How a smartwatch tells that a heart has stopped
Loss of pulse detection on watches like the Pixel Watch 3 looks for a pulse signal that suddenly vanishes while the wearer goes still, then checks again with extra light channels before it calls for help.
Pulse, covering wearables and health sensors
, 3 min read
67%simulated arrests Google's watch caught in its study
What to know
- The watch reads the same optical pulse signal it uses for heart rate, plus its motion sensor.
- Several checks must agree, then the wearer gets two chances to cancel before the watch tries to call emergency services.
- Google's study caught about two in three simulated arrests and made one unintended call per 21.67 user-years of wear.
Cardiac arrest is the one emergency where the person it happens to can't ask for help. The heart stops pumping in an effective way, blood pressure collapses and the person is unconscious within moments. If nobody sees it happen, nobody calls. A watch that notices is an attempt to be the witness.
The idea has moved from lab to wrist. Google's Loss of Pulse Detection runs on the Pixel Watch 3, and Honor promoted a similar feature for its Watch 6 Pro in China, though the European version is waiting on EU medical certification. Google has published the most about how its version works, so it's the clearest example.
A pulse that disappears
Every heartbeat pushes a wave of blood into the wrist's small vessels. An optical heart rate sensor shines green light into the skin and measures how much bounces back; the reading rises and falls with each wave. In cardiac arrest that rhythmic part of the signal flattens.
A flat signal on its own proves little. A loose strap, a cold hand or a watch on a nightstand can look the same. So Google's design stacks checks, all of which must pass within a short window:
- The green-light signal drops sharply and the accelerometer shows the wearer has gone still.
- A machine-learning model scores how likely the change is a real shift from pulse to no pulse.
- The watch switches on extra LEDs and light detectors, at other wavelengths, brightness levels and angles, to look for any weak pulse the first reading missed.
Google says the chain is built to flag an event in under a minute.
Then it asks
Before calling anyone, the watch checks whether the wearer is awake. The first prompt is a strong vibration and a message, cleared by a tap or a deliberate wrist twist. If that's ignored, a louder alert with sound follows. Only someone who stays unresponsive for 35 seconds moves on to a countdown and an emergency call, which tells dispatchers a wearable detected a suspected loss of pulse and gives a location. It's the same wait-then-escalate logic as fall detection, with a different trigger.
The trade-off in two numbers
You can't stop healthy volunteers' hearts to test this, so Google used a stand-in. It first recorded wrist signals from patients whose hearts were deliberately stopped during scheduled defibrillator testing, then showed that squeezing off arm blood flow with a tourniquet produces a similar signal. In that tourniquet model, the watch detected 67.23% of simulated arrests, according to the paper's abstract. Put plainly, about one in three was missed.
The other side is false alarms. Across two studies of everyday wear, there was one unintended emergency call per 21.67 user-years. That's roughly 7,900 days of wear per stray call (21.67 × 365).
Google tuned for few false calls and accepted misses, and its own limits say so: the feature may not catch every event, isn't meant for people with known heart conditions who need cardiac monitoring, and needs a charged device with a connection. Neither figure comes from real, unplanned arrests, which is the evidence still to come. Our explainer on how wearable accuracy studies work covers why a simulation number and a real-world number can differ.
Why bother with two in three?
Because the alternative, for an arrest nobody sees, is often no help at all. Google's blog cites research that half to three quarters of out-of-hospital arrests are unwitnessed, and that survival falls by 7% to 10% for each minute without resuscitation. A watch that catches two in three of those cases is still a witness most of those people wouldn't otherwise have. That's also why a regulator wants to see the numbers before the feature ships, the process behind an FDA clearance.
This article is general information about how technology works. It is not medical advice. Talk to a clinician about your own health.
Sources
- Loss of Pulse Detection on the Google Pixel Watch 3, Google Research, March 20, 2025
- Automated loss of pulse detection on a commercial smartwatch, Google Research (Nature paper abstract), 2025
WearablesHeart health techDigital health and AI in care
Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.