Latest news with #EdwardShi
Tom's Guide
24-06-2025
- Tom's Guide
The Pixel Watch 3's biggest safety feature sounds like a gimmick, but I just read the study and I think it might actually save lives
When Google announced the Pixel Watch 3 last summer, the company promised it would get a life-saving new feature — Loss of Pulse detection — at a later date. I was skeptical, and not just because you shouldn't buy devices on what they might do. Even though it's clearly one of the best smartwatches (how could it not be after Google decided to bury the best Fitbits and incorporate that brand's fitness tracking smarts into its own watch), a life saving software update sounded like vaporware to me. But then in February this year, six months after the Watch 3 launched, Google was granted FDA clearance for the feature, and, in March, it began rolling out to all Watch 3 owners in the U.S. That was one objection overcome — the company did release it — but I was cynical that a feature on the Pixel Watch 3 would be as revolutionary as Google claimed. But then I read the team's study in the journal Nature, and it changed my mind. To go alongside the launch of the feature, the research team at Google published a study in the scientific journal Nature detailing how they developed the Loss of Pulse detection feature and the algorithm that detects when your heart stops beating. There's a lot of detail in there, which surprised me as Big Tech companies are usually very reluctant to discuss how these critical health features actually work. I guess that's not shocking; they do have products to sell and need an edge over the competition. But if you spend your money on a new smartwatch because it could potentially save your life in a specific circumstance, you want to be sure it can actually do that effectively. A recent experience on vacation brought home to me just how critical that is. I was involved in a vehicle collision while traveling, and had been wearing the Amazfit T-Rex 3 for testing. Thankfully, everyone involved was okay, so when taking stock later on, I realized this would have been a rare test of some critical features. Unfortunately, the T-Rex 3 doesn't have collision detection like the Pixel Watch or Apple Watch Series 10, so all I could do was look nervously at how my heart rate rocketed at the moment the two vehicles smashed into each other. But this got me thinking; how would you know for sure that the collision detection feature worked, until it's too late? And similarly, how would you know if the Pixel Watch 3 Loss of Pulse detection can actually save your life? Of course, when you think about it, the tech companies have the same problem. You can't really test a feature like this in the real world; you'd need to wait until someone has a cardiac arrest and then delay care to see if the watch does it's job. So the team came up with a lab testing method instead. When my colleague Dan Bracaglia interviewed Edward Shi, the product manager on the Google Safety Team, Shi mentioned that they enlisted "stunt" participants. The team used "a pneumatic tourniquet to cut off blood flow in an arm, to simulate temporary pulselessness," and then would see if the watch picked up on these signals. This is similar to how Apple tests its collision feature with crash test dummies. It's not a real world test, but it's an approximation. Though this is a good reminder that these features aren't guaranteed to work, so if you have a watch with safety tools like this, it's a good idea not to be solely reliant on them. And while there's a good appeal to authority with the 'FDA cleared' label, this isn't quite as useful as it seems. There are two main FDA labels; cleared and approved. If a product or feature is "cleared" that doesn't mean that it's shown to be effective. What it really means is that the FDA is aware of the feature and believes it to be similar to something else already for sale or in use, so the applicant (ie, Google) can market it. But the FDA has not actually verified anything at this stage. To see whether it works as intended, it'd need to be FDA approved. But just because Google didn't aim for approval, that doesn't mean they don't have confidence in it. In a past life, I worked in regulatory affairs for a medical device company. To gain FDA clearance, it's (almost) as simple as filing some forms. But FDA approval is the standard medicines are held to, so you need a lot of evidence and testing to even begin the very lengthy approval process, so many companies don't bother with this. Okay, so Google has been a bit more open about the Loss of Pulse detection feature than it's peers, but that doesn't quite explain why I changed my mind and stopped seeing the LoP feature as a gimmick. According to Shi, one of the main motivations to develop LoP was that "many of these [cardiac arrests] are unwitnessed," so someone's heart could stop when they're alone and no one would know, so no one calls for help. And the way I see it, even if the Loss of Pulse feature can't detect all events with absolute certainty, there'll be enough people who suffer some form of heart-stopping emergency who will get life-saving treatment because the watch called for help. Plus, the Nature study notes that the algorithm "was trained on smartwatch sensor data collected from diverse participants (age, sex and skin tone)." This is super important because different skin tones absorb light at different rates, effecting optical heart rate measurements the system is based on. Historically, tech companies have thought very little about training their systems on diverse, less homogenous sources, so I was relieved to see Google had at least thought about it (even if you couldn't verify this as participants' data is hidden due to protect their privacy). Do I now believe that it'll work perfectly every time? No. But even if it only works half the time, that'll still save lives, as without it, no one would come to the aid of people who have suffered a potentially fatal health emergency.
