Wearable health sensors

Wearable health sensors are consumer devices worn on the wrist, finger, or torso that record optical pulse waveforms, movement, and skin temperature continuously, and infer heart rhythm, sleep, and activity from those signals. They are among the most widely deployed health-measurement devices ever made, worn by hundreds of millions of people who are mostly not patients. Almost nothing they report has been tested against a clinical outcome; the exception is detection of atrial fibrillation, which has been studied in cohorts of several hundred thousand and cleared by regulators as a notification feature rather than a diagnosis.

§What the sensors measure

Three sensor types do most of the work. Photoplethysmography shines green light into the skin and measures how much reaches a photodiode; capillary blood volume rises and falls with each beat, so the returning signal yields interbeat intervals, and from those come heart rate, heart-rate variability, and a measure of rhythm irregularity. A triaxial accelerometer records movement, supplying step counts, fall detection, and the raw material for sleep and wake classification. Several watches add a single-lead electrocardiogram: touching the crown, a button, or the metal frame closes a circuit across the body and produces a roughly thirty-second tracing comparable to lead I of a clinical ECG.

Rings use infrared light on the finger, where perfusion is stronger and motion artefact lower than at the wrist. Some devices add skin temperature relative to a personal baseline rather than core temperature, bioimpedance, and red-and-infrared pulse oximetry.

The hardware measures light, acceleration, and voltage. Sleep stages, stress, recovery, and readiness are inferences layered on top by proprietary models, and each has to be validated separately. A device that captures the pulse waveform faithfully has established nothing about the sleep classifier built from it.

§Development history

1. 1965
   The ten-thousand-step target
   A Japanese manufacturer markets a pedometer named manpo-kei, or ten-thousand-step meter. The number was a marketing choice and survives as a default goal in software sixty years later.
2. 1980s
   Chest-strap heart-rate monitors
   Wireless electrode straps reach endurance athletes, establishing continuous cardiac measurement outside a clinic.
3. 2009
   Fitbit ships its first tracker
   A clip-on accelerometer sold direct to consumers, framed around steps and sleep rather than sport.
4. 2015
   Wrist and finger optical sensing
   The Apple Watch ships with green-light photoplethysmography; Oura's first finger-worn ring reaches backers around the same time.
5. 2018
   First cleared consumer ECG
   US regulators authorize an electrocardiogram app and an irregular rhythm notification on a general-purpose smartwatch through the De Novo pathway.
6. 2019
   Apple Heart Study published
   A single-arm study of roughly 419,000 participants reports how often the notification fires and how often a mailed ECG patch confirms it.
7. 2021
   LOOP trial reports
   Implanted monitors roughly triple atrial fibrillation diagnoses in older adults at risk without a statistically significant reduction in stroke.
8. 2022
   Screening judged unproven
   The Fitbit Heart Study is published, and the US Preventive Services Task Force finds the evidence insufficient to recommend for or against screening asymptomatic adults for atrial fibrillation.
9. 2024
   Glucose warning and apnea clearances
   The FDA warns against devices claiming non-invasive blood glucose measurement; sleep apnea notification features are authorized for Samsung and Apple watches.
10. 2026
    First randomized consumer-wearable screening trial
    The EQUAL trial in the Netherlands reports that six months of smartwatch monitoring finds more new atrial fibrillation than usual care in adults over 65 at elevated stroke risk. The endpoint was diagnosis, not stroke.

The industry consolidated as the features became regulated. Google completed its acquisition of Fitbit in 2021; Apple, Samsung, and Garmin build sensing into general-purpose watches; Oura and Whoop couple hardware to subscriptions, which shifts the incentive from a single sale toward continuous engagement with the data.

