of the time.^[4] That doesn't rescue wearable accuracy, but it's a useful reminder that "clinical grade" means "the best available reference," not "flawless."

Why the accuracy gap exists

Most wrist and ring wearables estimate heart rate and SpO2 using photoplethysmography — shining light into the skin and measuring how blood volume changes absorb it — rather than the direct electrical signal an ECG reads or the arterial blood sample a lab oximeter uses. PPG is a genuinely clever, non-invasive workaround, but it's more sensitive to skin tone, tattoo ink, motion, ambient light, sensor placement, and blood flow at the measurement site (a finger clip reads differently than a wrist or a ring) than a hospital-grade sensor is. Sleep staging adds another layer of difficulty: PSG uses direct measurement of brain waves (EEG), eye movement, and muscle tone, while wearables infer sleep stage indirectly from movement and heart-rate patterns — a fundamentally lower-resolution proxy for what's actually happening in the brain.

What this means if you use one

The validation research supports using a wearable to watch your own trends — is resting heart rate drifting up, is time in bed shrinking, is nightly variability changing — where the device mostly just needs to be internally consistent with itself night to night. It does not support treating a single wearable reading as clinically equivalent to a doctor's-office or sleep-lab measurement, particularly for SpO2 during a suspected medical event or for a precise breakdown of how much deep or REM sleep you got last night. And because accuracy differs meaningfully brand to brand and generation to generation in the studies above, a claim about "wearables" in general should be treated skeptically — the real answer is almost always "it depends which one."