explainers
What exactly is OpenAI doing in health care?
You may have seen recent reports about hallucinations in ChatGPT-maker OpenAI's transcription tools, and how they may worm their way into hospitals. But how involved is the company in health care, and is it making any money there? I set out to answer these questions in my latest for STAT, where I spoke not only to OpenAI but a host of health care customers, including Color Health, Paradigm, and UT Health Houston. Among tidbits I picked up: The company typically offers two types of enterprise licenses — one to its GPT technology, and another to its API, allowing developers to weave AI into their existing products.
These early adopters told me they're aware that generative AI can hallucinate, and they're still not confident enough in the technology to unleash it directly on patients or allow it to dispense medical advice. Instead, they're using it to summarize medical records for internal use or suggest treatment to specialists, among other functions.
OpenAI, for its part, does not and has never considered itself a health care company, head of platform sales James Dyett told me. "We see ourselves as a partnerships company that's building a platform upon which customers from many sectors, including health care, can build great applications," he said.
Of course, that perspective conveniently distances the company from thorny questions about whether AI is safe enough to use in health care at all. Whether the technology should be closely involved in diagnosis, or interact with patients without a human in the loop, for instance, is a "decision that will be made by the health care industry, not by OpenAI in isolation," he said. Read more.
exclusive
LA Children's hospital tests gen AI for translation
Speaking of generative AI in health care, my colleague Katie Palmer has a case study on Children's Hospital Los Angeles, which is piloting the technology to instantly translate discharge notes for Spanish speaking patients. (Thanks to nondiscrimination rules requiring health systems getting federal financial assistance to provide meaningful access to health information to patients with limited English, we could see these applications proliferate.)
It typically takes hospitals hours or days to translate discharge notes without machine translation — and some clinicians have defaulted to using Google Translate, despite the risk of mistranslation, Katie writes. "There's been a language access gap in written materials that nobody has felt particularly motivated to do anything about," Elaine Khoong, an assistant professor of medicine at the University of California, San Francisco, told Katie. Read more.
from our readers
Can ChatGPT help patients? One advocate is testing it out for herself
In my last edition I called out for any other creative use cases for ChatGPT. Liz Salmi, a patient advocate and director of communications and patient initiatives for the data transparency research group OpenNotes, said she's been playing around with the paid, secure version for personal, medical and professional reasons — including, but not limited to, asking it to churn through almost 17 years and several thousand pages of medical records related to her tumor care. She's found that it can accurately answer queries like "when did this incident happen" or "where was this procedure" that might otherwise be difficult for her to find; but it does sometimes struggle to separate results from her participation in experimental research trials with clinical-level tests, she said. (Her solution, she said, was to funnel her medical records and her research trial results into two separate GPTs.)
She's also used it at work at OpenNotes to generate communications copy, including to actually write the research group's idea for an April Fool's Day post touting open data sharing for dogs. She's even, she told me, used it to scan the 600,000 word journal of an uncle who passed away last year — with permission from his family, and express wishes from the uncle that his journal be disseminated posthumously. Now, she and her relatives occasionally query the custom, pseudonymous "MikeGPT" to find out what he might have thought about various topics based on his extensive writing. (Asked what he'd say to an inquiring journalist, Salmi said the GPT said her uncle had "hoped some might find value" in his journal, and that he conveyed a hope that his words "would offer a 'better understanding' of his life and struggles.") Have you used the technology for yourself? Let me know what you've found.
Washington
How Trump's AI work could diverge from Biden's
It's been four years since Trump's occupied the White House, and artificial intelligence has evolved dramatically since then — especially with the onset of generative AI. The Biden administration's spent the last year playing catch-up, issuing sweeping executive orders and setting up task forces to get ahead of its risks in health care, but the incoming party has already signaled an intention to let the technology flourish with fewer guardrails. My colleagues Casey Ross and Mario Aguilar polled policy experts on what they're expecting.
"The appropriate balance between innovation with AI and safety of AI in health care is bipartisan," Lucia Savage, a former federal regulator who now works at Omada Health, told STAT. "There are already regulations that tell us how to use health care data in a fair, private, secure, and safe manner to develop AI, and what appropriate care by health care professionals (using AI or not) looks like." Read more from Casey and Mario.
wearables
Pfizer cancer trial shows promise for digital endpoints
Mario also has a story on Pfizer's research into cancer-related weight loss and an experimental antibody that could positively impact it — as well as the digital endpoints that the drug maker used to evaluate its effects in a clinical trial. Patients' physical activity was measured in the trial by wearable devices, and Carrie Northcott, head of digital sciences at Pfizer, told Mario she thinks the industry is close to "a revolution of using these measures within clinical trials." Read more.
venture capital
Summa Health and General Catalyst advance deal
Almost a year after it was announced, Ohio safety net hospital Summa Health's planned acquisition by venture firm General Catalyst appears to be moving forward, though regulatory approval is still pending. Summa said late last week that General Catalyst's Health Assurance Transformation Company, known as HATCo, has reached a definitive agreement to pay $485 million for the health system, which would allow Summa to eliminate $850 million in its own debt; HATCo also plans to invest $350 million within the first five years of the agreement, and $200 million for "strategic and transformative investments and to drive innovation over the first seven years," according to a release.
The deal is closely watched by patient advocates and tech entrepreneurs and investors alike: critics warn that the venture capital ownership could pressure clinicians into using technology they don't need on safety net patients who don't have other options for care, but others argue a direct ownership model could force the health system to embrace value-based care faster than it might have otherwise. If you're watching the deal, let me know what you think.
No comments