Dr. Dan Peters typically spends a shift in the emergency department dashing from patient to patient with a running list in his mind.
“I need to order meds for a patient. I need to order imaging studies. I need to call a consultant about a patient,” the ER doctor at UVM Medical Center described. “And on that task list is (that) I need to write a note about this patient.”
“To some extent, that task is a bigger task than everything else,” Peters said of writing a medical note. The note is a simple yet major part of doctoring that most patients hardly ever see. But for physicians, it looms large.
“I need to write a note that summarizes why this patient was here, what we did and what we thought about what’s going on or what wasn’t going on,” Peters said.
Medical notewriting is quickly changing for doctors across specialties with the advent of AI scribe technology. It’s been essential to reducing the burden of paperwork and the high levels of clinician burnout that result from it. Still, some worry about how to afford the expensive technology and about the risks of over-reliance on it for medical decision-making.
With an AI scribe, Peters begins his exams by asking his patient for consent and then hitting record on an iPhone. He asks questions and does a physical exam, narrating everything he’s seeing, feeling and thinking. All the while, his phone is listening.
When he stops recording, the AI scribe has cast the conversation into a full note — replete with the patient history, visit synopsis, exam results and medical decision-making, each separated into its own section.
For Peters, it has been game changing.
Existential exhaustion
Primary care providers describe the same thing.
In 2024, the University of Vermont Health network started to use Abridge, an AI scribe, as an initiative to reduce clinician burnout. A study of the implementation found that 69% of the 50 participating primary care providers reported burnout before using the technology. After four months, only 24% reported burnout. After one year, that number leveled off to 36%.
It mirrors national trends: A report from the medical technology company Doximity found that more than 36% of clinicians who use AI see it reducing the amount of time that they spend working outside regular hours.
That same report found that of 3,151 physicians surveyed in the U.S., 54% use AI technology in some way in their practice. The most frequent use, Doximity found, is in doing a literature search (that’s 35% of those using it). The second most popular use is for AI scribe technology.
Dr. Anne Morris, a family medicine physician at UVM Health and the associate dean of Lerner College’s primary care residency, described how profoundly the technology has lessened the psychological weight of doctors’ never-ending administrative to-do list.
“You can know that you’ve seen patients for the last nine hours and done some really important things. It’s the knowledge that you then have to spend three more hours documenting it that is exhausting. It can almost be like, ‘Well, I put all that work in, but it’s never enough,’” she said.
Now, using the Abridge technology has become second nature. It has whittled down the amount of work time that spills out of normal clinic hours and into the evenings — what she called “pajama time.”
“It helps to take away some of that existential exhaustion,” she said.
But she’s also seen it adding to the quality of her visits with patients, in ways that also seem to stave off burnout.
It’s changing the ways providers communicate, she said. There is no laptop propping up a barrier between the patient and provider, and there is not the same pressure to rush through a visit to get to the next one.
“When you can sit and fully give your attention to your patients and not have to worry about, ‘Did I document that correctly?’… The visit itself has more meaning,” Morris said. “That is what helps to bring that sense of fulfillment.”
In the emergency department, Peters has felt the same thing: The technology has driven him to articulate his thoughts more clearly and talk his patients through his decisions even more than before, he said. There’s this underlying incentive that, the more he communicates aloud, the easier his documentation will be later.
Medical decision-making
Peters sees the glimmers of how using an AI scribe could improve the scope of care he can give. As an emergency medicine doctor, he said, he needs to be something of a generalist, prepared to field a little of everything and anything. But, a large language model can hold a generalist’s breath of knowledge, with a specialist’s depth of detail.
Recently, he had a patient with an unusual pathology. As the patient described the symptoms, the AI scribe suggested a syndrome that Peters hadn’t known the name for. When Peters looked it up in a separate medical database, it was exactly what the patient was experiencing.
This type of diagnostic suggestion is still limited, he said, and still requires a level of gut-checking, but he is sanguine about what that power of suggestion could mean in the future.
“There is more knowledge in these (large language models) than could ever be consumed by a human being,” he said.
But at the same time, the work of distilling and understanding complex cases and crafting medical decisions is the art of the trade that gives him, and many doctors like him, fulfillment. He does not want to lose that.
Morris, too, worries about building an overreliance on AI for clinical decision-making, especially for her residents.
For the first six months of the residency, she said, the training doctors have to do all their notes by hand to learn the skill of creating an assessment without any artificial suggestion. She hopes that practice gives them the ability to check AI once they do begin using it.
Broader Applications
At UVM Health, AI technology has made its way into the exam room beyond scribing, in narrower applications. Dr. Justin Stinnett-Donnelly, who works as the hospital group’s chief health information officer and in internal medicine at Central Vermont and Porter medical centers, described how useful AI is in radiology. One tool, called Gleamer, uses AI algorithms to analyze X-Rays and highlight fractures.
Another, CathWorks, analyzes blood pressure and flow around the heart. Without it, doctors physically insert a wire with pressure monitors on either end to get a measurement. The use of the imaging technology, he said, prevents invasive surgery and saves money while achieving the goal.
In his work at Central Vermont Medical Center, Stinnett-Donnelly uses a band placed around a patient’s head that uses an AI algorithm to monitor whether their brain activity follows the patterns of a seizure. He uses it to gauge whether a patient can safely stay at CVMC, in Berlin, or needs to be transferred to the higher acuity medical center in Burlington.
The network also uses AI technology, Stinnett-Donnelly said, for “back office work” — like matching care with the right billing codes.
He sees it as a way to cut down all the paperwork that exists in the healthcare system between the moments where a patient is sitting in an exam room talking to their doctor.
Stinnett-Donnelly leads a group at UVM Health that guides the implementation of AI in the clinical space. The ethics, cost and security of the technology are concerns the group does not take lightly. Stinnett-Donnelly hopes that comes across to patients, especially the importance of privacy.
“It is a real fear,” he said, “and we do everything we can to respect individual autonomy and decision-making.”
Anecdotally, it’s rare that patients do not consent to doctors using the AI technology, he said. Morris and Peters agreed. The scribe does not save the recording and then its notes go directly into the electronic medical record that the network has been using for years. Morris said that the process ensures the notes are as secure as the medical record itself.
Still, these scribes embedded into the electronic health records can be so much more expensive — often prohibitively so for independent practices that are not connected to a big institution, like UVM Health.
Rick Dooley is a family practice physician assistant at Thomas Chittenden Health Center in Williston. He’s emphatic about how important AI scribe technology has been in his independent primary care practice. He uses a medical AI scribe, called Heidi, in almost the same way as Morris uses Abridge in her primary care practice — except, Dooley has to transfer the scribe’s note into the medical record, rather than having it already embedded within it.
OneCare’s primary care payment program had covered the cost of Heidi through July, at which point, Dooley suspects the practice will take over paying for it. At about $100 a month, Heidi is worth it, he said, and so much less expensive than paying for a transcription service or outsourcing to human transcriptionists — which ran about $150 to $1,200 a month, respectively.
A scribe that lives in the electronic health record system could be 10 times more costly. His clinic isn’t there yet, Dooley said, but he realizes there may be a day when the technology can take over enough busywork that it pays for itself, by opening up enough time for him to see more patients.
But, that’s not the world clinicians live in quite yet. The weight of paperwork surrounding healthcare still hangs heavy on providers.
“[AI Scribes are] a big piece, or a good-sized piece,” Dooley said, “but there’s so much other administrative burden that I think we just need to get rid of somehow.”
