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.
On Friday, Mehmet Oz, President Donald Trump’s pick to run the Centers for Medicare and Medicaid, turned on the charisma and cruised through a confirmation hearing before the Senate Finance Committee that featured a lot of substantive discussion of policy. For an excellent analysis read STAT’s Tara Bannow’s coverage of the hearing. Health tech, including telehealth and AI, came up a few times:
- On AI, Oz made a lot of statements most people would find totally uncontroversial, including about the ability of technology to augment care and reduce paperwork for clinicians, and about the need for guardrails to ensure bad things don’t happen.
- Oz noted that AI could be used to speed up pre-authorization for medical procedures. “We have AI support tools, navigation systems that could pretty quickly adjudicate whether you should have to wait even a day to get the medication that will get you out of pain, or even a week for the procedure that you should be allowed to have,” he said, “It’ll take a lot of the angst out of the system for the American people.”
- Pressed about AI being used by Medicare Advantage insurers to make determinations about whether to deny care, Oz suggested AI could be used to spot the problem: “We should be using AI within the agency to identify that early enough so that we can prevent it.”
Evaluating AI on health tasks that actually matter
Last year, a review of 519 health care large language model studies published between 2022 and 2024 showed that 45% of them tested the artificial intelligence models’ knowledge by asking them multiple-choice questions like those you might see on the U.S. Medical Licensing Examination. But performing well on multiple-choice questions does not necessarily translate to the tasks LLMs might be asked to complete in the real world, like being able to summarize a patient record, come up with appropriate billing codes, or diagnose a patient.
In a new story, Brittany Trang looks at a new AI language model evaluation framework from Stanford and talks to Stanford Health Care chief data scientist Nigam Shah about how the new evaluation framework — which the team has already run on models like Google’s Gemini 1.5, OpenAI’s GPT-4o, and Meta’s open-source Llama — can help health systems decide which models to use for the health care tasks that they care about.
Why disabled people are at risk from AI
O. Rose Broderick writes: A new report highlights how health technologies that rely on AI and algorithmic systems can fall short for people with disabilities. The report is another red flag about how a patients’ identities can affect the care they receive from health technologies and how such systems often struggle to adequately serve people from marginalized communities. Until recently, much of the research in this space has centered around race and gender, but more researchers are turning their eyes towards issues that affect nearly 20% of the population — Americans with disabilities.
“For people with disabilities, specifically, there’s a real risk of discrimination. Technologies are not being developed, deployed, and sometimes audited in ways that are inclusive of people with disabilities,” said Ariana Aboulafia, the disability rights in technology policy lead at the Center for Democracy and Technology.
Google’s big health event preview
Today, I’m heading to Google’s annual Check Up event in New York, where the company will share he latest about how its technologies, including consumer products and large language models are being applied to health. I may have more to say in the Thursday edition, after the event, but Google this morning published a preview of what the company will be discussing. You can stream the event yourself at 11 a.m ET.
- In search, Google wrote that it has made improvements to its AI Overviews for health related topics so that they are “more relevant, comprehensive, and continue to meet a high bar for clinical factuality” thanks to “recent health-focused advancements on Gemini models.” At Google and elsewhere there’s an increasing focus on making sure that generative AI products provide reliable answers to health-related questions.
- Google also announced a new update that could make it easier for users of Android phones to access information about care they received from providers. The company’s Health Connect app, that allows users to sync different kinds of data between apps, has a new API that allows apps “to read and write medical record information like allergies, medications, immunizations and lab results in standard FHIR format.”
- Google will talk about the loss of pulse detection feature for Pixel smartwatches and its new AI co-scientist, both of which we’ve discussed recently in STAT newsletters. Information distributed in advance makes no mention of health equity, as expected, after the company recently dumped the concept in favor of “health optimization” as President Trump attacks diversity, equity, and inclusion both in the federal government and across the private sector.
- In an announcement, no doubt timed to the event, Quest Diagnostics revealed a new deal with Google Cloud that includes using data analytics and generative AI technologies to “give patients personalized health insights.”
What we’re reading
- Opinion: HIPAA can’t keep up with health care’s security crisis, STAT
- Inside Elon Musk’s ‘Digital Coup’, Wired
- Is Google playing catchup on search with OpenAI?, MIT Technology Review
- Feds shutting down texting service platform Notify.gov for Medicaid benefits, Fierce Healthcare
- As AI nurses reshape hospital care, human nurses are pushing back, The Associated Press