A smartwatch's HRV and activity data flagged who later developed high blood pressure

Physiological data a smartwatch already collects — heart-rate variability, resting heart rate, and time spent in vigorous activity — carried a signal about who would go on to develop high blood pressure, in a year-long study of adults who started out normotensive. The finding is genuinely interesting and genuinely preliminary: it comes from a single observational cohort of 230 people, has not been externally replicated, and does not turn a watch into a blood-pressure test [s1].

It sits alongside the harder problem of measuring blood pressure on the wrist at all, where independent specialists have been reluctant to endorse the cuffless consumer devices. This study is about something different — not reading pressure, but predicting who will become hypertensive from other signals.

What the study did

Researchers followed 230 normotensive adults aged 30 to 60 for 12 months, using consumer smartwatches [s1]. They computed each person's baseline from the first 30 days of valid data, capturing heart-rate variability, resting heart rate, and time in moderate-to-vigorous physical activity, then checked who met the definition of hypertension at follow-up using standardised office blood-pressure measurements [s1].

Over the year, 28 participants — 12.2% — developed hypertension [s1]. Those who did had lower baseline heart-rate variability and spent less time in moderate-to-vigorous activity than those who stayed normotensive [s1]. In a multivariable analysis, lower heart-rate variability, lower physical activity and higher body-mass index were each independently associated with developing hypertension [s1]. The authors also found an interaction: people with both reduced autonomic regulation and low physical activity had the highest predicted risk [s1].

Why the caveats are large

Several things keep this in the "promising signal" category rather than a usable tool. It is one cohort of 230 adults with 28 events — a modest number of outcomes on which to build a prediction — and the authors themselves call for validation in larger, externally replicated cohorts [s1]. Machine-learning models improved statistical discrimination over clinical variables alone, but those were framed as exploratory complementary analyses, not a validated algorithm [s1].

The 12.2% who became hypertensive over a single year is also a high rate for adults aged 30 to 60, which suggests this was not a low-risk sample [s1]. A predictor's apparent performance depends on how many events it has to find, so a model that discriminates well in a cohort where one in eight converts may do less well in a general population where far fewer do — the same base-rate caution that governs any screening tool. The wearable signals here are risk markers layered on top of, not independent of, the conventional ones the study also found significant, such as body-mass index [s1].

There is also the direction-of-causation problem inherent to an observational design. Lower heart-rate variability and less activity are plausibly early markers of the same physiological drift that produces hypertension — but they may also be markers of other things, and an association in a cohort does not establish that acting on the wearable number changes anyone's blood-pressure trajectory. Notably, a meta-analysis of 21 randomised trials found activity trackers did not lower blood pressure, a reminder that a device predicting risk is not the same as a device reducing it.

A related limitation is measurement itself. The study's predictors — heart-rate variability and resting heart rate — are exactly the signals whose accuracy varies by device, as a validation across 536 nights found for the nocturnal HRV that feeds readiness scores. A prediction model built on a signal that one brand measures well and another measures poorly will not transfer cleanly between devices, which is part of what external validation would need to test [s1].

The wider context

A review guiding non-experts on consumer wearables situates work like this within a broader effort to build risk-stratification tools from wearable data, while stressing the same requirements for validity and careful interpretation before such tools reach clinical use [s2]. The signals in this study — low HRV, low activity, higher BMI — are also, notably, established cardiovascular risk factors on their own, so part of what the wearable is capturing is information clinicians already act on, now measured passively and continuously [s1][s2].

What it means

For now, treat this as a research result, not a feature to rely on. The useful, defensible reading is that sustained low activity and low heart-rate variability are consistent with higher cardiovascular risk, which is a reason to pay attention to the modifiable factors — activity in particular — rather than to a predictive score [s1]. Anyone concerned about blood pressure needs it measured properly, with a validated cuff, and interpreted against what actually counts as high. The watch may one day help flag who to check sooner; on this evidence, it has not yet earned that role [s1][s2].