As patients, we see only one dimension of a doctor’s work. They examine us, sometimes perform a procedure, provide us with a treatment plan, maybe refer us to a colleague. But for doctors, interacting with patients is only part of their job. They also handle insurance claims, come in patient records electronically, consult previous records. And they take notes and lots of notes.
Doctors take notes for everything related to a treatment:
patient history, diagnosis, insurance, second opinions. Notes are essential and take time. In the United States, doctors spend two hours completing administrative tasks for every hour with a patient, according to one study. It is a burden that contributes significantly to the high levels of burnout in the profession.
Now imagine if someone or something otherwise it could be taking all those notes whether it’s a patient’s history, diagnoses, insurance plans, second opinions and then organizing them and making sense of them, freeing doctors from the task.
In fact, it’s already happening, with AI for medical note-taking.
Most Americans they are uncomfortable with the idea of having AI in their doctor’s office, the Pew Research Center found out last year. Well, too bad. Such AI tools can already be found in tens of thousands of doctors’ offices across the United States.
How does AI work in healthcare?
There are many ways AI can be (and is) employed in healthcare. Doctors use it for imaging breast cancer detectionto treat strokes and detect cardiovascular diseaseit’s at assist surgical robots. It can now support doctors even during patient visits, thanks to generative AI.
The premise is simple. During a visit, an artificial intelligence tool listens to everything that happens: patient complaints, doctors’ comments. It then transcribes the exchange, as normal speech recognition or transcription software would do, but it does something more: it transforms the information into a well-organized document ready to be imported into a patient’s electronic medical record (FSE), sent to insurance companies, or shared with other doctors.
This is the kind of work that doctors and nurses hardly have the time to do during normal work hours and often take themselves home, organizing and updating notes and forms while preparing dinner or after the kids are in bed. Eat the little free time they have and their mental health by filling in the burnout and stress commonly encountered in the health professions.
Doctors love taking care of patients and hate what they call pajama time, said Alexandre Lebrun, co-founder of Nabla, a Paris-based AI clinical note-taking software startup. Nabla is integrated with OpenAIs GPT-4, a large language model capable of generating human-sounding prose. A couple of months ago, Nabla launched Copilot, a Chrome browser extension that works with audio input, transcribing a medical examination and turning it into notes.
A turning point
It’s a game changer, said James Schwartz, a Naples County, Florida, jail doctor who uses the Nablas tool. Schwartz has worked as a physician since the 1990s, practicing family medicine in Rhode Island for more than two decades before moving to Florida. For just as long, he wished he could do without taking notes. We all recognize that note-taking is key, but that’s also not what we went to medical school for, he says.
That desire to dispense with cumbersome paper notes turned Schwartz into an early adopter of electronic health records and later transcription software. But he was disappointed with the direction he was taking the EHR, focusing far more on patient data collection than on supporting the work of physicians and easing the burden of administrative tasks.
I wish there was a supercomputer that could listen to the clinical encounter and then synthesize a clinical encounter, Schwartz remembered thinking. Basically, I rubbed a genius lamp 15 or 20 years ago, and it finally granted my wish, he said of the Nablas tool.
In addition to managing notes, the program allows Schwartz to have a better setup for visits so he can give patients his full attention, without looking at a monitor or typing. He wears a lapel mic, he said, moving between prison cells with a mobile desk a few feet behind him.
Then, when I’m done with my laps, I look at all my notes, copy and paste them, and do whatever is required, Schwartz added. To my surprise, I actually thought the generator would have more hallucinations, but it really isn’t.
The kind of support the AI tool provides him would have had a huge impact on his career, Schwartz says. I wasn’t going to take my work home with me in the evening to finish my notes. For one thing, I could have kept going around my patients in the hospital, he said. And overall, it would have been a much less stressful career.
Trust the machine up to a point
Less than three months after its early spring launch, Nabla has just 800 users, though its customer base is growing 10% to 20% a month, according to Lebrun and Delphine Groll, its co-founder. But a host of other companies provide clinical note-taking software, including Nuance, the creator of Dragon Ambient eXperience, and DAX, the market leader, used by more than half a million physicians in the United States.
