BOSTON – Dr. Rebecca Mishlis remembers her mother, also a doctor, bringing patient charts home every night and working on them long after she went to bed.
Michelis, a primary care physician at Brigham and Women's Hospital, has repeated this ritual herself for years.
But no more.
Since last summer, she has been piloting two competing software applications that use large-scale language models and generative artificial intelligence to listen, transcribe, and summarize conversations with patients. At the end of a patient visit, it takes only a few minutes to check the summary for accuracy, cut and paste a few items into the patient's health record, and click save.
“I'm seeing patients right now (during consultations),” said Michelis, who is overseeing a pilot project across 450 Harvard-affiliated health care providers, with plans to expand to 800 within the next month. Told. “This is a technique that puts me back in the room with the patient, rather than putting a barrier between me and the patient.”
Mishlis, chief medical information officer and vice president of digital at Massachusetts General Brigham, was one of the early adopters of artificial intelligence in the medical field, which is notoriously slow to adapt to change. (“Yes, there is a fax machine in front of my clinic,” she said.)
While some other doctors are incorporating AI and large-scale language models, such as ChatGPT, which analyzes a range of online languages, into their practices, Dr. Mishlis and his team 200 miles away at NYU Langone Health. , one of the few physicians who has chosen to study the use of AI.
They want to ensure that the technology improves overall care before adopting it more widely.
“We're not in a hurry to get this out. We're really trying to tread a cautious path,” said Devin, director of digital innovation strategy in the Department of Information Technology at New York University Langone Medical Center.・Dr. Mann said. “We want to understand how these tools actually work before we release them.”
The much maligned electronic medical record
No one wants to make a mistake when using this technology that will cause them to lose the trust of patients or doctors.
After all, digital technology has so far disappointed both.
Electronic health records have become an essential tool in healthcare, replacing rooms full of paper documents that were difficult to manage and risked fire and other losses.
But patients hated the move to electronic medical records.
Rather than building a relationship with their doctor, making eye contact or listening to the murmurings of someone paying their undivided attention, they listen to the clicking of their fingers and listen to their caregiver's It felt like someone was talking in the back of my head.
Doctors hated them even more.
Dr. Kristin Shinskey, vice president of job satisfaction for the American Medical Association, calls the transition to electronic medical records a “major job shift.” Physicians, rather than nurses, medical assistants, or administrative staff, were suddenly responsible for recording most of the patient's data during office visits.
In a 2016 study, Shinsky et al. showed that after a “massive job change,” doctors spent two out of every hour they spent with patients at a desk.
“(Electronic medical record) time, especially physician order entry time, is a source of burden and burnout for physicians,” she said.
Burnout hurts everyone
Shinskey, who worked as a general internist in Iowa for 32 years, said burnout can lead to medical errors, increase the risk of medical errors, reduce patient satisfaction, damage an organization's reputation and reduce patient loyalty. It is said to reduce
She calculated the cost of doctors leaving the profession due to burnout to be between $800,000 and $1.4 million per doctor. Lost funds include recruitment costs, sign-on bonuses, and onboarding costs.
A recent AMA survey of doctors, nurses, and other health care workers found that nearly 63% reported symptoms of burnout at the end of 2021, up from 38% in 2020.
Inbox work also contributes to burnout, Shinskey said.
In March 2020, when the pandemic began, inbox workload increased by 57% and “has remained high ever since,” Shinskey said. Meanwhile, the remaining workload hasn't fallen enough to compensate for the increase, leading doctors to work more on holidays, she said.
The amount of time physicians spend on their personal time (commonly referred to as “off-the-job work” or “pajama time”) is often a good predictor of burnout. Physicians in the top quartile of pajama hours are far more likely to experience burnout than those in the bottom quartile.
One of the other new requirements adding to the burnout syndrome is the expectation that doctors will “text during the visit,” meaning they will be typing throughout the visit. The experience is deeply unsatisfying for both patients and doctors alike, Shinskey said.
Taking notes means synthesizing
Still, she is not convinced that generative AI and large-scale language models are the only or best solution to all these problems.
