Pixel Watch 3’s Loss Of Pulse Detection: The Algorithms That Tell Someone Is Dying

More and more of the ‘smart’ gadgets like watches and phones that we carry around with us these days come with features that we’d not care to ever need. Since these are devices that we strap onto our wrists and generally carry in close proximity to our bodies, they can use their sensors to make an estimation of whether said body is possibly in the process of expiring. This can be due to a severe kinetic event like a car crash, or something more subtle like the cessation of the beating of one’s heart.

There is a fairly new Loss of Pulse Detection (LoPD) feature in Google’s Pixel Watch 3 that recently got US FDA approval, allowing it to be made available in the US after previously becoming available in over a dozen European countries following its announcement in August of 2024. This opt-in feature regularly polls whether it can detect the user’s pulse. If not found, it cascades down a few steps before calling emergency services.

The pertinent question here is always whether it is truly detecting a crisis event, as nobody wants to regularly apologize for a false alert to the overworked person staffing the 911 or equivalent emergency line. So how do you reliably determine that your smart watch or phone should dial emergencies forthwith?

Budget Medical Devices

One of the amazing things about technological progress is that sensors and processing capabilities that were rather exotic a few decades ago are now being included in just about any smart device you can strap on your wrist. This includes motion sensors, pulse- and oxygen level meters, making these devices in theory capable of acting like ambulatory cardiac monitors and similar medical devices that monitor health parameters and respond to emergencies.

While for a long time the gold standard for heart function monitoring over a longer period outside a hospital setting involved a portable electrocardiogram (ECG) recorder, recently wrist-worn monitoring devices based around photoplethysmography (PPG) have proven themselves to be acceptable substitutes. In a 2018 study by Francesco Sartor et al. in BMC Sports Science, Medicine and Rehabilitation the researchers found that the wrist-worn PPG was not as accurate as the ECG-based chest strap monitor, but came close enough to be practical.

Here the difference is such that applications where precision actually matters the chest strap ECG is still the optimal choice, but wrist-worn PPG devices as integrated into many fitness bands and smart watches are an acceptable substitute, such as when monitoring heart rate for signs of atrial fibrillation. A 2022 study by Christopher Ford et al. in JACC: Clinical Electrophysiology examined two smart watches (Apple Watch 4, KardiaBand) for this purpose, finding that their accuracy here was 91% and 87% respectively.

Together with additional sensors like the commonly integrated motion sensor, these devices seem accurate enough to at least determine whether the person wearing them is suffering a cardiac event that requires immediate intervention.

Health Check

The idea of an automatic emergency call isn’t new, with for example the EU making such a system (called eCall) mandatory in new cars since 2018. The idea is that when a serious collision is detected, emergency services are contacted and provided with location and sensor data. Google added its Car Crash Detection feature to the Pixel 3 smartphone in 2019, and Apple added Crash Detection to its Apple Watch and iPhones in 2022. These use sensor data from gyroscopes, GPS, microphones, and accelerometers to determine whether a crash just occurred.

What users of these devices discovered, however, was that activities such as going on a rollercoaster ride could activate this feature, as well as snowmobile rides, skiing, and similar activities. In response, Apple had to adjust its algorithms on these devices to reduce the number of false positives. Despite this, rescue workers in e.g. Canada are still reporting a large number of false positives. One reason cited is that although there’s a time-out before emergency line is called with audible alarm, this can be hard to hear when you’re on a snowmobile.

As it turns out, defining what seems like a pretty clear event to us when you’re limited to just this handful of sensors is much trickier than it seems. After all, what is different between the sensor data from a rollercoaster ride, a car crash, dropping one’s phone or smart watch onto a concrete floor or forgetting said phone on the roof of the car?

In this context, the idea of taking a simple activity like measuring heart rate and pulse, and extrapolating from these that if they cease, an emergency has occurred is fraught with pitfalls as well.

