Dr. Dan Peters typically spends his shifts in the emergency department running from patient to patient with a to-do list in mind.
“Medications need to be ordered for the patient. Imaging tests need to be ordered. A consultant needs to be called about the patient,” explained an ER doctor at UVM Medical Center. “That task list says I need to write a note about this patient.”
“In some ways, that job is bigger than all the other jobs,” Peters said of creating medical records. This note is a simple but important part of clinical practice that most patients rarely see. But for doctors, it’s a big problem.
“We need to write a note summarizing why this patient came here, what we did and what we thought about what’s happening and what’s not happening,” Peters said.
With the advent of AI scribing technology, medical note writing for physicians across specialties is rapidly changing. This was essential to reduce the administrative burden and resulting high levels of clinician burnout. Still, some worry about how to purchase expensive technology and the risks of over-reliance on technology for medical decision-making.
Using an AI scribe, Peters begins the test by asking the patient for consent and typing a note on their iPhone. He will tell you everything he sees, feels, and thinks while asking questions and performing a physical examination. All the while, I keep listening to his cell phone.
When he stops recording, the AI scribe writes out the conversation into complete notes. The notes are filled with patient medical history, visit summaries, test results, and medical decisions, each in its own section.
For Peters, it was a game-changer.
existential fatigue
Primary care providers explain the same thing.
In 2024, the University of Vermont Health Network began using Abridge, an AI scribe, in an effort to reduce clinician burnout. An implementation study found that 69% of the 50 participating primary care providers reported burnout before using the technology. After four months, only 24% reported burnout. A year later, that number leveled off at 36%.
This reflects national trends. According to a report from medical technology company Doximity, more than 36% of clinicians using AI feel that AI reduces the amount of time they spend working outside of normal business hours.
The same report found that of 3,151 physicians surveyed in the U.S., 54% use some form of AI technology in their practice. According to Doximity, the most frequently used search is literature search (35% of users). The second most common application is AI scribing technology.
Dr. Anne Morris, a family medicine physician at UVM Health and associate dean for primary care training at Lerner University, discussed how this technology has significantly reduced the psychological weight of physicians’ never-ending administrative to-do lists.
“You can spend the last nine hours seeing a patient and know you did some really important things, and then you know you have to spend another three hours documenting it, so it’s exhausting. You might feel like, ‘I put in all this effort and it’s never enough,'” she says.
Now, using Abridge technology has become second nature. That cut down on her regular office hours and into the evening, or what she called “pajama time.”
“It helps relieve some of the existential fatigue,” she said.
But she also feels it has improved the quality of her patient visits and prevented burnout.
The way providers communicate is changing, she says. There are no laptops as barriers between patients and providers. There’s also no pressure to rush through one appointment to get to the next appointment.
“When you can sit back and give your full attention to the patient and not have to worry about, ‘Did I record that correctly?’…the visit itself becomes more meaningful,” Morris said. “That brings a sense of fulfillment.”
Peters feels the same way in the emergency department. He said technology has allowed him to express his thoughts more clearly and explain his decisions to patients more thoroughly than before. There is an underlying incentive that the more he communicates aloud, the easier it will be to document later.
medical decision making
Peters sees glimmers of how using an AI scribe could improve the scope of care he can provide. As an emergency physician, he said, you need to be something of a generalist who can do a little bit of everything. However, a large language model can hold the rich knowledge of generalists and the detailed knowledge of specialists.
Recently, he had a patient with an unusual medical condition. As the patient described his symptoms, the AI scribe suggested a syndrome Peters didn’t know the name of. When Peters looked it up in another medical database, it was exactly what patients were experiencing.
He says this kind of diagnostic suggestion still has its limitations and requires some gut-checking, but he’s optimistic about what its power will mean in the future.
“These (large-scale language models) contain more knowledge than humans can consume,” he said.
But at the same time, identifying and understanding complex cases and making medical decisions is a fulfilling profession for him and many doctors like him. he doesn’t want to lose it.
Morris also worries that we will become overly reliant on AI for clinical decision-making, especially for trainees.
She says that during the first six months of training, residents are required to make all their notes by hand to learn the skill of creating evaluations without artificial suggestions. She hopes that with practice, she will be able to check the AI when she actually starts using it.
Wider application
At UVM Health, AI technology is being brought into the laboratory beyond scribing for narrower applications. Dr. Justin Stinnett Donnelly, chief health information officer for the hospital group and internal medicine specialist at Central Vermont Medical Center and Porter Medical Center, discussed the utility of AI in radiology. One tool, called Gleamer, uses an AI algorithm to analyze X-rays and highlight fractures.
Another, CathWorks, analyzes blood pressure and flow around the heart. Without this, doctors would have to physically insert and measure a wire with a pressure monitor on each end. The use of imaging technology can prevent invasive surgeries and save costs while achieving goals, he said.
Dr. Stinnett Donnelly of Central Vermont Medical Center places a band around a patient’s head and uses an AI algorithm to monitor whether a patient’s brain activity follows a seizure pattern. He uses this to determine whether a patient can safely stay at CVMC in Berlin or needs to be transferred to Advanced Medical Center in Burlington.
Stinnett-Donnelly said the network also uses AI technology for “back-office operations” such as matching care to the appropriate billing code.
He sees it as a way to cut out all the red tape that exists in the health care system while patients sit in exam rooms and talk to their doctors.
Stinnett-Donnelly leads a group at UVM Health that guides the implementation of AI in clinical settings. The ethics, cost and security of this technology are concerns the group does not take lightly. Stinnett-Donnelly especially hopes that patients will be reminded of the importance of privacy.
“This is truly scary. We will do everything we can to respect individual autonomy and decision-making,” he said.
Anecdotally, patients rarely consent to doctors using AI technology. Morris and Peters agreed. The scribes do not keep records and their notes are written directly into the electronic medical records that the network has used for years. Morris said this process ensures that the notes are as secure as the medical records themselves.
Still, the cost of these scribes embedded in electronic medical records can be much higher, and is often cost-prohibitive for independent clinics not affiliated with large institutions like UVM Health.
Rick Dooley is a family physician assistant at Thomas Chittenden Health Center in Williston. He highlights how important AI writing technology is in independent primary care practices. He uses a medical AI scribe called Heidi in much the same way that Morris uses Abridge in his primary care practice. However, the difference is that Dooley must transfer the scribe’s notes to the medical record, rather than embedding them already in the medical record.
OneCare’s primary care payment program covered Heidi’s costs until July, but Dooley suspects the provider will take over the payments at that point. At about $100 a month, Heidi is worth the money, he says, which is much cheaper than paying for a transcription service or human transcription expert, which each run about $150 to $1,200 a month.
The cost of a scribe using an electronic medical record system can be 10 times higher. Dooley said his clinic is not yet complete, but he recognizes that one day there may be enough time to see more patients and technology to take over enough busy work to pay for itself.
However, clinicians do not yet live in such a world. The burden of administrative procedures surrounding healthcare continues to fall on healthcare providers.
“[AI Scribes are] Dooley said, “There are large pieces, or good-sized pieces, but I think there are other things that are just too much of an administrative burden that we need to get rid of somehow.”