Unlike the Nablas product, DAX does notor not yetdeliver notes in real time. However, for clinicians, the perceived benefits of using an even slower note-taking system they are hugeIncluded greater retention in a field with severe staff shortages. Everyone should want this, because it reduces the burden on the doctor, allows them to see more patients, reduces paperwork and the wait time to access primary care or urgent care [center] it would go down, in theory, said Marc Succi, a radiologist and chair of innovation and commercialization at Mass General, Brigham Enterprise Hospital.
Succi, who did research on the integration of artificial intelligence in the doctor’s workflow, says the provider still needs to review and sign the notes, because the AI can be inaccurate. I don’t think I would feel comfortable using the transcription software and just not reading what he wrote, he said. You might imagine that he could misinterpret things.
This is where AI may fall short of expectations for the potential to create dramatic workload reductions, said Dev Dash, an AI researcher at Stanford University School of Medicine. Sometimes in clinical practice, it’s much easier to generate a note yourself than relying on an AI tool and having to review the entire note, he said. This becomes all the more true the longer the note is.
In other words, reviewing the AI-generated notes could be burdensome, or even more, than writing the note yourself.
AI for diagnostic support
But with AI’s note-taking capabilities already reaching new levels of complexity and accuracy, it’s becoming easier to conceive of a time when generative AI will be involved in other medical tasks, such as clinical decision support. In fact, the first companies to experiment with some level of diagnostic support are already here.
One is Glass health, a Silicon Valley startup that uses large language models to compile medical knowledge and assist physicians with their diagnoses and clinical plans. Using the software, a healthcare professional can enter a paragraph describing the patient and any symptoms and receive a tentative diagnosis or treatment plan along with references to relevant medical articles and literature.
This does not mean that AI does a doctor’s job. Our artificial intelligence is a doctors assistant. It does not replace doctors in any way. It’s not a substitute for physician judgment, said Dereck Paul, co-founder and CEO of Glass Health, who was inspired to build the support tool after being an internal medicine resident during covid when he experienced the limited resources available to physicians. It’s something you want doctors to think about at the level of an intern…an assistant who can make suggestions and might help you think about something you weren’t thinking about before.
With 40 academic physicians training the model to deliver more accurate diagnoses and treatment plans, and already tens of thousands of users, Glass Health hopes to become the ultimate repository of medical knowledge, ultimately helping physicians diagnose complex problems.
This is the kind of team that maybe 10 years ago would have been working on an online textbook for medicine and 20 years ago would have been writing print textbooks, Paul said. But that team of experts is now training in how to assist doctors with a machine.
And the need for humanity in healthcare?
As the industry grapples with new applications for AI, there are some in the field who fear the technology will replace essential parts of the care process. Among them is Emily Silverman, a doctor at Zuckerberg San Francisco General Hospital and founder of The Nocturneswhich produces podcasts and live events that promote storytelling by healthcare professionals.
For me, writing notes is an important time to think quietly, says Silverman. I work in the hospital, where patients are seriously ill and complex with multiple disease processes. I often arrive at my diagnosis and treatment plan in the process of putting my thoughts on the page. So I’m not sure how an AI tool would fit into that process.
Silverman is fine with outsourcing mindless administrative tasks, but is disturbed by the lack of humanity she finds in the AI-generated clinical notes. I can tell which sections were written by humans and which weren’t.
Human sections have their own voice, Silverman says, which makes reading medical records that much more enjoyable and keeps doctors alert to important information. “I’m concerned that the doctor’s voice will be further homogenized by the AI and my eyes will just stare as I try to read the notes and figure out who the patients are,” he says.
But as AI robots get more sophisticated, that is to say, more human-like concerns may start to disappear, if only to make room for concerns about the next areas of medicine where AI is to intervene.
#Artificial #intelligence #arrived #doctors #office #stakes
Image Source : qz.com