Smartwatch and health: which metrics matter and which ones fail

A smartwatch gives you dozens of numbers a day, but they are not all worth the same. Some are measured well and backed by studies linking them to your health; others are estimates so imprecise that they can lead you to bad decisions, like eating more because the watch says you burned 700 kcal. Here is, metric by metric, what validation studies say and what to do with each number. We are not reviewing specific models: algorithms change with every update.

What a smartwatch measures well and what it measures badly

The broadest review to date (Doherty and colleagues, Sports Medicine, 2024) pooled 24 systematic reviews covering 249 validation studies. Its first conclusion is uncomfortable: only about 11% of the watches and bands released so far have been validated for at least one measurement. The second is useful: heart rate and steps come out reasonably well; energy expenditure, sleep detail and estimated aerobic capacity, considerably worse.

Steps: the most reliable number and the one that says most about your health

Counting steps is what a watch does best. A review of 158 studies (Fuller and colleagues, JMIR mHealth and uHealth, 2020) concluded that, in the lab, several of the most studied brands count steps accurately, and the 2024 review puts mean error between 9% below and 12% above: more than enough to compare weeks. It is also the metric with the strongest health evidence. A meta-analysis of 15 cohorts with 47,471 adults followed for a median of 7 years (Paluch and colleagues, The Lancet Public Health, 2022) split them into quarters by daily steps. Compared with the quarter that walked least (median 3,553 steps), the risk of dying from any cause was 40% lower at about 5,800 steps, 45% lower at about 7,800 and 53% lower at about 10,900. But the benefit levels off: around 6,000-8,000 steps a day in adults aged 60 and over and around 8,000-10,000 in those under 60. It is an observational study: it shows an association, not cause and effect. If you walk to lose weight, how many steps to lose weight has the numbers.

• Look at your weekly average, not a single day: one 3,000-step day does not undo a week of 8,000.
• If you are well below 6,000, the biggest jump in that data is the first one: from about 3,500 to about 5,800 steps a day.
• The watch also helps you move: a review of 39 systematic reviews (Ferguson and colleagues, The Lancet Digital Health, 2022) estimated about 1,800 more steps and 40 more minutes of walking a day among activity tracker users.

Resting heart rate: compare yourself with yourself, not with others

The optical sensor on your wrist measures resting heart rate well. In a Stanford study of 60 volunteers and 7 devices (Shcherbina and colleagues, Journal of Personalized Medicine, 2017), checked against an electrocardiogram, six of the seven had a median error below 5% while cycling; error grew when walking and as intensity went up. Another study (Bent and colleagues, npj Digital Medicine, 2020) found that during activity the absolute error was, on average, 30% higher than at rest. The valuable thing about this number is that it is yours. Among 92,457 adults with a median of 320 days of recordings (Quer and colleagues, PLoS One, 2020), mean resting heart rate was 65 beats per minute, but between people it ranged from 40 to 109. Within each person, by contrast, it was very stable, although 20% had at least one week in which it moved by 10 beats or more.

Calories burned: the least reliable number on your watch

This is the most expensive error. In the Stanford study, measured against indirect calorimetry (analysis of the air you breathe), no device estimated energy expenditure with less than 20% error: median error ranged from 27% for the best to 93% for the worst. The 2024 review confirms it: depending on the review, mean error runs from 21% below to 15% above, and for a specific person on a specific day it can be larger. Put numbers on it: if the watch says 600 active kcal and it is 27% off, the real figure could be about 160 kcal higher or lower. If you “eat back” those calories, the watch’s error goes straight onto your plate and can swallow a good part of your deficit.

Sleep: how long you sleep, yes; the stages, not so much

The watch infers sleep from movement and pulse, without seeing the brain activity that polysomnography records. One study (Chinoy and colleagues, Sleep, 2021) compared several consumer devices with polysomnography in 34 young adults over three nights. All of them detected very well when you were asleep (sensitivity of 0.93 or higher), but struggled to recognise when you were awake in bed (specificity from 0.18 to 0.54); sleep stages came out inconsistent and every device did worse on the night of disrupted sleep. That is why they tend to overcount time asleep: the 2024 review reports typical errors above 10%. To sleep better, how to sleep better covers what works.

• Trust the timings: when you go to bed and get up, and whether they are regular.
• Do not read much into the stage breakdown or the “sleep score”: if you wake up rested, little “deep sleep” in the app is not a problem.
• If you snore loudly, wake up gasping or feel very sleepy during the day, bring it up with your doctor: the watch will not solve that.

Blood oxygen and VO₂ max: estimates with a wide margin

Wrist blood oxygen (SpO₂) differs from the reference by up to 2 points on average, and on this scale 2 points matter. Even hospital pulse oximeters have not solved it: at the University of Michigan (Sjoding and colleagues, New England Journal of Medicine, 2020), with readings of 92-96%, true arterial saturation was below 88% in 11.7% of measurements in Black patients, compared with 3.6% in White patients. A watch is no tool for monitoring a respiratory illness. Estimated VO₂ max has a wide margin. A meta-analysis of 14 studies (Molina-García and colleagues, Sports Medicine, 2022) found that watches calculating it at rest overestimate it on average and, for an individual, are off by anywhere from −13 to +17 ml/kg/min; those using a workout have no average bias, but the individual margin is still about ±10 ml/kg/min. If your watch says 40, your true value could be anywhere between 30 and 50: use it to see whether your fitness goes up or down, not as an exact figure.

Atrial fibrillation: what the Apple Heart Study really showed

Atrial fibrillation is an irregular heart rhythm that can go unnoticed and raises the risk of stroke. The largest study of whether a watch can detect it (Perez and colleagues, New England Journal of Medicine, 2019), funded by Apple, followed 419,297 people who said they did not have it:

• Over a median of 117 days, only 0.52% (2,161 people) received an irregular pulse notification.
• Those notified were sent an electrocardiogram patch, which they put on 13 days later on average. Of the 450 with analysable data, 34% showed atrial fibrillation.
• When a new notification arrived while the patch was on, 84% matched atrial fibrillation on the electrocardiogram.
• 57% of notified participants who answered the 90-day survey contacted a healthcare provider.

How to use the data: trends, not single numbers

A watch’s value lies in what you see when you put weeks together, not in today’s number. A simple routine:

• Once a week, compare three averages with previous weeks: daily steps, resting heart rate and hours of sleep.
• Set goals for what the watch measures well and what you control: steps, walking after meals, going to bed at a set time.
• Treat VO₂ max and “recovery” or “stress” scores only as trends.
• If a number makes you anxious, hide that view: the watch is there to help you move and sleep better, not to keep watch over you.

When a watch alert is a reason to see a doctor

A smartwatch does not diagnose: its alerts are a signal to talk to a professional, who can confirm or rule things out with the right tests. Do not start, change or stop any medication because of what your watch says. Book an appointment if:

• Your watch flags an irregular rhythm, especially if it happens again.
• Your resting heart rate stays clearly outside your usual range for days, with no obvious explanation.
• Your blood oxygen reads low again and again.
• Watch or no watch, with chest pain, shortness of breath or fainting, call your local emergency number.

A smartwatch is a good pedometer, a good monitor of your resting heart rate and a reasonable record of how long you sleep; it is a poor calorie counter, and its alerts are warnings, not diagnoses. Use it to look at trends, move more and book an appointment when something falls outside your normal, not to decide how much to eat.