Admin_Adham
It doesn’t wear a white coat, never needs a break, and may help with your diagnosis in the emergency department (ED). AI will see your patient now.
After Harvard Medical School released a new study that found that AI “matched or exceeded” the diagnostic accuracy of attending physicians in the ED, it would be easy to conclude that AI should play a more pivotal role in the ED.
About Medscape Insights
Medscape continually surveys physicians and other medical professionals about key practice challenges and current issues, creating high-impact analyses. For example, the Medscape Physicians and AI Report 2025 found that
- 45% of physicians surveyed were enthusiastic about AI.
- 62% of physicians reported being somewhat knowledgeable about AI.
- 30% of doctors were concerned about AI-driven diagnosis and treatment.
However, before extrapolating that this research means that large language models (LLMs) should take over the critically important work of ED physicians, it’s important to note that in the study, the AI model was given only patient information available via electronic health records before being asked to generate a diagnosis and recommendation for next steps.
This is an important detail because a patient’s health record is just one piece of the puzzle emergency room (ER) doctors consider when assessing patients, especially during times of extreme time pressure, overcrowding, and hectic shifts.
“We have patients coming in with fragmented charts [and] limited information, and even though we are good at very efficient structured history-taking, it’s hard to get pieces of information from patients, especially when someone is in acute distress,” said Tehreem Rehman, MD, associate medical director at The Mount Sinai Hospital Emergency Department in New York City and a spokesperson for the American College of Emergency Physicians.
Rehman is all for continued study of the use of AI in the ED but urges clinicians to exercise caution and not treat it as a one-size-fits-all solution.
“What we do is different than what a doctor does once a patient is admitted,” she said. “In the ED, I’m focusing on not missing something life-threatening, not necessarily identifying a diagnosis. To put it crudely, I’m looking for something that could kill a patient — soon.”
That said, there is a definite place for AI in the ED, said Karen Yeager, MD, an ER doctor at Phoenix Children’s Hospital in Phoenix.
“AI integration is great for workflow management in the ED rather than decision management,” Yeager said. “It can reduce the administrative and cognitive burden on clinicians, but it’s not a replacement for all the important decisions ED doctors are making quickly.”
Rehman argues that AI might be best seen as comparable to the role first-year medical students play in the ED.
“I see AI as something that can assist with structured history-taking,” she said. “That takes a lot of time, and for ED physicians who are running around, unable to dive deeper and working with a 5- to 10-minute intake, I could see that this would be helpful.”
AI tools can also give clinicians more time to spend with patients.
“That is a big challenge in ED settings, where we’re volume centered unlike other appointment-based clinical settings where you have 20 patients a day,” Yeager added. “We don’t control who comes through the door, but we have a responsibility to each patient, whether it’s 100 or 300. Having AI assistance can give us time to build rapport with the family, answer questions, work with the team and make sure everything is done in a standardized way.”
In the end, the goal of spending time with ED patients is to best assess their health concerns and keep them paramount.
“What I’m afraid of is that AI isn’t the whole picture,” said Stephan A. Mayer, MD, director of Neurocritical Care and Emergency Neurology Services for the Westchester Medical Center Health Network in Valhalla, New York. “It’s a stripped-down kind of a tool that performs one basic essential function and isn’t open to nuance.”
And in the ED, AI tools should always be used in a controlled setting, Rehman added.
“The clinician has to be in the loop because there is so much danger if you anchor a diagnosis on [an LLM],” she said. “You could miss out — after all how many really bad life-threatening things might you miss based on a patient’s risk profile?”
And no matter how much research is being done, these LLMs will never take the place of personalized care.
“I worry that the next generation of providers aren’t honing their skills in terms of listening to patients, taking a history, examining a patient, and looking for outside-of-the-box issues that may be happening,” Mayer said. “An AI tool isn’t open to the art of medicine and the fine points of making a diagnosis.”
At least not yet.
Sources reported no financial disclosures.
Lambeth Hochwald is a New York City-based journalist who covers health, relationships, trends, and issues of importance to women. She’s also a longtime professor at New York University’s Arthur L. Carter Journalism Institute.
