FDA moves on AI tools, bionic pancreas research, & equity in digital therapeutics

You’re reading the web edition of STAT Health Tech, our guide to how tech is transforming the life sciences. Sign up to get this newsletter delivered in your inbox every Tuesday and Thursday

FDA’s new AI guidelines create uncertain future

The Food and Drug Administration roared into the week with a handful of dense documents that hold big implications for how it will oversee technology in health care. The most significant is the agency’s final guidance on what kinds of decision support software it intends to regulate, spelling out four criteria, pictured above, that determine whether things like AI software that flags medical conditions for doctors are considered diagnostic devices that must be reviewed by the agency.


The guidelines specifically call out tools that alert clinicians of sepsis as requiring agency review. The mention feels hardly accidental. Sepsis kills more than 200,000 people in the United States each year and is notoriously difficult to detect. STAT has published multiple investigations detailing the shortcomings of a widely-used tool developed by Epic.

Still, it’s unclear what impact the guidelines will have on dozens of tools that are already on the market. As Casey reports, the new rules have received a mixed reception. While one lawyer went so far as to call them illegal, others applauded elements of the rules which push vendors to provide clearer labels explaining how AI models were built and validated.


Elsewhere, an FDA report put the agency’s software pre-certification program to rest. The program was intended to explore streamlined review for firms that could show they maintained high quality standards, addressing concerns that the FDA’s clunky apparatus is poorly suited to the nimbler world of software development. Ultimately, regulators determined that while a quicker process might be feasible and desirable, the existing regulatory framework cannot be changed without legislation.

A big new study on a bionic pancreas

Since 2016, automated insulin delivery has helped people with type 1 diabetes avoid dangerous swings in their blood glucose. But so-called ‘artificial pancreas’ systems are still hard to use, requiring careful carb counting to help algorithms calculate the right insulin dose. Data from the pivotal trial of a new system from Beta Bionics, though, suggests that patients might be able to give up more power to the machines and still get good glucose control.

The diabetes device company submitted an application to the FDA earlier this year for its iLet bionic pancreas, which only requires a user to submit their weight and qualitative log of whether a meal is high, low, or average in carbs. “I think about my diabetes 24/7, and I think these systems just allow us to offset some of that,” said Jen Sherr, a pediatric endocrinologist who also has type 1 diabetes. Read more about the race to robotic insulin delivery in Katie’s latest.

A look at Akili’s equity tracking

To help ensure that its video game treatment for ADHD does not only serve people who are white and wealthy, Akili Interactive created an initiative called Equitable Benefit, which company leaders presented at DTx East this week. As part of the effort, the company scores how well the gender, race, ethnicity, and income of the users of its FDA-cleared product, EndeavorRx, align with the target population of kids with ADHD. The framework allows the company to analyze decisions like creating new versions of the game for more kinds of devices. As expected, when Akili expanded from the original iPad version of the game to iPhone, its equitable benefit score increased. They assumed that expanding to Android phones would have a similar effect, but it didn’t. That’s because though Android users were lower-income, they were also more likely to be men.

“This just highlights that even really kind of simple choices that we make, like expanding to Android can have complicated effects on how that impacts equity of your patient base,” said senior data science manager Andy Heusser.

The company is also looking at whether different groups use the product as intended and will ultimately study whether all groups get positive outcomes. It’s notable that the company is considering the issues before the app has been widely used in the market.

Industry news

  • Building on its growing stable of consumer health devices, Amazon announced Halo Rise, a new bedside sleep tracker and wakeup light. You’ll recall Google launched sleep tracking in its Nest Hub last year.
  • Rippl, which hopes to improve mental healthcare for seniors, raised a whopping $32 million seed round led by ARCH Venture Partners and General CatalystGVF-Prime Capital, and Mass General Brigham Ventures, also participated.
  • Grocery delivery company Instacart follows a parade of tech and retail companies hoping to make their mark on health. The very intuitively named Instacart Health is billed as a wide-ranging initiative including “product innovations, new partnerships, cutting-edge research and policy advocacy,” targeting “nutrition security, making healthier choices easier, and food as medicine.”
  • Mayo Clinic published new research about an AI screening tool that reads electrocardiograms to detect atrial fibrillation.
  • Headspace Health launched a program inviting outside researchers to submit proposals for projects that study the company’s products.

Personnel file

  • Babylon Health hired Andrew Hine as a vice president to lead the company’s overseas expansion. Hine joins from Ernst & Young.
  • Therapeutic chatbot company Woebot Health hired Brad Gescheider as chief commercial officer. Previously, he was global head of digital innovation and patient services for Sanofi’s immunology business.

What we’re reading

    • Can a telehealth startup add a layer of support to pregnancy care?, New York Times
    • Virtual drug screening identifies possible new antidepressants in LSD-like molecules, STAT
    • Health digital twins as tools for precision medicine: Considerations for computation, implementation, and regulation, NPJ Digital Medicine
Source: STAT