Tom's Guide
19-06-2025
- Tom's Guide
Google Exclusive: How the Pixel Watch 3 got a life-saving feature the Apple Watch can't match
Onboard safety features are a huge selling point of modern wearable devices. These days, the best smartwatches can automatically contact emergency responders and/or loved ones if you take a nasty fall or are involved in an accident, regardless of whether you're wearing the latest/greatest Garmin, the best Apple Watch, or the best smartwatch for Android. While fall, crash and incident detection are all but par for the course on high-end, full-feature smartwatches, a new, more advanced safety feature surfaced last summer that's currently only available on the Google Pixel Watch 3. That's right, not even the Apple Watch Ultra 2 or Samsung Galaxy Watch Ultra offers anything like Google's Loss of Pulse Detection tool. Like fall detection, Loss of Pulse Detection is designed to help users out during an emergency — in this case, a medical one, when there may otherwise be no one around. Better yet, setting up Loss of Pulse Detection takes less than 2 minutes, which is not a lot of time considering it could be a literal lifesaver. To find out more about Loss of Pulse Detection, including insights into the development, testing and FDA approval process, I had an exclusive interview with Edward Shi, the product manager on the Google Safety Team who spearheaded the project. Our 30-minute conversation covered a lot, but it's Google's creative approach to testing the new safety feature — something that's crucial for avoiding false positives — that most fascinated me. For one, Shi and his team had to figure out how to simulate a loss of pulse in a living subject, for testing purposes, of course, which is no easy feat. His team also worked with stunt actors to understand how a user may fall when a loss of pulse is experienced. Beyond that, our conversation touched on whether older Pixel Watch devices could get Loss of Pulse Detection in the future, how long until the competition replicates the feature and what the Google Safety Team is up to next. Get instant access to breaking news, the hottest reviews, great deals and helpful tips. Edward Shi: I'm a product manager here on our Android and Pixel Safety Team. Our team works on safety products with a goal of giving users peace of mind in their day-to-day lives. These products include, in the past, features such as car crash detection and fall detection. For Loss of Pulse, specifically, I'm one of the main product managers on the project, working across the teams, with our clinicians, our engineers, etc., to bring Loss of Pulse Detection to the Pixel Watch 3. It uses sensors on the Pixel Watch to detect a potential loss of pulse and prompt a call to emergency services with either the user's smartwatch or their connected phone. Shi: It's really for any Pixel Watch 3 user who meets our eligibility criteria. It uses sensors on the Pixel Watch to detect a potential loss of pulse and prompt a call to emergency services with either the user's smartwatch or their connected phone, who can then intervene and potentially provide life-saving care. A loss of pulse is a time-sensitive emergency, and it can be caused due to a variety of different factors, such as a cardiac arrest, a respiratory or circulatory failure, poisoning, etc. Many of these events are unwitnessed today. So around 50% of cardiac arrests, in particular, are unwitnessed, meaning that no one's around to help. Shi: The two main sensors are the PPG sensor as well as accelerometer. We use PPG to detect pulselessness, as well as the accelerometer to look at motion in particular. So if a loss of pulse occurs, what we anticipate is that the user is unconscious, so there shouldn't be excessive motion. So those two sensors combined help form the foundation of the algorithm. The algorithm is trying to balance both detecting that emergency, so in this case, a loss of pulse, while minimizing accidental triggers. Shi: There are a lot of similarities in the sense that all are emergency detection features. Essentially, these are for potential life-threatening emergencies in which a user may not be able to call for help themselves. In those events, we would need to be able to detect that emergency and then help connect [the user] with emergency services. Much of the design and the principles remain the same. The algorithm is trying to balance both detecting that emergency, so in this case, a loss of pulse, while minimizing accidental triggers. That's a really key part of all three of the features. We don't want to overly worry and bother the user with accidental triggers. Also, in particular, we don't want to burden [emergency] partners with accidental triggers in the case where a user doesn't need help. Shi: Once a loss of pulse [or] a car crash [or] a fall is detected, the experience is designed to try to quickly connect the user over to emergency services. If, for whatever reason, the user doesn't actually need help, the user experience is [also] designed so that they can easily cancel any call. We actually worked with stunt actors to induce pulselessness and simulate a fall within a reasonable timeframe to see if it was still able to detect a loss of pulse in those scenarios. Shi: I don't know if I could precisely say exactly how long, but definitely over a year and a half, but it can really vary. One particular [safety] feature isn't necessarily the same as the others. They may look similar on the surface, like a fall or a car crash