Merging Data

How do you know as a human being that someone has just suffered cardiac arrest? You confirm that they don’t have a noticeable carotid pulse, and the reason why you checked is because they clearly collapsed. This is when you’d pull out your phone and dial emergency services. The LoPD feature that Google has introduced has to do effectively exactly these steps, except that it starts from the loss of pulse (LoP) rather than from seeing someone pass out and collapse to the ground.

Thus the tricky part is establishing whether said collapsing has occurred, not whether the pulse has been lost. After all, the user may have simply taken the watch off. According to Google, to verify their algorithms they hired stunt actors to simulate LoP using a tourniquet, cutting off blood flow, and simulating falls like a person suffering cardiac arrest would suffer.

On the sensor side they use the heart rate monitor (PPG sensor), which initially uses the green light to check for pulse, but can switch to infrared and red lights when a LoP condition is triggered. Simultaneously the motion sensor data is consulted, with a lack of motion taken as a sign that we’re dealing with a LoPD. This starts an auditory alarm and visual countdown on the screen before emergency services are contacted with an automated message plus the user’s location.

Then they needed to calibrate the response to this merged sensor data with clinical data on cardiological events before trialing the result with said stunt actors and volunteers. An article on this research was also published in Nature (paywalled, but here is a gift article), detailing the algorithm and the way they tested its effectiveness. In the paper the authors note one false positive event and subsequent emergency call across 21.67 user-years across two studies, with a sensitivity of 67.23%.

A Matter Of Time

Chain of survival in case of cardiac arrest. (Credit: European Resuscitation Council)
Chain of survival in case of cardiac arrest. (Credit: European Resuscitation Council)

In the case of cardiac arrest, time is of the absolute essence. This is also clearly noted in the Google paper on the LoPD feature, who note that ideally there is a witness on-site who can immediately begin CPR or ideally get a nearby automated external defibrillator (AED). Unfortunately in most cases of cardiac arrest, this event goes initially unnoticed. The LoPD feature on a smart watch thus would be for cases where nobody is around to notice the emergency and respond to it. Although it isn’t explicitly mentioned, it seems that the watch can also detect whether it’s being worn or not, which should prevent false positives there.

With over half a million US citizens each year suffering cardiac arrest, and over half of these occurring outside of a hospital setting, this could potentially save thousands of lives each year. Following cardiac arrest and in the absence of resuscitation the lack of blood and oxygen being circulated means that within minutes organs begin to suffer the harmful effects, depending on their oxygen requirements. The brain is generally the first to suffer ill effects, which is why the application of CPR is so crucial.

Because of the intense urgency following a major cardiac event like this, the practical use of this LoPD feature will be highly dependent on the location where the emergency occurs. In the case of someone collapsing while alone at home in their city house or apartment, this could conceivably save their life if emergency services can arrive within minutes. Even faster and more useful in less urban settings would probably be having your smart device notify nearby people who can then perform CPR while calling 911 or equivalent.

That said, perhaps the real killer feature that’s missing here is an integrated AED in smartphones since everyone has one of those things on them at all times, or even smart watches that can automatically perform defibrillation while also notifying emergency services.

38 thoughts on “Pixel Watch 3’s Loss Of Pulse Detection: The Algorithms That Tell Someone Is Dying

    1. I know of someone who had one while sitting on the toilet in bathroom at night.
      Must be embarassing to be found that way by the ambulance helpers.
      Even worse if you had passed away just shortly before.

      1. KISS means “keep it simple and stupid”.

        These smart watches can be bad to wearers of pacemakers/defibrillator under some circumstances.
        Because the electric sensor pulses may trigger the pacemakers unnecessarily.
        The optical sensor method is not problematic, though.

      1. That’s a PDA, silly! 😆 Aka palm pc, pocket pc, handheld pc.

        3com Palm Pilot had been in common use since 1996!
        https://en.wikipedia.org/wiki/PalmPilot

        Nowadays “smartphone” are just PDAs with a cellphone modem tacked on!
        Not creative, at all. We had them years before the iPhone.

