Blood-Oxygen and Stress Metrics: Which Wearable Numbers Mean Anything

Nearly every wearable sold today, from a £40 fitness band to a £700 flagship smartwatch, reports a blood-oxygen percentage and some flavour of a stress score. Both numbers come from the same piece of hardware — a PPG sensor never designed as a medical instrument, repurposed through algorithm and marketing into a headline health feature. Some of what these sensors report is genuinely useful. Some of it is a plausible-looking number with far less rigour behind it than the confident decimal-point display suggests. The gap between the two is worth understanding before trusting either metric with a real health decision.

How PPG actually works, briefly

Photoplethysmography shines light — usually green, sometimes a combination of green and infrared or red — into the skin and measures how much of it scatters back to a photodiode. Blood absorbs more light during the pulse of each heartbeat than between beats, so the fluctuating signal maps to heart rate reasonably reliably; this part of PPG is mature, well-validated technology, and heart-rate accuracy from wrist and finger wearables has improved to the point that it’s genuinely close to chest-strap accuracy at rest and during steady effort.

SpO2 estimation asks more of the same basic sensor. It compares the absorption ratio of red and infrared light to estimate the proportion of oxygenated versus deoxygenated haemoglobin in the blood — the same underlying physics a clinical fingertip pulse oximeter uses. The difference is calibration and validation rigour: a dedicated medical pulse oximeter is tested and certified against a known reference standard before it can be sold as a medical device, while a consumer wearable’s SpO2 feature is explicitly sold as “not intended for medical use” in the fine print of nearly every product that offers it — a distinction that matters enormously and gets lost in the marketing screenshot showing a clean 98% reading.

Where SpO2 on a wearable is known to fall short

The specific, well-documented limitation of consumer wrist and finger PPG SpO2 sensing is accuracy degradation on darker skin tones, because melanin absorbs light in a way that interferes with the red/infrared absorption ratio the algorithm depends on. This is a well-documented concern, the subject of public scrutiny of pulse oximetry broadly, including clinical-grade devices, following documented research showing systematic overestimation of blood oxygen in patients with darker skin during the Covid-19 pandemic, when pulse oximeter readings were being used for real triage decisions. Consumer wearables inherit the same fundamental sensor physics and the same limitation, and manufacturers across the category have been slow to prominently disclose the scale of that accuracy gap in consumer-facing marketing, even as the underlying research on pulse oximetry accuracy across skin tones has become well established.

Motion is the other consistent failure mode. SpO2 readings on a wearable are typically taken during stillness — most watches actively prompt you to hold still for the measurement — because motion artefact swamps the subtle absorption-ratio signal the algorithm depends on. A reading taken during any real movement is essentially noise, which is why sleep-time SpO2 tracking (measuring during genuine stillness) tends to be the more consistent implementation across the category compared to on-demand daytime spot checks.

Stress and HRV scores: a different kind of estimate

Stress scores — Garmin’s, Amazfit’s, Fitbit’s, Samsung’s, each under a different brand name — are built primarily from heart-rate variability (HRV), the beat-to-beat variation in time between heartbeats, which genuinely does correlate with autonomic nervous system state and is a legitimately studied physiological signal in sleep and exercise-physiology research generally. The problem is not the underlying signal; it’s the leap from “HRV pattern consistent with sympathetic nervous system activation” to a single “stress” number displayed on a screen, since the same HRV signature that shows up during genuine psychological stress also shows up during a hard workout, digesting a large meal, recovering from illness, or simply waking up — all states an algorithm reading only HRV cannot reliably distinguish from each other without additional context the sensor doesn’t have.

This is why stress scores across the entire wearable category are prone to the same well-known false positive: reporting “high stress” shortly after a genuinely good, hard workout, because the cardiovascular signature looks similar to the algorithm even though the subjective experience is the opposite of stress. It is a known, structural limitation of HRV-only stress scoring rather than a defect unique to any single brand, and manufacturers have made real efforts to correct for it — factoring in recent activity data to avoid flagging exercise as stress — with mixed, incomplete success.

What each major platform actually gets right

Whoop’s Recovery score benefits from a genuinely larger accumulated dataset and longer field-validation history than most rivals, drawing on years of a large, engaged subscriber base whose feedback has helped refine edge cases — illness, travel, altitude — that a newer algorithm hasn’t had the same volume of real-world correction against. Garmin’s Body Battery has a similarly long refinement history, having been iterated across roughly six years of watch generations. Oura’s Readiness score leans on the ring’s continuous overnight temperature sensing as an additional input beyond HRV alone, which gives it a genuinely useful extra signal — illness detection specifically — that pure HRV-based scores on other platforms don’t have access to. None of these should be read as clinically validated diagnostic tools; all are legitimately useful as personal trend indicators, which is a meaningfully different and lower bar than what the polished in-app presentation sometimes implies.