or a loss of pulse, but each of them has its own unique challenges in validating both the algorithm and developing the user experience. And of course, with laws, we had to go through working with our regulatory partners and regulatory bodies in different regions [for Loss of Pulse Detection]. So there are different complexities for each of them, so the timeline can definitely vary. Shi: It's a bit of both. So, it's definitely algorithmically tested. We also collect hundreds of thousands of real-world user data and run our algorithm over that data to take a look at how often it could be triggered. Internally, we have "dog foods." And then we ran clinical studies. All of that is run to measure how often we're seeing accidental triggers in particular. In addition to honing the algorithm or user experience design, we run user research studies to look and walk [users through the] 'flow,' both during onboarding, as well as when an actual loss of pulse is detected. [We're] seeing that users understand what's happening and are able to cancel out of that flow if they don't need help. So, it's both algorithmic as well as user research. Basically, using a pneumatic tourniquet to cut off blood flow in an arm, [we were able] to simulate temporary pulselessness. Shi: It is pretty difficult, and it took a lot of creativity from our research scientists, in particular. Basically, using a pneumatic tourniquet to cut off blood flow in an arm, [we were able] to simulate temporary pulselessness. We were able to do that and then put our watches on the user at the same time to ensure that our algorithm was detecting that [loss of] pulse when it occurred. We actually worked with stunt actors to induce pulselessness and simulate a fall within a reasonable timeframe to see if it was still able to detect a loss of pulse in those scenarios. Shi: We're very fortunate at Google to have great team members who are familiar with the process and are regulatory experts. Receiving U.S. FDA clearance does go through a rigorous process to ensure quality and understandability of the products that are coming through. So really, it's taking a look at the U.S. FDA established regulatory frameworks and regulations, knowing what we have to conduct in terms of necessary performance testing, what we have to show to prove that the feature is doing what it [says], and in particular, that it's understandable to users who choose to use the product. Shi: The biggest thing that we inform users about, essentially during onboarding, is that it's only meant to detect an immediate loss of pulse. So it's not meant to diagnose or treat any medical conditions, and it's not meant to be a feature that gives you a pre-warning of any health condition. That's a really important distinction that we do try to make as clear as possible within the product itself, so that you don't change any health regimens, etc, and you don't change anything that you've heard from medical professionals. As always, go to your healthcare professional to discuss all of your well-being, etc., and what's best for you. Shi: It's something we can't go into detail about at the moment. We have to look at both the hardware that's available on the older Pixel Watches and see if it's possible. Also, we have to ensure that there is hardware equivalency on each of the different devices. So we have to make sure on the older Pixel devices, if we were to do [Loss of Pulse Detection], that it still performs as expected within the guidelines that we set. We would like to make [the feature] available as widely as we possibly can, so that's what we're going to try to do. Shi: Our top priority when we released this feature was to make sure that it maintains its quality and is able to do what it says it does within the guiding principles that we have. What we anticipate is that as new Pixel Watches are released, it's available on all different Pixel Watches. Of course, it's going to be a hardware-by-hardware validation. We would like to make it available as widely as we possibly can, so that's what we're going to try to do. Shi: I think this is definitely speculation and subjective, but I think in the tech world, people are always looking at other competitors and trying to close the gap or match different features. So I wouldn't be surprised if that's something that people did. In some ways, I think for our team, that this would be a good thing — with safety in particular — if other competitors started trying to copy features. I think as long as everyone maintains high quality, of course, then it's not necessarily a bad thing. But yes, I think it's fair to assume that people are looking at it and they attempt to copy it. Shi: We're always looking at helping users get connected with help if they aren't able to themselves. We know emergencies, hopefully, are a bit of a rare event in users' daily lives, but there could be other scenarios where users may feel unsafe. So, one of our existing features is a Safety Check. When users are going out for a run or going out for a hike and they want that extra peace of mind, they can start a Safety Check, and we can check in with them, and then if they don't respond, we can automatically share their location and reason and context with their emergency contacts. That's an existing feature, and also things that we're thinking about on the safety side. We're looking across the spectrum from emergencies to daily use cases of how we can help, how we can deliver a little bit more peace of mind in your daily life.