        There were Nokia Communicator, T-Mobile MDA, Handspring Treo 160,
        the Siemens SX-45 (based on Casio Cassiopeia E-125) and many more.

        In fact, the iPhone might have been inspired by a Palm OS device.
        The Handspring Visor had an optional 2G cellphone module for Springboard slot.

        The people of Handspring even demoed it to Mr. Jobs, according to a documentary,
        who then was convinced to center everything around a phone instead of a Mac.

        1. This. The Treo was a great phone/PDA and before that a usable albeit less integrated phone module for the Handspring Visor. Apple followed by merged a phone into their iPod Touch unit which was a large screen iPod music and app device.

  1. Can’t speak for others but my heart stopped beating once and it was anything but subtle. Apple watch ECG caught it and that earned me a 2 week cardiac event monitor.

    Regarding the AED idea – they have pacemakers with build in defibs for patients who need that. You don’t really want a phone OS doing those things.

    1. My father had a pacemaker with defibrillator.

      The defibrillator kicked in once – for real, really required.

      It saved his life.

      The next visit with his heart doctor after he got out of the hospital, he had the defibrillator function turned off.

      The experience of being defibrillated while conscious was so ghastly that he decided he’d rather die than go through that again.

      Die he did, the next time the defibrillator would have kicked in.

      The one defibrillator experience extended his life by several months. Whatever he experienced that one time was apparently bad enough that not even extra months of life was worth it.

      1. Sorry about your dad Joseph. Can’t imagine it (waking defib) myself either. Was friends with a guy who coded in ER, got shocked, started to come around and heard “shock him again”. Said it felt like all his cells were flying apart. Closest I’ve been is respiratory failure during a procedure and waking up with incredible chest pain – not even comparable.

  2. The future will be a direct pipeline from preschool into hospice care with nothing in-between, hooked up to tubes and gadgets monitoring you like a lab rat the entire way through. It will be very ‘safe’ but it will have an enormous suicide problem and nobody will ever want to have kids

  3. A deep dive into how the AED devices work would be a good article. Acting as a defibrillator is the least of what they do, first they have to talk someone through attaching the pads correctly, then they have to analyse a heartbeat, and only if it detects fibrillation does it fire.
    Fun fact, it can take at least a minute for the heart to start up after a defib, so you really need someone there performing CPR as soon as possible.

    1. And the problem is? Right, poor support for people working in healthcare.
      Working as, say, an ambulance helper or carer in the nursing home should be rewarded. Both socially and financial.
      There’s something deeply wrong in modern society that such social-oriented people aren’t being welcomed with open arms.

      We need them more than ever, and we needed them to be uplifting
      and positive. Which won’t happen if they’re being mistreatet lik sub-humans.
      But if they are negative and unhappy, they are likely to loose engagement and merely work for the little bit of money.

      To make life excellent to all of us, people should be encouraged to work in social activities.
      There should be an overcapacity of people working at call centers,
      be it in case a big catastrophe happens or for sake of them being well rested.

      Same goes for education sector. Children should be raised in a way by their parents and family members early on that they show at least a minimum of respect to teachers.
      They don’t have to “obey” but at least behave normally.

      1. “Working as, say, an ambulance helper or carer in the nursing home should be rewarded. Both socially and financial.”

        I see you have jokes.

        EMS will hire the cheapest person with the least amount of experience, as those who burn out leave, since they are easy to replace.

        Those that make a lifetime of it either a) have had extremely easy on shift or b) are mentally ready to just crumble.

