HRV: The Number Every Wearable Shows and Almost Everyone Misreads

The full story

Heart-rate variability has an unusual problem for a consumer metric: the science underneath it is considerably better than the science of the thing being sold. The link between low variability and dying is among the more replicated findings in cardiovascular epidemiology. The link between last night's overnight RMSSD and anything you can act on is much thinner rope.

Both statements hold because they describe different objects. Epidemiology compares populations under specified recording conditions. Your ring compares you to you, overnight, through skin, using an algorithm nobody outside the company has audited. The gap between those two exercises is where almost every misreading lives.

What the number is, and why the acronym matters

HRV is the variation in the interval between heartbeats, and on most interpretations a proxy for parasympathetic (vagal) influence on the heart. It is not one number. The joint task force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology set out the standard definitions in a 1996 guideline in Circulation: time-domain indices including SDNN, RMSSD and pNN50, and frequency-domain indices that split the signal into low- and high-frequency power. That guideline also laid down a rule consumer apps routinely ignore. Recording length and measurement conditions must be matched before two HRV values are compared. A five-minute seated reading and an eight-hour sleep average are not the same measurement in different clothes.

You will also see values reported log-transformed, as lnRMSSD or lnHF. That is not decoration; the cohort studies that found the mortality signal generally worked on the transformed scale, and a linear-looking dashboard trend is not what they analysed.

The mortality signal is real

A 2022 systematic review and meta-analysis in Neuroscience and Biobehavioral Reviews pooled 32 studies plus two individual-participant datasets, covering 38,008 people, and found lower HRV predicted higher all-cause and cardiac mortality across ages, sexes and recording lengths. Participants in the lowest quartile of five-minute RMSSD carried a pooled hazard ratio of 1.56 (95% CI 1.32 to 1.85) against the other quartiles combined.

This is old news in the best sense. A 1994 Framingham Heart Study analysis in Circulation followed 736 participants with a mean age of 72 for four years and found each one-standard-deviation decrement in log-transformed low-frequency power came with a 1.70-fold higher hazard of all-cause mortality (1.37 to 2.09) after adjustment for conventional risk factors. In people without known cardiovascular disease, a 2013 meta-analysis with dose-response meta-regression in Europace pooled eight studies and 21,988 participants and found the lowest versus highest level of SDNN carried a relative risk of 1.35 (1.10 to 1.67) for a first cardiovascular event, with low HRV overall associated with a 32 to 45 percent increased risk depending on the index used.

Those are respectable effect sizes for a single non-invasive measurement. They are also, note carefully, quartile comparisons within populations rather than thresholds for a person.

Which is exactly why your number is not comparable to anyone else's

The single most useful paper here for a wearable owner is a 1996 analysis in the Journal of the American College of Cardiology of 2,722 Framingham participants, asking what actually determines HRV. The answer is unglamorous: age and heart rate dominate, with partial R-squared values of 0.125 to 0.389. Lower HRV was also independently associated with beta-blocker use, prior myocardial infarction or heart failure, diuretic use, diastolic blood pressure of 90 mmHg or higher, diabetes, smoking, and three or more cups of coffee a day.

Sit with that list. An absolute HRV value carries information about your age, your resting heart rate and your medication list before it carries anything about your recovery. Two people with identical rings and identical numbers can occupy entirely different physiological situations, and a percentile rank against strangers is close to meaningless unless it has at least been adjusted for age and heart rate. Most dashboards do not tell you whether it has.

Alcohol: the one lever with unusually clean data

If you want a demonstration that the nightly number responds to something concrete, alcohol is the cleanest case in the literature. A 2019 review in Pharmacology Biochemistry and Behavior covering 33 studies found acute intake reduced resting HRV in healthy subjects and, more soberingly, that resting HRV in people with alcohol use disorder improved only after at least four months of abstinence. A 2014 review of 24 studies in Alcoholism: Clinical and Experimental Research found acute consumption decreases parasympathetic and increases sympathetic HRV indices. A 2019 meta-analysis in Drug and Alcohol Dependence pooling 15 studies found people with alcohol use disorder had significantly lower parasympathetic HRV than healthy controls (Hedges' g -0.43, 95% CI -0.76 to -0.10), with total variability and RMSSD separating the groups most reliably. Note the width of that interval: the upper bound nearly touches zero.

One complication worth reporting rather than burying. That 2014 review describes the relationship between habitual intake and HRV as J-shaped, with low intake associated with higher HRV and heavier intake associated with lower HRV than abstainers. That is an observational shape across reviewed studies, not a recommendation, and nothing in this evidence set tested drinking as an intervention on HRV.

What the evidence does not show

Nothing here shows that raising your HRV lowers your risk of anything. Every mortality finding above is associational: the cohorts measured variability and counted deaths. No trial in this set randomised anyone to an HRV-raising programme and then measured survival, and that absence is the whole distance between a prognostic marker and a treatment target. Treat "improve your HRV" as an untested hypothesis with a plausible mechanism attached.

Nothing here defines a good absolute value either, not for your age, not for your sport, not for your device. And the optical sensors have a narrow competence window. A 2023 validation study in Psychosomatic Medicine found that in 57 healthy adults performing paced breathing, phone-based photoplethysmography agreed near-perfectly with simultaneous ECG across lnRMSSD, lnpNN50, lnLF and lnHF, with every intraclass correlation above 0.90. Encouraging, under those conditions. A 2021 study in the International Journal of Psychophysiology compared the two signals in 96 youths and found excellent agreement at rest (Lin's concordance 0.93 to 0.99) that degraded during a mental-arithmetic task (0.88 to 0.91), with photoplethysmography biased toward higher HRV values than ECG. Optical HRV is trustworthy when the wearer is still. Nobody in this evidence set has shown it is trustworthy when the wearer is not.

How to read your own number without fooling yourself

• Find out which index your app reports and over what window. RMSSD, SDNN and pNN50 answer different questions and are not interchangeable.
• Compare yourself only to yourself, on the same device, under the same conditions, asleep and still. That is the condition under which optical measurement has actually been validated.
• Judge trends over weeks. A single night's deviation has too many candidate explanations to function as a signal.
• When a night drops, the only acute mover with review-level evidence behind it is alcohol. Everything else on the usual list is plausible and was not tested in the studies cited here.

There is a version of HRV genuinely worth your attention, and it is not the one on the dashboard. It is the version in the cohort studies: a cheap, non-invasive signal carrying real prognostic information about autonomic function, in large part because it sits downstream of your age, your heart rate, your medications and your disease burden. A wearable can plausibly show you the alcohol effect and the ordinary noise around your own baseline, which is more than most consumer metrics manage. What it cannot do is place you on a curve that was drawn in populations, under matched conditions, on a transformed scale. If your ring reports a number you like tonight, enjoy it. Just don't mistake a rank against strangers for a measurement of your own physiology.