I wear a watch that gives me an HRV every morning. I will not publish that number, and it is not out of modesty. It is because compared with yours, it means nothing, and that is precisely the point.
Heart rate variability, HRV, has become the fetish number of watches and rings. It has a solid basis: a healthy heart is not a metronome, and the small irregularities between two beats reflect the dialogue between the nervous system and the heart. The problem is not the measurement. It is the number one use made of it, comparing yourself with others, which is also the number one mistake.
healthy adults across 44 studies, with between-individual differences of several orders of magnitude
Nunan 2010
of its youthful value: where RMSSD falls before 60, before stabilising
Umetani 1998
error on RMSSD depending on the consumer device, against an electrocardiogram
Stone 2021
What the number measures
The heart does not beat at equal intervals. Between two beats, the gap varies by a few milliseconds, under the effect of breathing and the vagus nerve, the nervous system's brake. The 1996 standards set the indices: SDNN, the standard deviation of all intervals over the recording, and RMSSD, which only looks at the differences between successive beats. RMSSD is what consumer devices report, because it is the reference index of the vagal brake and can be computed on short recordings.
Shaffer and Ginsberg add the warning product sheets forget: reference values established over 24 hours, over five minutes and under five minutes are not interchangeable, and recording length, age and sex weigh on any baseline value. A watch that measures overnight and an app that measures five minutes on waking do not produce the same number, even in the same person.
Why your number does not compare with mine
Nunan's review gathered 44 studies and 21,438 healthy adults to establish normal values for short-term HRV. It did establish some, with decent homogeneity between studies for the simple indices. But it also reports between-individual variations reaching several orders of magnitude, up to 260,000% for some spectral indices, and method differences from one study to the next that are enough to explain discordant values in perfectly healthy people.
Age accounts for a lot. Umetani and colleagues followed 260 healthy subjects aged 10 to 99 over 24 hours: RMSSD falls to 47% of its second-decade value before the sixth, then stabilises; SDNN falls more slowly, to 60% by the tenth decade. Before 30, women's HRV is lower than men's, and the difference disappears after 50. In 1,906 healthy subjects measured for five minutes, almost every index changes with age, and sex differences vanish in the last two decades.
And part of it is written before any habit. A genetic analysis of 53,174 people identified seventeen variants in eight regions of the genome associated with HRV, several of them in genes expressed in the sinoatrial node, the heart's natural pacemaker. These variants explain only 0.9 to 2.6% of the variance, but they are a reminder that two people of the same age, same sleep and same training have no reason to show the same number.
Then the device
Stone and colleagues compared seven consumer devices with a multi-lead electrocardiogram, over 148 resting trials. All give a correct mean heart rate. On RMSSD, the mean error ranges from 4% for the best app to 7% for the Oura ring, and up to 112% for an app that reads the pulse through the phone's camera. A watch number is therefore only comparable with the number from the same watch.
What HRV measures well: you against you
The right unit is not the millisecond, it is the deviation from your own baseline. The Finnish study by Pietilä is the cleanest demonstration: 4,098 employees, each recorded on at least one day with and one day without alcohol, compared with themselves. During the first three hours of sleep, alcohol lowers HRV-measured recovery by 9.3 points at a low dose, 24 at a moderate dose, 39.2 at a high dose. The effect is visible from a low dose, stronger in the young, and neither regular exercise nor youth protects.
Nobody would have seen that effect by comparing the 4,098 numbers with each other. It only appears when each person is compared with themselves, and that is what a watch can do, provided it stays the same watch, worn at the same time.
Plews and colleagues drew the same lesson in Olympic athletes: HRV can rise or fall during a negative adaptation to training, which makes an isolated value unreadable. Their answer is methodological: average over several days rather than read an isolated value, and follow each athlete long enough to know their individual fingerprint. I take seven days, the length of an ordinary week. An HRV number taken one morning is a point. A seven-day average is a line. Only the line can be read.
So what about the link with health?
It exists, at population level. Hillebrand's meta-analysis, eight studies and 21,988 people without known cardiovascular disease, finds 32 to 45% more risk of a first cardiovascular event in the lowest HRV group compared with the highest, and about 1% less risk per percent of extra SDNN. Those recordings are standardised, in cohorts followed over time.
What that says: HRV carries information about the health of the autonomic nervous system, and it is real. What it does not say: that a watch number below a table puts you in the risk group. The table describes groups, measured differently from your watch, and your position in the distribution of healthy people depends first on your age, your genes and your device.
How I read mine
| Rule | Why | Source |
|---|---|---|
| Same device, same time, always | Norms for 5 minutes, 24 hours and overnight are not interchangeable; error ranges from 4 to 112% by device | Shaffer 2017, Stone 2021 |
| Seven-day rolling average | An isolated point rises or falls for opposite reasons; the line can be read, not the point | Plews 2013 |
| Deviation from my baseline, not from a table | Healthy values span several orders of magnitude, age and genetics weigh before health does | Nunan 2010, Umetani 1998, Nolte 2017 |
| Find the cause of a drop before worrying | Alcohol, effort, infection, short night: one night's drop almost always has a cause from the day before | Pietilä 2018 |
In practice, I look at the week's average, not the day's number. When the day's number drops, I look at the day before first: bedtime, meal, effort, the start of a cold. And I never look for my number in an age table, because I know what is in there: other watches, other protocols, other genes.
What this article does not say
- That HRV is useless. It is linked to cardiovascular risk at population level, and it detects, in the same person, the effect of an evening out, a workout or an infection. It is the comparison between people that is useless.
- That age tables are wrong. They describe the decline with age well. They do not say where you should be, because the spread within each age is too wide.
- That your watch is accurate. Only a comparison with an ECG says so, device by device. An imprecise but consistent device still tracks a trend.
- That a drop is a sign of disease. One night's drop almost always has a cause from the day before. A lasting drop from your own baseline, on the other hand, deserves a question to a doctor.
- That I have proven anything with my watch. I describe a way of reading, consistent with the literature, not a measurement.
« HRV is not compared like height. It is compared like a signature: with yesterday's, from the same hand. »
What Helix does
HRV is the measurement with the most to lose from being compared with a norm, and the most to gain from being compared with itself. That is the principle of the whole Helix score: a personal baseline, measured with the same device, and deviations from that line rather than positions in a table.
In practice, the engine reads HRV as a rolling average, links it to the day before, alcohol, effort, bedtime, and never shows a bare number next to an age average. When a baseline is missing, it says so, and waits.
A score that compares your HRV with your own, not with strangers measured differently.
Join the beta