  4. “as nobody wants to regularly apologize for a false alert to the overworked person staffing the 911 or equivalent emergency line” … I have called 112 (I’m Dutch – 112 is the European emergency number) dozens of times and they have all been extremely cool.
    I’ve called them to test new VOIP installations, about large items on the driving lanes of the highway (heavy ratchet straps, wooden beams), a car that drove 30 km/h on the motorway, a generator that heavily smoked which I saw from the highway in a pinewood forest in dry conditions (“Yes, we’ve had hundreds of calls about that. This one has been checked and is ok, but please keep calling for similar events”), a traffic jam in an unexpected place with obscured view, and once about someone who had made a suicide attempt (fortunately a failed one). Each and every time they were really cool, and for the more minute things I asked them whether that was ok to call them for it “Yes, that’s perfectly fine!”.

    I notice there is some fear about calling 112 for things where noone is actually dying right now, I think instilled into children for not calling that number as a joke. When in doubt, I think it’s always better to call emergency services.

    1. We had 1922 in Germany for calling an emergency doctor, but it nolonger worked last time I tried in an emergency.
      Instead a new number should have been called instead, to manage the calls. Modernization or something. Or lack of personell.
      However, the computer voice held me in a telephone queue.
      For 20min – and it was a real emergency.
      After that, I hung up and called 112 (firebrigade in Germany, 110 is/used to be police) and got a human being on the phone, who then called the ambulance at hospital real quick.
      If we hadn’t been lucky that someone was there doing the cardiac massage and mouth-to-mouth resuscitation already, that unconscious person wouldn’t have made it because I had to wait on phone.
      Long story short: Time matters. If you can’t get through, consider try calling police or firebrigade or any other emergency service.
      They can make contact with their colleagues.

  5. Last paragraph:
    “That said, perhaps the real killer feature that’s missing here is an integrated AED in smartphones since everyone has one of those things on them at all times, or even smart watches that can automatically perform defibrillation while also notifying emergency services.”

    1) “killer feature”- Interesting choice of words given the topic!

    2) AEDs, along with non-automated external defibrillators, work by passing hundreds of volts ACROSS the heart to restore rhythm. For external defibrillation to work, the paddles/pads are placed on either side of the heart specifically so current flows through/across the heart. Defibrillation is not as simple as just shocking the body. So… even if watches or phones had enough battery capacity as well as the circuitry to boost to hundreds of volts, no single device strapped to your wrist or in your hand could ever really perform defibrillation as we know it. Implanting electrodes in/around the heart can resolve the challenge of getting the current to flow along the required physical path, but at that point a dedicated pacemaker variant makes way more sense than relying on your smartphone/watch to be the defibrillator controller….

  6. sometimes i wonder if life saving features are there so that you live long enough to buy the next version.

    i could do with fewer death stealers. after observing the way the world treats old people i dont want to live past 50.

    1. 50? Don’t be silly. You’re still a boy at that age.

      Don’t eat crap, or smoke, and restrict alcohol to special occasions rather than a regular part of your diet, and with a modicum of reasonable exercise it won’t be until you hit your mid 70s that you’ll start to notice age.

    1. Life-saving features are great, but I’m quite curious if it works in practice. Even with GPS, the watch can only give the approximate location. In a densely populated area, there is very little chance of finding the person. Hopefully it’ll work and save many lives.

      As for the black mirror part, it’s an interesting technology to monitor your heart rate 24/7. But I’m not looking forward to the heart rate being coupled to other data. I can only guess what Google can do with it at scale.

  7. I was finally able to disable the smartwatch emergency calling “feature”… Last time it called 911 when using power tools. Before that, dozens of times during enjoyable bicycle rides. Not a convince…

  8. Now I can give data aggregation companies access to my heart rate, respiration rate, and galvanic skin response so they can align it with my realime gps location and accelerometer data for “linger time” at any given location!

    Why wouldn’t I want my physiological impression data associated with whatever thing I happen to be standing next to?

    How else are the marketing companies going to generate ads on demand, specifically for me, if they don’t know my subconscious preferences for fonts/colors/kerning?

Leave a Reply to rclarkCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.