§Heart rhythm, the best-supported use

The Apple Heart Study enrolled roughly 419,000 US participants and monitored them for a median of about four months. Just over half a percent received an irregular pulse notification. Those who did were mailed an ECG patch, and among the minority who returned usable recordings, atrial fibrillation was present in about a third during the patch period.¹ Two readings of that fraction are both fair. It understates the algorithm, because atrial fibrillation is often paroxysmal and the patch arrived days after the alert. It also flatters the algorithm, because only notified participants were investigated, so the study could not estimate how many cases were missed.

The Fitbit Heart Study used a similar design in a comparable number of people and reported a much higher positive predictive value, defined against atrial fibrillation present during the window the algorithm had flagged rather than across a later monitoring period.² The two headline figures are not comparable, and the difference lies in the definition rather than in the hardware.

Clearances for rhythm features are narrow by design. Labelling restricts them to adults without a prior atrial fibrillation diagnosis and states that they are not a substitute for clinical evaluation; what the device produces is a prompt to seek one.

§Sleep and breathing

Sleep tracking is heavily marketed and weakly validated. A comparison of seven consumer devices against polysomnography found that they estimated total sleep time reasonably well in healthy sleepers and classified sleep stages poorly, with performance varying by device and by stage.³ The reason is structural. Polysomnography scores stages from EEG, eye movement, and muscle tone; a wrist device infers them from movement and cardiac features, and no wrist-accessible signal maps cleanly onto the stage definitions.

A second failure mode matters more. Devices that infer sleep from stillness tend to score motionless wakefulness as sleep, so they overestimate sleep in precisely the people with insomnia who consult the data most closely. That also constrains how the epidemiology summarized under Sleep and longevity can be translated into a nightly score.

Orthosomnia

Sleep clinicians coined the term for patients whose insomnia is driven or worsened by the pursuit of a perfect tracker score, and who discount a normal sleep study in favour of the device.⁴ The report is a small case series rather than a prevalence estimate, and it remains the clearest documented instance of a consumer health metric aggravating the condition it claims to monitor.

Sleep apnea notification is among the most recent cleared functions on these devices. Samsung received US authorization in early 2024 and Apple later the same year; both detect breathing-related disturbance across multiple nights and are framed as a reason to seek testing. Confirmation still requires a sleep study.

§What these devices do not measure

No consumer wearable measures blood glucose. The FDA issued a safety communication in 2024 warning against smartwatches and smart rings sold with claims to measure blood glucose without piercing the skin, and stating that it had authorized no such device.⁵ A watch that displays glucose is receiving the value by radio from a continuous glucose monitor whose filament sits in interstitial fluid under the skin.

Cuffless blood pressure from optical sensors is likewise unvalidated, and cardiology bodies do not recommend the current devices for diagnosis or treatment decisions. Samsung's estimate must be recalibrated against an upper-arm cuff roughly monthly, and it reached users in the United States in 2026 as a general-wellness feature rather than a cleared medical one. A hypertension notification cleared for Apple's watch in 2025 sidesteps the accuracy problem by reporting a pattern consistent with chronically raised pressure over about a month while deliberately producing no number. Hydration status, blood alcohol, and circulating cortisol have all been claimed for optical sensors without published validation. Stress and recovery scores are repackaged heart-rate variability, which shifts with posture, breathing rate, caffeine, alcohol, and illness.

The glucose claim is the field's clearest falsehood

Non-invasive optical glucose measurement has been announced repeatedly for decades and demonstrated by nobody. Products making the claim are sold anyway, and a person with diabetes acting on a fabricated number can be harmed within hours. This is the sharpest example of a general pattern in consumer diagnostics, where the marketed capability runs ahead of the physics.

§The screening problem

The central critique of consumer health sensing is arithmetic rather than technical. A test with imperfect specificity applied to a population in which the condition is rare produces mostly false positives, however good it looks in a clinic. Atrial fibrillation is uncommon before fifty, and the cohorts wearing these devices skew young. Each false alert generates a clinic visit, often a monitor, sometimes an echocardiogram, and reliably some anxiety, plus incidental findings that start cascades of their own.

