Imagine waking up one morning feeling out of sorts. You call your doctor’s office and don’t get put on hold. Instead, a virtual receptionist, powered by artificial intelligence, determines from your conversation that you need to speak to a nurse, who immediately sends you to the hospital.
At the emergency department, a physician begins her exam by activating an AI-powered recorder on her smartphone, which automatically links to your electronic health records. As you describe your symptoms — a severe headache, dizziness, blurred vision — the AI transcribes and summarizes the conversation, allowing the doctor more time to review your medical history and delve into the cause of your symptoms.
Suspecting that you’re having a stroke, the doctor orders a head CT scan. An AI system that analyzes the scan flags a blood vessel in your brain with a potential blockage. The AI automatically alerts the hospital’s stroke team and texts your scan to a neurologist at another hospital for a human confirmation. Within minutes, the neurologist calls in a prescription for a clot-busting medication that likely saves your life.
“In acute stroke, timing is really critical. So that’s a case where minutes really do matter,” said Dr. Justin Stinnett-Donnelly, chief health information officer for University of Vermont Health, the state’s largest health network.
RapidAI, a system that helps physicians detect and identify strokes faster, isn’t science fiction or even experimental technology. This and other forms of AI are being used in Vermont every day to transform the way patients and medical providers interact. They’re now routinely used in hospitals, outpatient clinics and doctors’ offices statewide.
Medical professionals say AI is enabling them to provide better care by doing their jobs more efficiently — from performing routine administrative tasks, such as medical coding and billing, to detecting breast tumors to predicting which hospital patients will be readmitted within a month of discharge.
Such virtual assistance comes at a critical time in medicine. Vermont’s large population of baby boomers, now in their sixties to late seventies and experiencing more severe and complex ailments, is straining a health care system that has already lost a significant portion of its workforce to retirement and burnout, in part due to the pandemic. AI can help reduce the burnout rate by allowing medical professionals to focus on the work they were originally trained to do: treat patients.
“So much of our jobs … has been sitting behind a computer and trying to document and capture a patient encounter. Obviously, that’s not why everyone signed up to do medicine,” said Dr. Daniel Peters, an emergency medicine physician who divides his time between University of Vermont Medical Center in Burlington and Champlain Valley Physicians Hospital in Plattsburgh, N.Y.
Peters, 31, is also a clinical informatics specialist, which means he helps UVM Health implement new technologies, including AI, in its emergency departments. One such innovation is Gleamer BoneView, an AI fracture-detection agent that UVM Health started using several months ago.
“For a sizable number of our practitioners, it’s been really life-changing.”
Justin Stinnett-Donnelly
Gleamer doesn’t replace the work of a radiologist, Peters explained, but offers a preliminary interpretation of X-rays, which provides physicians with insights during their examination, such as where to look for point tenderness. If Gleamer detects a fracture, the physician can apply a splint until a radiologist reviews the X-ray and confirms it. If there’s no indication of a broken bone, the patient can be discharged sooner, freeing up hospital resources for other patients.
A “game changer” for Peters is a new AI scribe tool called Abridge, which records, transcribes and summarizes patient encounters, then enters them into the patients’ medical records. Peters, who might see 20 to 30 patients during a typical shift, estimated that Abridge saves him hours of typing notes into a computer each week. Previously, he did that task throughout the day as time allowed or when his shift ended, sometimes hours after those conversations took place.
“I can spend more time with patients because I don’t have to worry as much about remembering every single piece and having to document it myself,” he said. “It’s been a significant reduction in the cognitive load that I carry day-to-day.”
Abridge does not recommend treatments or therapies, and Peters must still review the notes Abridge generates to ensure there are no errors or AI “hallucinations.” All treatment decisions are made by a medical professional, not an algorithm, Peters assured.
Since Abridge was piloted earlier this year, about half of UVM Health’s emergency physicians and physician assistants have begun using it. (Since July, it’s been available to all primary care doctors in the network.) It’s an optional tool, as some practitioners still prefer to enter their own notes, Peters said. Some are wary of AI, while others are concerned that those recordings and transcripts will be used in ways not originally intended, such as in malpractice suits — or to supplant clinicians altogether.
Last month, Mass General Brigham in Boston launched an AI app that questions patients, reviews their medical records and offers a list of potential diagnoses. As the Boston Globe reported, the chatbot, called Care Connect, is meant to relieve a critical shortage of primary care physicians. But some doctors said the adoption of Care Connect doesn’t address the underlying causes of the physician shortage — namely, inequitable pay and poor working conditions.
Thus far, UVM Health practitioners have mostly reported that Abridge has eased their workload and improved their job satisfaction. An internal study of the technology by UVM Health and Abridge found that Vermont clinicians spent 60 percent less time documenting their work than they did previously.
“For a sizable number of our practitioners, it’s been really life-changing,” said Stinnett-Donnelly, who will appear on a panel about AI in health care, “The AI Will See You Now,” at the Vermont Tech Jam on Saturday, October 25. “I’ve heard colleagues say, ‘I don’t think I’m going to retire in six months anymore. I can do this for a little longer.’ Or, ‘I don’t feel exhausted when I get home from work. My spouse has noticed a difference in me.’ And that’s pretty cool.”