What none of these platforms should be trusted for

No consumer wearable’s SpO2 or stress score is a substitute for a medical-grade pulse oximeter or a clinical stress or anxiety assessment, and every manufacturer’s own regulatory disclosures say exactly this in language most users never read. Anyone managing a real respiratory condition, sleep apnoea investigation, or diagnosed anxiety disorder should treat these features as a rough personal trend line worth mentioning to a doctor, never as a diagnostic tool, and should not delay seeking medical attention for a genuine symptom because a wearable’s reading looked reassuring. This is the single most important caveat in this entire product category, and it is worth repeating precisely because the polished, decimal-point precision of the in-app display actively works against readers internalising it.

How manufacturers actually validate these features

Regulatory clearance is where the gap between marketing confidence and underlying rigour becomes clearest. A small number of wearable SpO2 and even ECG-adjacent features have obtained specific regulatory clearances in certain markets for narrow, defined uses — the kind of clearance that requires real validation data submitted to a health regulator. The much larger set of health-adjacent features on the market — general stress scores, wellness-oriented SpO2 spot checks, readiness scores — ship as general wellness features rather than medical devices, a regulatory category with a dramatically lower evidence bar. This isn’t a hidden scandal; it’s disclosed, if quietly, in the terms every wearable maker publishes. The practical takeaway for buyers is to check specifically whether a given feature carries a genuine medical clearance for a defined use, rather than assuming the presence of any given sensor implies clinical-grade validation behind it.

Third-party academic validation studies exist for some of this category’s flagship health features, comparing wearable readings against clinical reference equipment in controlled settings, and the published results are consistently more measured than the marketing copy: generally reasonable heart-rate accuracy at rest and steady effort, wider and more variable error margins for SpO2 specifically, and stress/readiness scores validated more as showing plausible correlation with known physiological states than as precise, individually diagnostic numbers. This is exactly what should be expected of a wellness feature built on repurposed consumer sensor hardware, and it is a genuinely useful, if unglamorous, category of product — provided it is judged against that standard rather than a medical one.

The marketing-versus-disclosure gap, concretely

The clearest way to see the gap between marketing confidence and underlying disclosure is to compare the product page against the terms of use for the same feature. A product page will show a clean SpO2 reading in a hero image with no caveat visible; the same company’s terms of service, usually several taps away, will state plainly that the feature is not intended to diagnose, treat or monitor any medical condition. Both documents are accurate to what the company can defend, but only one of them is what most buyers actually read before purchasing, which is precisely why the accuracy limitations covered in this piece are less widely understood than they should be given how openly they’re actually disclosed in the fine print.

Using these numbers sensibly day to day

The practical approach that gets the most genuine value from these features without over-trusting them is to watch trends rather than single readings. A single SpO2 spot check of 94% means very little on its own, given the sensor’s known variability; a sustained downward trend across several nights, alongside other symptoms, is worth paying attention to and mentioning to a doctor. The same logic applies to stress scores: a single “high stress” alert after a hard gym session is very likely a false positive from the exercise-versus-stress confusion covered above; a stress score that stays elevated across several consecutive days, disconnected from any obvious exercise or life event, is a more meaningful signal worth taking seriously as a nudge to actually check in with how you’re doing.

This trend-over-spot-reading approach is, not coincidentally, exactly how the more credible platforms in this category (Whoop’s Recovery, Oura’s Readiness, Garmin’s Body Battery) are designed to be read by the companies that built them — multi-day graphs and rolling baselines are the actual product, and the single daily number is really a summary of that trend rather than a precise instantaneous measurement, however precisely it’s displayed on the screen.

The verdict

Buy — as a trend line, never as a diagnosis. Heart-rate tracking from modern PPG wearables is genuinely accurate enough to trust for training and general fitness purposes. SpO2 readings are worth having as a rough overnight indicator but carry known accuracy limitations, particularly across skin tones, that no amount of algorithm refinement has fully closed industry-wide. Stress and readiness scores are legitimately useful personal trend indicators once you learn each platform’s specific false-positive patterns (mistaking hard exercise for stress, in particular), but none of them, from any manufacturer, is validated to a medical diagnostic standard, and all of them say so in the small print.

Worth reading alongside Oura Ring Gen 4: the sleep promise after three months for how one specific readiness algorithm performs in sustained daily use, and Whoop vs a £40 band: does subscription fitness earn it? for how much the algorithm maturity behind these scores is actually worth paying a subscription for.