AI Can Be an Asset to Clinicians and Health Systems — If It Is Properly Introduced

—
At one large health system, it’s saving time and increasing doctors’ satisfaction

by
Joyce Frieden, Washington Editor, MedPage Today
May 11, 2025 • 4 min read

PHILADELPHIA — Artificial intelligence (AI) can be seen as an asset to health systems rather than a threat to the workforce — if it is sold properly, several speakers said here at the Population Health Colloquium hosted by Thomas Jefferson University.

“For us, AI gets thrown around a lot,” said Arshad Rahim, MD, MBA, senior vice president and chief medical officer for population health at Mount Sinai Health System in New York City, during a panel discussion on Wednesday. He said he sometimes feels as if “this is just data mining, and we’ve been doing [it] for 20 years, but I’m really glad [our AI team] is here, and they’re probably overqualified for what we’re trying to do, like finding people with LDLs greater than 190.”

Rahim said he’s looking at using AI to scale up performance of rote activities. “I don’t think it’ll ever necessarily replace a team member, but I think I can augment scale further, whether it’s some parts of ambulatory care management or whether it’s patient outreach — I think there is definitely a ‘there’ there, and the huge part is going to be the change management,” he said.

AI is already being put to use at Mount Sinai, where “all of our radiology, — especially our scans, CTs, and MRIs — the first read is AI, and then [the radiologists] do an overview,” Rahim explained. “One of the radiologists that I really respect in our institution said, ‘AI will not replace radiologists, but a radiologist who knows how to work with AI is going to be a lot more valuable to organizations they’re serving, versus one that doesn’t, and doesn’t embrace it. And that’s how I’m selling it to our team members, too … because there is a lot of change management, and it’s naturally threatening.”

But not everyone on the panel was quite so far along. “As it relates to AI, we’re really just starting to dabble in it from the perspective of clinical decision-making,” said Grayling Dwayne Yarbrough, MHA, vice president for population health at Sentara Healthcare, in Virginia Beach, Virginia. “I wouldn’t say that we’re highly sophisticated quite yet in the AI space.”

At Kaiser Permanente, AI’s introduction — at least in one area — happened at lightning speed, Stephen Parodi, MD, executive vice president of the Permanente Medical Group in San Francisco, said in a separate panel on Thursday. “One of the technology leaders at Kaiser Permanente said, ‘You know, we need to roll out ambient AI, and we need to do it fast, because people are burning out and they’re frustrated,'” he said; ambient AI refers to an AI system that operates in the background without direct user interaction, doing things like transcribing a patient encounter. “The traditional process that we would have done a decade ago, or even just 5 years ago — we would have planned it and we would have taken about a year to figure out how to roll it out.”

Instead, “it rolled about in 1 month to 24,000 physicians across the country, and they put together a 1-pager [on] how to use it — ‘If you’ve got a question, hit the hyperlink’ — and I’m like, ‘Oh my god, this is a disaster,'” he said. “But you know what? It was well received. Yes, there were some hiccups, there were a couple of failures along the way, but we didn’t bring down any systems. It was well received, and we didn’t take years to roll it out.”

After the AI system transcribes the encounter, the physician reviews it, edits it, approves it, and then moves on. “In the primary care specialties, we’ve seen significant reductions in the time clinicians spend documenting — and when I say significant, I’m talking on average, about 1 or 2 hours a day, which translates into less time that somebody’s spending probably off-hours doing that work, which leads to satisfaction from a clinician perspective,” said Parodi. In addition, “we started to see actual improvement in the number of patients that can be seen in a given unit, in particular in the primary care setting.”

The other thing that’s a little surprising is the level of patient acceptance of AI, he added. “You would think maybe people would be worried about security, and maybe people would be worried about privacy. But we have not had people decline having AI in the room listening, and that was very interesting to me. We have satisfaction surveys across the country now [in which] 92% to 93% of patients are saying, ‘This is a better interaction with my clinician. They didn’t spend time with the computer; they were actually talking to me. I feel heard.'”

The experience with ambient AI “gives us insight into some additional applications of what AI would look like,” Parodi said. “I think you’re going to have AI assisting in a more direct manner,” including helping with figuring out how to code visits and also with reporting to various regulatory and payer organizations. “I know we’re actively looking at that.”