In previous practices, Shinskey said it worked well for nurses to share information with doctors in the same room, pull additional information from the electronic medical record and enter instructions in real time. That way, the doctor can focus on the patient and the nurse is familiar enough with the patient's care to answer most of the follow-up questions that may arise during the visit.
“When you integrate care, when you build systems that integrate care, when you prioritize relationships between people, between doctors and patients, between doctors and staff, that's when the magic happens. That's when quality… It’s better and it costs less,” she said. she said. “I see her AI as a technology solution to a technology problem, but the balance of risks and benefits is yet to be determined.”
Shinskey said she worries something will be lost if doctors no longer fully dictate or write their own notes.
As anyone who writes regularly knows, it's in the act of writing that you begin to truly understand your subject matter, she said. Without that connection, the requirements for synthesizing the materials, Sinsky worries that doctors will miss clues about a patient's health.
“It remains to be seen how much (AI) will help us and how much it will distract us,” she said. “I worry that some physicians will simply accept the output of AI without pausing and reflecting to enhance their understanding.”
Offer a hug or other sign of commitment
Still, early reactions to Harvard University and New York University Langone's AI note-taking technology have been positive.
“Some people say it's okay, but it's probably not for them,” Michelis said, but most are more enthusiastic. Many people report that their paperwork has changed dramatically, and in some cases, for the first time, they are able to leave the clinic without having to fill out paperwork. “Some people even offered to give me a hug.”
Mishlis' research also measured how much time doctors spent after office hours reviewing consultation notes in electronic medical records, and how much they changed the notes created by AI. If the doctor makes significant changes, it suggests that the doctor is dissatisfied with the draft.
Each physician participating in the study will complete a survey after using either of the two technologies for two weeks, eight weeks later, and again three months later. At this point, participants are just about to hit the eight-week mark, so data on strain and burnout will be released soon, Michlis said.
She hopes studies like hers will determine for whom the technology may or may not be useful. “This technology may not be suitable for oncologists yet, it may not be suitable for every visit, but that's where we're trying to determine,” she said.
At New York University Langone, where AI is being tested on a small scale, early results show that the technology was able to translate visit notes that doctors typically write at a 12th-grade level or higher to a 6th-grade level. I am. It also makes sense to patients, said Dr. Jonah Feldman, medical director of clinical innovation and informatics in Langone Medical Center's information technology division.
When doctors wrote notes, only 13% broke the content into simple chunks, but 87% of Chat-GPT4 notes were written in easy-to-understand pieces, he said.
Feldman said the purpose of using AI is not to put someone out of a job (typically the biggest fear workers have about artificial intelligence), but to get more done in a limited amount of time. said.
He said this is expected to allow doctors to spend more quality time with patients, improving interactions and care and reducing burnout. “We are focused on increasing physician efficiency and improving the patient room experience,” Feldman said.
Mann, who oversees digital innovation at New York University Langone, found that notes written by AI were difficult to read and that clinicians spent more time rewriting notes than writing them in the first place, resulting in “double work.” He said he wanted to avoid wasting money. . For this to work, he said, “it needs to get better and it needs to get easier.”
The Langone team is also experimenting with using AI to respond to patient emails. Mann said providers want their emails to be personal, so doctors who used to send “haiku” to their patients won't suddenly start sending “sonnets.”
Next, the team wants to expand into home monitoring. This would allow, for example, someone instructed to measure their blood pressure at home every day and upload that information to their doctor to have their questions answered through AI, rather than having it “follow us”. I am thinking of doing so. I put a phone tag on it and kept it,” Mann said. “Many simple answers can be done faster, freeing up limited time and energy to do more complex tasks.”
We also focus on being the first to provide these types of services, as people with limited resources are often the last to receive technological advances.
Ultimately, the success of this type of technology will depend on physicians' willingness to adopt it and patients' comfort with it.
Rachel Albrecht, a recent Michelis patient, had no problem with the AI eavesdropping on her doctor's appointment.
“I think this is a good tool,” Albrecht, 30, an accountant from Boston, said at the end of his appointment. She liked the idea of having an easy-to-understand overview of her results after her visit. She said, “I'm generally an AI advocate.”
Karen Weintraub can be reached at kweintraub@usatoday.com.
