UCSF Chancellor Sam Hawgood aims to use AI to ‘de-clog’ health care

By Ron Leuty – Senior Reporter, San Francisco Business Times
May 23, 2024

In the $7 billion business that is UCSF Health, Sam Hawgood knows there’s no magic bullet for improving health care. But nearly two years into an effort to insert AI where it can, the University of California, San Francisco, chancellor is seeing the technology’s benefit.

And his team is already course-correcting when artificial intelligence doesn’t work as planned.

It’s no easy task, Hawgood said Wednesday at Biocom California’s Converge Summit, an event designed to map the intersection of life sciences and technology. AI is ready to be part of the UCSF patient care team, for example, Hawgood said, making suggestions for how to treat patients. At the same time, UCSF Health must keep in mind patient privacy and trust, he said.

The key to AI’s introduction in parts of the UCSF system, Hawgood said, has been constant evaluation. AI was deployed to see which patients would be at risk of post-surgical disorientation, for example, a reaction that often leads to giving patients more drugs. The application didn’t work.

“It was bad. We pulled it,” said Hawgood, who had rolled out UCSF’s enterprise-wide AI push during his “state of the university” address last year.

With Microsoft Corp., UCSF developed what it calls Versa — short for “conversation” — a secure generative AI platform that includes a web interface that allows interactive chats using OpenAI’s GPT-3.5 and GPT-4 and a programming component, or API, that allows programmers and researchers to develop their own tools to automate tasks and process large databases.

Versa has 1,400 users in UCSF Health after it was turned on in the clinical space last year, Hawgood said. It was rolled out across the broader UCSF enterprise last month.

“The field is moving incredibly quickly thanks to companies like Microsoft and many other companies in the Bay Area,” Hawgood said. “We’re trying to both keep up but keep our eye on the ball in terms of what can we do that is not a pilot, is not a test, but is an enterprise-wide solution.”

In a space-constrained UCSF system that takes in 10,000 patients a year and turns away another 4,000, AI can help with the transfer of thousands of pages of medical records as well as medical images, Hawgood said, with the technology essentially triaging patients and selecting those patients UCSF is best-equipped to help.

“We can use these tools to say, we have to take that patient, we have no other alternative — if we say ‘no,’ it’s the end of the road for that patient,” he said. “Whereas another (patient), we could send them to X, Y or Z (facility) and they’d get good care.”

UCSF also is linking patient preferences and doctor preferences to an automated AI call center to make setting an appointment less of a friction point, Hawgood said.

“Our experts told me at the beginning of 2023 that they thought that would be the next 24-month journey and it would be quite a while before we got to more interesting applications,” Hawgood said. That timeline has accelerated.

AI assistants are “fundamentally ready” to be part of the patient care team, alongside doctors, nurses, residents and medical students during patients rounds, Hawgood said he was told this week during an AI update session he holds with staff every two months.

“For some of our more-adept users, that’s already in place,” Hawgood said.

The initial drive with AI was to free clinicians from inputting notes on the keyboard, back turned to patients during visits, Hawgood said. Now UCSF Health hopes to have all its clinics with patient-consented AI scribes by the end of this year, “so doctors don’t have to type anything.”

Ideally, that means more patient-facing time and less doctor burnout, he said. It also potentially means better access for patients who struggle to see a primary-care doctor and sometimes use the emergency room as a doctor’s office.

“If we can help — and I’ll use a term that sounds wrong — ‘de-clog’ the system for those kinds of issues … that’s an incredibly inefficient and expensive use of a very valuable, highly trained workforce,” Hawgood said. “I see that as a huge use-case, and people are working on it.”