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.
Good morning health tech readers!
Last evening, after a dreary, rainy Memorial Day weekend in New York, the sun finally poked its head out. Just in time for us to get back to work.
Reach me: [email protected]
What have we learned from Doctronic’s AI experiment in Utah?
After five months, the state of Utah released some data about its experiment with AI startup Doctronic. The state earlier this year authorized the company to use an AI chatbot to renew prescriptions for roughly 200 medications. The pilot operates under a program that allows the state’s Office of AI Policy to waive regulations for AI companies that agree to oversight and safeguards. The hope is that the state will learn more about what kinds of rules make sense for AI in health care. The Utah Medical Licensing Board recently came out forcefully against the pilot. At the time of the launch, experts told me
, and that it’s not clear an AI bot can legally renew prescriptions without
the plan is riskyFood and Drug Administrationauthorization.
Currently, the pilot is in its first phase, where all of the AI’s decisions are being reviewed by clinicians. We don’t know is how many sessions Doctronic has done in Utah, but the state says the number is “limited.” So what exactly have we learned?
The AI renews prescriptions 72% of the time without escalating to physician for reasons like missing lab work. In 69% of cases, reviewers determined escalations were appropriate.
The AI doesn’t always agree with humans. In cases where AI granted the renewal, physicians agreed 91% of the time. In cases where there was a disagreement, second clinicians disagreed with the first reviewer most of the time. In 3% of cases, neither clinician agreed with the AI.
Bias. A main drawback of the data is that all of this supervision and review is being done by Doctronic’s team. But state officials are expected to do an independent review. Zach Boyd, director of Utah’s AI office, told me the state has obtained an anonymized sample of conversations that experts will review “to both confirm Doctronic’s reports and give qualitative feedback for how to improve the pilot where that may be necessary.”
Doctronic Co-CEO Adam Oskowitz said the early data underscores the safety of the system and its guardrails: “No adverse events or contraindicated prescriptions were identified in the early results of Phase 1,” he wrote to me. “Phase 1’s pre-approval gate functioned as intended with no inappropriate prescriptions identified because nothing got through that shouldn’t have.”
While the numbers are interesting “We should be careful about what they prove,” said Girish Nadkarni, chief AI officer at the Mount Sinai Health System in New York:
“We should not let early operational numbers get translated into ‘AI prescribing is safe.’ The fair conclusion is narrower: a constrained AI system may be able to help process selected refill requests, but we still need independent evidence that it is safe, equitable, and better than a well-designed physician-supervised workflow.”
BigHat Biosciences CEO on what creates an edge in AI drug design
BigHat Biosciences works with some of the biggest drugmakers: It has partnerships with Merck, Amgen, Abbvie, Lilly, and recently completed work with Johnson & Johnson. The company also has two candidates of its own headed to the clinic.
Brittany Trang caught up with CEO Peyton Greenside in San Mateo last week to talk about the actual hard parts of AI drug design. “If you want me to design you a protein right now in six hours, I’m happy to do it,” she said.
AI developers aren’t taking advantage of FDA policy that allows speedy updates
A common complaint from AI developers is that that the technology moves so quickly that the plodding pace of FDA review can’t keep up. Developers must lock their software for studies and submissions, so by the time a product is on the market, it might be several years old.
New research suggests developer aren’t taking advantage of a policy that allows developers to get pre-approval from FDA for changes. Researchers analyzed public documents from 794 AI-enabled medical devices authorized from 2023 to 2025, covering roughly the time period since FDA finalized guidance on predetermined change control plans for AI devices. Just 42 devices had plans on file.
The researchers noted a modest increases starting in 2025, which may reflect a lag in taking advantage of the policy. Still, “low overall adoption indicates that there remain unidentified barriers to PCCP use or that certain AI-enabled devices may simply be unamenable to prespecified design changes that preserve safety and effectiveness,” the authors wrote.
Oura inches toward public filing
Smart ring maker Oura last week announced it had confidentially filed a draft registration to go public. Speedy work, considering the company just announced it had raised $900 million last October.
Exits & Outcomes analyzed filings in other countries that suggest the company’s revenues were $924 million for its fiscal year that ended in September 2025, more than double the $396 million it earned the year before.
E&O author, Brian Dolan writes: “One of the thresholds for filing for an IPO as an ‘Emerging Growth Company’ is total annual gross revenues under $1.235 billion. While I’m not sure that’s incentive enough for Oura to hurry up and IPO in 2026, it seems likely the company would not qualify for that status if it filed next year. This way, Oura would only need to disclose its past two years of financial performance among other things.”
Dolan cautions that “accounting practices are different country-by-country, so these numbers may look a bit different” in the filing to go public in the United States.
