Why Your Resting Heart Rate Differs by Device: Apple Measures Awake, Garmin and Oura Measure Asleep

Suppose a person wears an Apple Watch and a Garmin on the same wrist for a night. Given what each computes, Apple will report a resting heart rate a few beats higher than Garmin for the same heart. Oura, for the same night, would show two numbers of its own, an average and a lowest, and across Oura’s membership those two figures sit 8 bpm apart.

Every one of those figures can be correct. They are answers to different questions, and the devices do not say which question they are answering.

This post lays out what each device actually computes, how far those definitions diverge, how that compares with the error in the sensors themselves, and why it matters more for resting heart rate than for almost any other number a wearable shows. The short answer to the last one: across 1.2 million people, every 10 bpm of resting heart rate carries a 9% difference in mortality risk, and the gap between two devices’ definitions is about 4 bpm.

What each device actually computes

Five devices, at least four definitions. Where possible this is from the manufacturer’s own documentation; where the manufacturer does not publish it, from the validation literature that had to work it out.

Read down the “when” column. Apple excludes sleep by design. Garmin, Oura and WHOOP measure almost entirely during it. Fitbit blends both. That single difference is the largest source of disagreement between devices, and no device labels its number with it.

How far the definitions move the number

Night is about 4 bpm below day. Speed and colleagues followed 433 adults wearing a Fitbit Charge 4 for up to three months, accumulating 19,242 days and 18,520 nights of usable data. Mean resting heart rate was 54.5 bpm during daytime rest and 50.5 bpm at night, a difference of 3.9 bpm [7]. That is roughly the gap between an Apple definition and a Garmin one on the same person.

Lowest is about 8 bpm below average, on the same device. Oura publishes both of its figures for its own members. Across 2024, restricted to members aged 18 and over, the average nighttime RHR was 65 bpm and the average lowest RHR was 57 bpm [8]. Same ring, same nights, same people, eight beats apart, depending on which of Oura’s two numbers you read.

Sleep stage moves it again. Grosicki and Presby note that resting heart rate is 3.5% lower during slow wave sleep than during REM [2]. A device that weights toward slow wave sleep, as WHOOP does, will sit below one that averages the whole night, from the same heart.

And time of night matters. In the Speed study, 53% of each person’s lowest readings fell between 03:00 and 07:00 [7]. A device that takes the minimum of the night is largely measuring the small hours; one that averages the night is diluting them.

For scale, the effect of fitness in the same study was 6.7 bpm between the least and most active participants [7]. The definitional gap between devices is more than half the size of the difference between a sedentary person and an active one.

How far the sensors move it

Less. Dial and colleagues put five wearables against ECG over 536 nights for nocturnal resting heart rate [9]:

At a resting rate of 60 bpm, those errors are roughly 1 to 2 bpm. Garmin was excluded from the resting heart rate comparison not because its sensor failed but because, at the time, the timing of its calculation was undisclosed, so the authors could not construct an equivalent ECG reference [9]. That is the definition problem showing up inside a validation study.

For Apple, a living meta-analysis of 82 studies and 430,052 participants found a heart rate bias of -0.27 bpm (95% CI -0.72 to 0.17), with limits of agreement from -7.19 to +6.64 bpm [10]. The average error is negligible; the spread on any individual reading is not, and that spread is the honest caveat on every figure in this post.

Why this matters more for resting heart rate than for most metrics

Two reasons, and the second is the one that should worry a product team.

It is the single most prognostic number a wearable produces. Zhang and colleagues pooled 46 prospective cohort studies covering 1,246,203 people and 78,349 deaths. Each 10 bpm increase in resting heart rate was associated with a relative risk of 1.09 (95% CI 1.07 to 1.12) for all-cause mortality and 1.08 for cardiovascular mortality. A resting rate above 80 bpm carried a relative risk of 1.45 (1.34 to 1.57) against the lowest category. The results “did not differ after adjustment for traditional risk factors for cardiovascular disease” [11].

That is a real, graded, well-replicated signal, and it means the definitional gaps above are not academic. A 4 bpm difference between two device definitions corresponds to about a 3.5% difference in the mortality risk estimate, conjured by a change of watch.

And it is everywhere downstream. Resting heart rate was one of the three wearable features most correlated with insulin resistance in the study behind Google’s Health Guardian, which we covered in what a watch can and cannot detect about insulin resistance. It sits inside readiness and recovery scores across the industry, including ours. A score that ingests Apple’s awake-only figure for one user and Garmin’s slow-wave-weighted figure for another is treating a 4 bpm definitional artifact as physiology, and every derived value inherits it.