Whether finding more atrial fibrillation helps is a separate question, and the strongest evidence is discouraging. The LOOP trial randomized older adults with stroke risk factors to an implanted loop recorder or usual care. The monitor roughly tripled diagnoses and increased anticoagulation, and the reduction in stroke or systemic embolism was not statistically significant.⁶ The EQUAL trial in the Netherlands, the first randomized test of a consumer device, reported in 2026 that six months of smartwatch monitoring produced roughly four times as many new atrial fibrillation diagnoses as usual care in adults over 65 at elevated stroke risk. It counted diagnoses, not strokes.⁷ In 2022 the US Preventive Services Task Force concluded that evidence was insufficient to assess the balance of benefits and harms of screening asymptomatic adults for atrial fibrillation.⁸

Detection is not benefit

Device makers cite cases where a notification preceded a diagnosis, which is real and unsystematic. Cardiologists point out that the randomized tests of intensive rhythm monitoring have raised diagnosis and treatment rates without demonstrating fewer strokes. Both positions are consistent with the evidence, because the trials run so far, on implants and on watches alike, have counted diagnoses rather than events.

§Activity, longevity, and behaviour change

Step counts are the oldest wearable metric and the one with the most epidemiology behind it. Pooled cohort data associate higher daily step counts with lower all-cause mortality, with the association flattening well below the familiar ten-thousand target, and at lower counts in older adults.⁹ These are observational findings and carry the obvious confound in both directions: illness reduces walking as well as following from it.

Evidence that wearing a device changes behaviour durably is weaker than the market implies. In a randomized trial of a behavioural weight-loss programme in young adults, the group given a wearable activity monitor had lost less weight at two years than the group self-monitoring without one.¹⁰ Nothing about that result argues against physical activity, which remains the best-supported intervention for the functional outcomes the longevity field cares about. It argues that measurement is not motivation.

Wearable streams increasingly feed other consumer health products: Quantified self tracking, Biological age estimates sold alongside methylation tests, the personalized models discussed under Human digital twins, and the supplement regimens of the Biohacking and grinders community. Weight-management programmes built around GLP-1 receptor agonists have an obvious use for activity and lean-mass tracking, whose contribution to results has not been established. None of these composites has cleared the validation bar set out under Aging biomarkers, and a watch-estimated maximal oxygen uptake is a model output rather than the treadmill measurement whose association with mortality gave the metric its reputation.

§Outlook

The technology is mature and the evidence is not. Sensors are cheap and accurate enough at the physical layer, and body-worn measures have begun to serve as endpoints in regulated drug trials. What is missing is the study that would settle the consumer question: randomizing device use against usual care in a defined population and counting strokes and deaths rather than diagnoses.

Governance lags further. Health data from a consumer device sits largely outside medical privacy law in the United States, in the hands of advertising companies and insurers, and life insurers have offered premium discounts for sharing activity data. The concerns are those raised under Genetic discrimination and, for neural signals, under Mental privacy and Neurorights, transposed to a stream that is continuous, behavioural, and voluntarily surrendered. How voluntary it is depends on whether the device arrived through a purchase or through an employer or insurer programme, a distinction treated under Quantified self. The devices also cost money and reach the people least likely to need screening, which makes them a case for Access and inequality.

The contrast with implants clarifies what these devices are. A Brain–computer interface accepts surgical risk for signal quality that surface sensing cannot approach, and Non-invasive neuromodulation accepts weaker effects to avoid the surgery. Wearables sit at the far end of that trade, buying scale and reversibility at the cost of resolution. The open question is whether a low-resolution signal collected continuously from millions of people is worth more than a high-resolution signal collected from patients occasionally, and whether the diagnoses it produces amount to anything like the Compression of morbidity the marketing invokes.

§See also

• Continuous glucose monitoring
• Quantified self
• Consumer blood testing
• Human digital twins
• Sleep and longevity
• Aging biomarkers
• Exercise as a geroprotector
• Healthspan

§References