You’re reading the web edition of STAT’s Health Tech newsletter, our guide to how technology is transforming the life sciences. Sign up to get it delivered in your inbox every Tuesday and Thursday.
A closer look at Google’s ‘loss of pulse’ feature
With Google’s annual health-focused PR event coming up in March, the company this week announced a flurry of updates.
First, Google announced that the Food and Drug Administration cleared a loss of pulse detection feature for its Pixel 3 smartwatch. It’s a 510(k) clearance, and as of writing, it’s not posted to the FDA’s website. The feature uses the wearable’s sensors to detect that your heart has stopped beating — Google cites cardiac arrest, respiratory or circulatory failure, and overdose or poisoning as reasons why this might happen. After a brief check-in period, the smartwatch sounds an alarm and starts a countdown. If there’s still no response, the watch uses uses built-in cellular or a connected smartphone to contact emergency services with context and location. Google plans an education campaign with emergency services groups so they know what exactly to make of these new calls. The feature is not meant for people with preexisting heart conditions or who need cardiac monitoring.
Google this week also published a Nature paper describing the development of the loss of pulse feature. Since it’s not feasible to hand out wearables and sit around waiting for hearts to stop, the company used simulated events to train its model, including planned testing of implanted defibrillators by electrophysiologists and using a tourniquet to cut off circulation to an arm. There was concern that false positives would put undue strain on emergency services, so Google opted for absolutely certainty of crisis in its design. The feature has sensitivity of 67.23% and a specificity of 99.987%. In other words, roughly 67 out of 100 loss-of-pulse events will be detected, but the product will only improperly flag an event .013% of the time. The company claims there was only one errant emergency call in its prospective testing.
My big question when this feature was announced last fall was how big a difference it could make. As the Nature paper notes, unwitnessed cardiac arrest is “nearly unsurvivable.” CPR needs to be started almost instantly or brain damage and death quickly follow. I imagined collapsing alone in my home while making my mid-morning toast. Even if the feature detected I was in trouble, by the time emergency services arrived and got in, I myself would be toast. But Jake Sunshine, a research scientist at Google, pointed out to me that many times, loss of pulse happens unbeknownst to people around who can help — maybe your spouse in the next room or sleeping in bed next to you. If they hear an alarm, your chances might greatly improve. Given the feature was designed with a low false positive rate, I admit, I find myself thinking — why not?
Also this week, Google released a report about the company’s “impact on health,” which provides a nice survey of the company’s disclosed work.
Teladoc reports yearly revenue decline, ongoing BetterHelp troubles
On Wednesday, Teladoc Health reported that its 2024 revenue dropped 1% to $2.57 billion. The company projected a revenue range between $2.47 billion and $2.58 billion for 2025, meaning Teladoc expects a sliver of growth at best, and another yearly decline at the lower end of the projection.
Teladoc’s virtual mental health company BetterHelp continues to be a drag on overall financial performance, recording an 8% drop in revenue compared to last year. Looking ahead, Teladoc expects another yearly revenue decline of between 3.75% and 9.75%. BetterHelp’s direct-to-consumer model is dependent on pricy advertising to draw in new users who can be cost sensitive and quick to churn out. On the company’s earnings call, CEO Chuck Divita cited product improvements to keep users engaged, international expansion, and a new weekly pricing model as initiatives that could lead to stability and growth.
BetterHelp’s struggles overshadow Teladoc’s modicum of success with the integrated care services it sells to health plans and employers, which grew 4% in 2024. The company said it could grow between o% and 3% in 2025.
Teladoc experienced huge growth during the pandemic but has since struggled to turn its tremendous user base into a profitable business. Since taking over last year, Divita has been making changes he claims will right the ship. “I am confident that the actions we’re taking this year will provide a strong foundation for us to deliver improved performance over the longer-term,” he told investors. Analysts, writing after the earnings call, speculated that things could start looking up in 2026.
Oura’s fresh women’s health push
Smart ring maker Oura announced a number of new women’s health-focused initiatives. First, a study with Scripps Research Digital Trials Center to investigate physiological changes during pregnancy and complications in hopes of identifying possible warnings signs. The study hopes to enroll 10,000 participants. (The research reminds me of the BUMP study that I wrote about a few years ago, which used, among other sensors, an Oura ring.)
The company is also working with researchers at Stanford on a Study on Typically Ignored Groups of Menstruating Adults, which will aim to recruit people typically not represented in women’s health research.
Finally, the company announced that the readiness score that it shows users “now takes into account the biometric changes that occur throughout the menstrual cycle.” The change will prevent naturally occurring fluctuations in physiology from registering as signs of strain.
A digital therapy for chronic cough
Hyfe, which develops cough-detection technology, this week announced a new deal with Japanese pharma company Kyorin to develop a digital therapeutic to address chronic cough for the Japanese market. The treatment will deliver Behavioral Cough Suppression Therapy, which includes education and suppression strategies. Hyfe’s cough-detection technology, which can be embedded in smartphones and other devices with microphones, will be used to track cough counts, frequency, and bouts, which the company hopes will increase efficacy by promoting, “engagement, motivation, and adherence.”
Kyorin distributes Merck’s drug for chronic cough, gefapixant, in Japan and for right now, the company plans to market the new digital treatment independently of the drug. The FDA rejected gefapixant in the United States, and Hyfe sees an untapped opportunity here. The company eventually hopes to bring its digital treatment through the FDA and is open to partnerships to bring it to the U.S. market.
Details on Color’s work with OpenAI
Brittany Trang writes: Last fall, we reported on how health care organizations are using OpenAI’s models, even though they aren’t optimized for health care. Many uses weren’t very specific or transparent, but today, cancer screening and care management company Color Health published the details of its OpenAI and UCSF partnership to test a “Large Language Expert” for helping to provide treatment plans for cancer patients. The publication delves into how the companies combined reasoning LLM models like OpenAI’s GPT-4o and o1 with clinical criteria decision logic to develop the tool.
Note: In the Tuesday edition of STAT Health Tech, I misspelled California Representative Ro Khanna’s surname in the subject line. I regret the error.
What we’re reading
- Are we adopting AI scribes based on… vibes?, STAT
- Human therapists prepare for battle against A.I. pretenders, New York Times
- Under siege after a tragedy, UnitedHealth grapples with fresh security threats and a customer backlash, STAT