VO2max From a Smartwatch: How Accurate Is the Measurement and Can You Trust It?

For most people who train, that is enough to use the watch value as a rough indicator of direction: is fitness rising or falling after a few months of regular training? The problem starts when the number on the screen is treated like a test result: compared with norm tables, with a friend's result on another brand of watch, or used to assess health risk. Then an error of several, or even more than ten, ml/kg/min can change the category from 'average' to 'excellent' or the reverse.

In this article we explain step by step how VO2max estimation algorithms work, what independent studies showed for Apple Watch, Garmin and Pixel Watch, where errors come from and how to get the most reliability out of a watch. We also show when it is worth investing in a laboratory test. For broader context, what VO2max is and how to improve it, see our entries on VO2 max, zone 2 training and heart rate variability (HRV).

VO2max (maximal oxygen uptake) is the greatest amount of oxygen the body can take up and use per minute during maximal exercise, most often expressed in millilitres per kilogram of body mass per minute (ml/kg/min). It covers the whole cascade: lung ventilation, oxygen diffusion, cardiac output, oxygen transport in the blood and its use in the muscles. The gold standard for measurement is an exercise test to exhaustion with direct analysis of exhaled gases (spiroergometry, CPET), that is, with a mask or mouthpiece and a metabolic analyser.

A watch has no mask and no analyser. Instead it relies on a relationship that exercise physiology has known for a long time: at a given running pace (or power), heart rate rises higher the lower the fitness. The algorithm therefore compares the heart rate trace with speed, usually from GPS and an accelerometer, during a run or brisk walk, takes into account age, sex, body mass and resting and maximum heart rate, and then 'converts' this into VO2max. Importantly, the effort is submaximal, so nobody has to exhaust themselves. As a systematic review points out, most studies concern Garmin devices with the Firstbeat algorithm and estimates from submaximal outdoor runs, compared against a laboratory treadmill test.

Different manufacturers use their own variants of this idea (some require a treadmill or an outdoor run with GPS, others use multi-day data from everyday wear), but the logic is similar: from indirect data, namely heart rate and movement, a quantity that cannot be seen directly is estimated. An important consequence follows: any error in the input data, in heart rate, in pace or in the assumption about maximum heart rate, is carried into the result.

Such a large error is not a feature of one manufacturer alone. In a study of 19 people with a Garmin fenix 6, the mean error against laboratory data averaged over 30-second windows was 7.05% (Lin's concordance coefficient 0.73), which met the authors' validation criteria (MAPE below 10% and CCC above 0.7). In the same paper that watch failed at measuring blood oxygen saturation, which shows well that 'works in one function' does not mean 'works in all'. In turn, in a study of 24 healthy people comparing five different bands and watches, all the devices evaluated had an error above 10% except the Garmin Forerunner 920XT (7.3%), which significantly underestimated VO2max.

A study of the Pixel Watch 3 (34 participants) is also striking: the mean systematic bias was small (about -0.5 ml/kg/min, MAPE 8.65%), but the limits of agreement ran from about -12.6 to +11.5 ml/kg/min. In other words, the device as a group does not err in one direction, but in a specific person the error could be very large. The authors judged that the estimates lack sufficient individual-level validity for clinical use, although the bias suggests usefulness for population health monitoring.

The first source of error is the assumption about maximum heart rate. The algorithm often assumes it from age (using a general formula) or from earlier workouts. If your real maximum heart rate differs substantially from the estimate, and in many people the deviation is a dozen or so beats per minute, the result shifts systematically. The second source is the quality of heart rate measurement: optical wrist sensors are less accurate with rapidly changing effort, in the cold, with tattoos or a loosely worn strap, which is why studies often use a chest strap.