Which definition is right

The Speed study answers part of this directly. Heart rate stabilised in most subjects after four minutes of inactivity, with a further decrease of less than 1 bpm between four and twenty minutes [7]. So a daytime resting measurement needs about four minutes of stillness, and Apple’s five-minute background cadence and Fitbit’s five-minute no-movement rule are both consistent with that.

Beyond that, the honest answer is that there is no single right definition, only a right definition for a purpose.

For comparison to the clinical literature, Apple’s awake-only figure is closest to what the cohort studies measured, which was a seated or supine reading while awake. A sleep-based number is systematically lower and should not be read against a chart built on daytime values.

For tracking one person over time, sleep-based definitions are better, because sleep holds posture, stress and activity roughly constant. Oura’s own thresholds show how tight that can get: a deviation of more than 3 to 5 bpm above or below a personal average built over about two months “may be a sign of low recovery or excessive stress,” and a lowest RHR 0 to 10 bpm below the long-term average is read as good recovery [4].

Notice what that implies. Oura’s meaningful-change threshold is 3 to 5 bpm. The gap between device definitions is 4 to 8 bpm. A user who switches devices will trigger a recovery or stress signal from the definition change alone.

What this means if you show resting heart rate to users

Never pool resting heart rate across devices without normalising for definition. An Apple value and a Garmin value are not the same quantity. If your pipeline receives both, they need a source tag and either separate baselines or an explicit correction, and the correction is not a constant, because it depends on the person’s day-night difference.

Reset the baseline when the source changes. A switch from a daytime definition to a sleep definition drops the number by roughly the same amount as a meaningful physiological change. A baseline that carries across that switch will misfire.

Say which one it is. “Resting heart rate” on its own is a label covering at least four calculations. “Lowest during sleep” or “at rest while awake” costs six words and removes most of the confusion a user has when their partner’s device disagrees.

Judge change against the person’s own definition-consistent history. Within one device, the sensor noise is 1 to 2 bpm and Oura’s 3 to 5 bpm threshold is a defensible signal boundary. Across devices, nothing is.

Use the clinical gradient carefully. The 9% per 10 bpm figure is real and worth showing. It was measured with a daytime definition, on a population, and describes risk across a distribution. Presenting it against a sleep-based individual reading applies a population gradient to a number that is already 4 bpm lower than the studies’ construct.

Where we sit

Sahha reports resting heart rate as a biomarker and uses it in readiness and other scores, so this is a constraint on our own product, and the pipeline code is specific about how.

We do not compute resting heart rate from raw heart rate samples. We ingest the device’s own resting_heart_rate data type. That means an Apple user’s value is Apple’s awake-only estimate and a Garmin user’s is Garmin’s sleep-based one, and our biomarker carries whichever definition the source used. Every data log stores its source, recording method and device type, so that fact is never lost, but it is not corrected either.

Two consequences follow, one good and one to be clear about. The readiness factor built on resting heart rate computes its 30-day rolling average per source, so a user who switches from an Apple Watch to a Garmin starts a new baseline rather than contaminating the old one, which is the rule this post recommends. But a demographic or global comparison of resting heart rate across users is comparing an awake-only figure against a sleep-based one wherever the cohort mixes devices, and that gap is 4 bpm on the evidence above. For this metric, the baseline lens is the one the evidence supports and the population lenses need the device held constant or stated.

The general case, and why source is not metadata to be dropped at aggregation time, is in what is a good HRV and normalising wearable data across providers.

The short version

Apple defines resting heart rate as the lowest rate at rest while awake and excludes sleep. Garmin, Oura and WHOOP measure it during sleep. Fitbit averages all rest periods. Night runs about 3.9 bpm below day in a 433-person study, Oura’s own average and lowest figures are 8 bpm apart, and slow wave sleep sits 3.5% below REM.

The sensors, by contrast, agree with ECG to within 1 to 2 bpm on the better devices. The definition moves the number two to four times more than the measurement does.

That would be a curiosity for most metrics. For resting heart rate it is not, because across 1.2 million people each 10 bpm carries a 9% difference in mortality risk, and because the number feeds nearly every recovery and readiness score in the industry. A device switch produces a shift the size of a real recovery event, from nothing but a change in what “resting” means.