Diagnosis on your wrist? What the new smartwatch heart studies actually show

What this is about

The first week of November 2026 brought several similar-sounding headlines: your smartwatch detects high blood pressure without a cuff, an AI spots weak heart pumping in a watch ECG, wearables predict blood pressure crises three days ahead, and an AI reads the state of 18 body systems from how you move. The triggers were Google's Nature paper from November 5 and the American Heart Association (AHA) Scientific Sessions. The US cardiology conference ended on November 9 with a dedicated late-breaking session on AI in cardiac diagnostics, where brand-new study results are presented for the first time.

Most reports passed these studies along individually and without context. We put the numbers side by side. Our thesis: the watches can deliver a warning sign, nothing more. They miss many affected people. Depending on the device, they reach different results. And several of the studies haven't been peer-reviewed yet or come from companies that profit from the results.

The four studies compared

You need two terms to read the numbers. Sensitivity tells you how many of the people who are actually sick a test finds. Specificity tells you how many healthy people it correctly classifies as healthy.

Google: Better than the doctor's office, but not significantly

Google's model analyzes the pulse waveform and the ballistocardiogram over about seven days. The ballistocardiogram is the tiny recoil your body makes with every heartbeat, picked up by the motion sensor. The reference was 24-hour ambulatory monitoring, where a cuff measures automatically day and night. The watch used was a Pixel Watch 3. These details come from the preprint on medRxiv, whose numbers match those in Google's blog.

The confidence interval (CI) is the range in which the true value very likely lies. The ranges for the watch and the office reading overlap to a large degree. The preprint contains no statistical test for the difference, just the statement that the watch was "higher." According to the physician portal Epocrates, the gap was not statistically significant. 38.8 percent of participants had high blood pressure, so about 76 people. By our math, the watch found about 50 of them and missed 26. The preprint gives the same number: 26 false negatives, 17 of whom would also have slipped through the office reading. At 90 percent specificity, roughly one in ten healthy people also got a false alert.

The study's setup matters too. Anyone with known heart disease and anyone taking blood pressure medication was excluded. Only 196 of 296 enrollees made it into the analysis, mostly because their 24-hour monitoring was incomplete. Google itself calls the work "early feasibility research."

WATCH-LV: Which watch, which result?

The WATCH-LV study targets left ventricular systolic dysfunction (LVSD), meaning the heart's main chamber pumps too weakly. It was defined as an ejection fraction of 40 percent or less, i.e. how much blood the chamber pushes out per beat. At Incheon Sejong Hospital in South Korea, 2,211 adults recorded a 30-second ECG with an Apple Watch and a Galaxy Watch, one right after the other. According to the AHA release, the AI discriminated "about equally well" on both watches. But for 200 people (9 percent), the result was positive on one watch and negative on the other. The release doesn't give sensitivity, specificity or the area under the curve, a common overall measure of accuracy.

In the at-home part, 146 people measured with their own watch, 94 percent of them with a Galaxy Watch. We recalculated the numbers from the release:

Nearly half of the at-home group (49 percent) got a mix of positive and negative results. Of these, 40 percent actually had weak pumping. Study author Hak Seung Lee says it himself:

"A single smartwatch ECG is a snapshot, and repeated recordings may yield different results."

Lee is a cardiologist and also chief medical officer of Medical AI. That company developed the model and supplies the algorithm for Samsung's watches. Medical Xpress nevertheless calls the study "independent." The AHA points out that conference abstracts are not peer-reviewed. The setting also favors the results: patients came to a specialist clinic for a cardiac ultrasound, and many already had known heart failure. So the condition was far more common than in the general population. Whether treatment after an AI finding extends lives was not studied.