The third is that the estimate is submaximal and rests on a certain 'typical' relationship between heart rate and speed. If someone has unusual running economy, runs uphill or into the wind, is dehydrated, sleep-deprived, has had coffee or is in the heat, heart rate at a given pace will rise and the watch will 'decide' fitness is lower. The fourth, revealed in studies, is so-called regression to the mean: in people of low fitness estimates tend to be too high, and in very fit people too low, because the algorithm 'pulls' the result toward the typical population. This is why an endurance athlete with a real VO2max of 65 may see a clearly lower figure on the watch.

Finally there is method mismatch. In the Apple Watch study the reference was measured on a cycle ergometer while the watch relied on running, and in every validation, differences in protocol, data averaging and population add to the result. For the user this means one thing: comparing the number from a watch with a norm table or with a friend's result on another device makes little sense. Comparing it with your own result from three months ago, from the same watch and in similar conditions, is far more justified.

A study of the Garmin Forerunner 265 during a 12-week half-marathon preparation programme (20 people) showed a phenomenon worth knowing about: the watch overestimated VO2max by about 2.5-2.9 ml/kg/min on average both before and after training, but the error was relatively constant and the mean percentage error stayed below 10%. A constant systematic error is much less harmful for tracking a trend than a random error, because it does not mask change.

Aerobic fitness is among the strongest known predictors of survival. In a large retrospective study of 122,007 patients undergoing treadmill exercise testing, all-cause mortality fell as fitness rose, and people in the highest ('elite') group had, after adjustment, about 5 times lower risk of death than those in the lowest group (HR 0.20), with no upper limit of benefit observed. The American Heart Association, in a scientific statement, proposes treating fitness as a 'vital sign' in clinical practice, because it may be a stronger predictor of mortality than smoking, hypertension or diabetes.

Since differences between fitness categories carry such prognostic weight, the question arises whether an error of 4-9 ml/kg/min destroys the usefulness of a watch. The answer has two levels. At the level of one person, the error is too large to assign yourself to a normative category from a single reading. Over time (the same watch, several months, a rising or falling trend line), the information about direction of change remains useful, because it reflects a change in fitness and not an absolute value. Just remember that a watch trend is a signal, not proof.

If the result is to be used for health decisions, for example assessing cardiovascular risk, planning an intensive training programme after an illness or adjusting load in a person over 60, it is better to rely on a test performed under supervision. For healthy, active people who simply want to know whether their form is going up, a watch is usually entirely sufficient.

The most accurate is a spiroergometric test (CPET), performed in a sports medicine, cardiology or exercise physiology laboratory. The participant performs progressively increasing effort on a treadmill or cycle ergometer while an analyser measures oxygen consumption and carbon dioxide production breath by breath; thresholds of ventilation are also obtained, which matter a great deal when planning training zones. The test is a maximal effort, so it should be performed under staff supervision, and in people with heart disease or risk factors it should be preceded by medical assessment.

There are also field tests, cheaper and more accessible but less accurate: the Cooper test (distance covered in a 12-minute run), the 1.6 km walking test and equations that estimate VO2max without exercise from age, body mass and resting heart rate. Their error can be comparable to a watch's error, and their advantage is that they are fully repeatable under identical conditions (the same distance, route and pace of effort). For people who want to track progress without a laboratory, a regular Cooper test or a repeatable control run over the same distance often gives information as useful as a watch estimate.

A good analogy is a bathroom scale that is off by 1-2 kg but consistently. In the morning, on the same scale, in similar conditions, a drop of 3 kg is a signal. The same scale for two different people is no longer a good tool for absolute comparison. VO2max from a watch should be read the same way: do not assume that 52 for you means more than 49 for a friend with a different device, but do notice if your own 49 from half a year ago has changed to 53.

A watch is a wonderful tool for collecting data about your own activity, and the VO2max it shows is a useful clue as long as you do not expect laboratory accuracy from it. If you treat it as a compass rather than a map, it is very helpful: it shows the direction your form is heading. If you treat it as a test result, you expose yourself to conclusions the data do not justify. Before an important health or training decision, and at least once every few years if the stake is cardiovascular risk assessment, consider a real exercise test with gas analysis.