medical devices
Proving to FDA that brain implants work
As much as anything we talk about in health technology, brain-computer interfaces sound like futuristic marvels straight out of science fiction: Implant hardware in someone's skull and grant them new powers to communicate, or move. Only they're not science fiction — they're medical devices meant to help people with disabilities. And like all risky medical devices, these products must demonstrate to the Food and Drug Administration that they are safe and effective.
Easier said than done, Rose Broderick reports in a fascinating new story. The companies furthest along in development, like Neuralink and Synchron, haven't yet declared primary endpoints for pivotal trials, which highlights a key hurdle: Unlike typical medical devices, these brain implants don't cure or change underlying medical conditions. Rose's story dives into the quest to develop clinical outcome assessments that capture how the technology improves the health of people with ALS and paralysis by enabling them to move, speak, and expand their overall functioning.
Read more here
Artificial intelligence
A first crack at studying ChatGPT Health's medical capabilities
One of the interesting things about large language models in health care is that researchers can very quickly spin up and complete studies of their medical capabilities. Within weeks of the launch of ChatGPT in 2022, researchers pushed out a pre-print illustrating the bot's performance on the U.S. Medical Licensing Examination.
I was reminded of that speedy turnaround this week when researchers published a paper examining ChatGPT Health's medical judgement. The paper was submitted on January 15th — eight days after OpenAI announced the product.
The high-level findings have been widely discussed on my feed: Presented with 60 vignettes, the product under-triaged 52% of cases and researchers noted that "crisis intervention messages activated unpredictably across suicidal ideation presentations." The researchers say their findings suggest consumer-facing AI-triage systems should be prospectively validated before launch.
These performance questions are super interesting and important, and I'm glad that researchers are developing rigorous tests, if only because it will push developers to improve their products. ChatGPT Health, it should be noted, is not intended as a tool for medical triage.
One thing I've been asking people in interviews is how we are going to measure the population health impacts of products like ChatGPT Health? One hypothesis is that consumer chatbots can give people better access to health care information and that this will help them live healthier lives. And that this benefit outweighs the negative health outcomes tied to bad health advice people take from the bots. Can we prove it?
business
Teladoc ups and downs
Teladoc reported earnings this week shedding light on its progress on turning its mental health subsidiary BetterHelp into an insurance covered business from one marketed to cash-paying consumers. By the end of the year, the company expects to do $75 to $90 million in insurance-covered revenue for BetterHelp. Its competitor Talkspace, which made a similar transition several years ago, just reported that its total payer revenue for 2025 was $171 million.
Overall, BetterHelp continues to sag. Revenues for 2025 were $950 million, down 9% from the year before. In 2026, Teladoc expects BetterHelp revenue to decline be .5% to 7%.
The more interesting development is that Teladoc is predicting it will lose members in its integrated care segment this year as some health plan customers reduce spend owing to the expiration of Affordable Care Act subsidies. It projects it will finish 2026 between 97 million and 100 million members versus nearly 102 million at the end of 2025.
No comments