According to Peters, few of his patients have objected to his use of AI during an exam. As he put it, “When patients come to the ER, or our health care system in general, they expect cutting-edge technology.”
Some of his patients’ acceptance of Abridge may be due to their familiarity with its delivery device — a smartphone. Broadly speaking, however, the public tends to be far more skeptical of the long-term benefits of AI than the experts who develop it.
A Pew Research Center report from April found that while 47 percent of AI experts said they are more enthusiastic than concerned about the positive impacts the technology will have on their lives, only 11 percent of American adults expressed such optimism. That said, both groups shared similar views in wanting more regulation and personal control over how AI is used. But that, too, may change as more people encounter AI on a regular basis.
For many Vermonters, their most frequent interaction with the health care system is with a primary care provider. But a simple phone call to schedule an appointment or refill a prescription can often become a frustrating and time-consuming process.
“I can order a pizza online and see the exact moment they’re placing the pepperoni on the top and the exact minute it goes in the oven,” said Ethan Bechtel, cofounder and CEO of Burlington medical tech company OhMD; he will also appear on the Tech Jam panel. “Yet I call [a doctor] to make an appointment for something that could be a real problem, and I’m listening to flutes play for seven minutes.”
Founded in 2016, OhMD developed an AI agent that streamlines communications between physicians’ offices and their patients. In the process, it’s helping alleviate a national staffing crisis in medical practices, which in recent years have been hemorrhaging employees due to long hours and stressful work environments.
In essence, OhMD automates the most routine calls that come into medical practices — appointment requests, prescription refills and billing questions — by converting those calls into text messages and routing them to the appropriate staff. OhMD’s virtual assistant will ask the caller questions such as “What day works best for you?” “Which doctor do you want to be seen by?” and “Which location do you prefer?” Patients always have the option to speak to a human, and all communications are confidential. The system then responds with a text message confirming the appointment or indicating that the prescription was ordered.
OhMD is now used by more than 1,200 physician practices nationwide, including Northeastern Reproductive Medicine in Colchester. In 2019, when the clinic’s staff were spending hours each day listening and responding to dozens of voicemails, it adopted OhMD. Since “automating the easy stuff,” Bechtel said, Northeastern Reproductive has grown the practice by 25 percent, while reducing staff workloads and improving overall patient satisfaction.
Much of the promise of AI in health care is still in the developmental phase. At UVM, researchers are working on future AI agents that will create new vaccines and drug therapies and better detect eye diseases.
Dr. Gary An, a general surgeon, trauma and critical care specialist, as well as a professor at the Larner College of Medicine, has been working for years on developing a treatment for one of medicine’s most confounding diseases: sepsis. It’s a goal that even pushes the boundaries of AI’s current capabilities.
Sepsis is a complex, dynamic and often fatal condition in which the body overreacts to an infection, causing inflammation and tissue and organ damage. In intensive care units, it consumes enormous amounts of time and resources. Each year, sepsis is responsible for one in five deaths worldwide — approximately 11 million people — according to the World Health Organization.
Currently, there is no cure for sepsis. As Dr. An explained, clinicians can give patients a cocktail of antibiotics, control their infection with surgery or a drain, and support their breathing, blood pressure and kidney functions. But because the disease is so biologically complex, and each patient’s immune response varies dramatically over time, a cure is elusive.
“As a clinician, it’s frustrating because you try and support them, but you know that you’re not actually making them better,” An said. “You’re just essentially doing damage control.”
As a researcher, An pointed out the limitations of AI in treating sepsis. Unlike, say, teaching AI to create videos by drawing from millions of existing digital images, medical and biological data are comparatively sparse, incomplete and not “granular” enough. This makes modeling human physiological responses particularly difficult, he said.
To overcome that data scarcity, An’s AI generates synthetic biological data, or “digital twins,” to simulate the progression of sepsis and how therapeutic interventions will effect the body. Unlike real-world medical trials, simulations allow researchers to run very detailed “what if” scenarios, including those that are counterintuitive, to compare treatment outcomes.
Ultimately, An’s goal is to develop AI that will tailor a treatment for sepsis to each individual patient. While such cutting-edge technology sounds “almost sci-fi,” An added, he estimated that its implementation could happen in as little as 10 years.
With AI’s computational power doubling every six months, it could be even sooner.
Hear Justin Stinnett-Donnelly, Ethan Bechtel and Juniper Lovato in a panel discussion, “The AI Will See You Now,” at the Vermont Tech Jam on Saturday, October 25, 1 p.m., at Hula in Burlington. To attend the free event, register at techjamvt.com.
The original print version of this article was headlined “Healing Algorithms | From scheduling appointments to detecting fractures and strokes, AI is revolutionizing health care in Vermont”
This article appears in The Tech Issue 2025.
Got something to say?
Send a letter to the editor and we'll publish your feedback in print!