Blood pressure crises and movement: What's behind them

The crisis study followed 4,190 people with severe (stage 2) hypertension for twelve months at a hospital in Pakistan's Sindh province. The data didn't come from a watch alone: a wearable cuff device measured blood pressure, and a fitness band tracked pulse and heart rate variability. The AI analyzed 47 features. 312 people had a total of 418 crises. Only about 32 percent of patients who got an alert actually had a crisis. Without an alert, 97.4 percent stayed crisis-free. The widely quoted "61 percent fewer crises" isn't a measure of accuracy. It compares people whose doctor adjusted their medication within 24 hours of an alert with people who didn't get a quick response. The release doesn't describe how these groups came about. Without randomization, meaning random assignment, you can't infer an effect from this. In the same text, the AHA itself recommends upper-arm cuffs and advises against wrist and finger devices.

The Nature Medicine study is miscategorized in many reports. It doesn't rely on smartwatch data but on 3D skeletal movements captured by a depth camera during five movement tasks. Across 3,328 adults, the model found significant associations with 1,887 of 3,210 health measures. These are statistical associations, not validated diagnoses. Co-authors work at Newton Tech, the company that sells the camera system. It describes screening and diagnostic uses itself as "investigational," i.e. still being tested.

A fifth AHA study, where AI was deliberately used as an assistant, offers a useful contrast. Across 1,200 cardiac ultrasound exams at 21 Kaiser Permanente facilities, cardiologists using AI drafts needed 4.6 instead of 8.5 minutes per report. They substantially revised the AI draft in 50.6 percent of cases, and the human draft in 49.7 percent. A doctor always signs off at the end.

What's available in Germany

Apple's feature runs on the Series 9 or newer and the Ultra 2 or newer. It's intended for people aged 22 and older without a hypertension diagnosis who aren't pregnant, and analyzes 30-day windows. According to the FDA documents, sensitivity for stage 2 was 53.7 percent. Google's feature runs for free on the Pixel Watch 3 to 5. On the Fitbit Air, you need Google Health Premium for 9 euros a month. Google says the Nature study "helped shape" the feature. But it's built on the WavesFM foundation model, and Google doesn't say the Nature model itself is in the app. Google explicitly states the feature is "not a pre-screening for hypertension." Samsung's weak-pumping detection was demoed in Berlin in 2025. We found no EU approval or launch date for Germany.

False alarm or all-clear: How to read your watch

The riskiest outcome is a false all-clear. Apple's FDA documents say it outright: if no notification appears, that doesn't mean you don't have high blood pressure. A research letter in JAMA ran the numbers on US population data (summary at the AAFP). For people over 60, the probability of high blood pressure without a notification only drops from 45 to 34 percent. With a notification, it rises to 81 percent. Among under-30s who get a notification, by contrast, only 47 percent actually have high blood pressure.

In practice, that means: - After an alert: Apple recommends measuring with an upper-arm cuff for seven days and then taking the readings to your doctor. That makes sense, because many alerts are false alarms, especially in younger people. - Without an alert: If you're over 40 or have risk factors, you should still measure regularly. The German Heart Foundation (Deutsche Herzstiftung) stresses that smartwatches can complement a doctor's visit, but not replace it. - With acute symptoms like chest pain or shortness of breath: don't look at your watch, call emergency services.

Our take

What's proven: in studies, the watches find more cases than chance would predict, and they measure over days without you doing anything. A single office reading can't do that. In April 2026, the German Cardiac Society (Deutsche Gesellschaft für Kardiologie) described them as an "early warning system" whose strength lies more in raising awareness than in precise measurement.

What's not proven is that people live longer or healthier lives as a result. None of the four studies measured that. Two results have only appeared as conference abstracts. Two involve companies that profit from the outcome. And one study that many outlets lumped in with smartwatches doesn't measure at the wrist at all.

Our assessment: the most important sentence of the week isn't in a press release but in the WATCH-LV study itself. Which watch you use changes the result for one in eleven patients. As long as that's the case, a smartwatch is a reason to check your blood pressure or see your doctor. It isn't a diagnosis, and a silent watch isn't an all-clear.