ChatGPT rivals doctors in diagnosing emergencies

[New Sancai Compilation and First Release] One day, emergency physicians may rely on consultations from artificial intelligence (AI) programs such as ChatGPT to help them quickly and accurately diagnose patients’ illnesses.

A new study found that ChatGPT performed as well as human doctors at diagnosing patients when given the same clinical information.

"In fact, they are very comparable," says Jeroen A. Nijmegen, Netherlands. Steve. Clinical chemist and senior researcher at Jeroen Bosch Hospital. Steef Kurstjens said. "Because they are so similar, AI may help speed up the process or increase the number of diagnoses in the emergency department."

In this study, a team of two doctors and an AI separately reviewed the results of the study in Jeroen in March 2022. Physician notes and laboratory tests of 30 patients treated in the emergency room of Boss Hospital.

The researchers used both free and subscription versions of ChatGPT. The AI tool and medical team then compiled a list of the top five basic diagnoses for each patient based on available information. Because these were past cases, researchers already knew what the exact diagnosis was.

Doctors' correct diagnosis rate for these five diseases was 87%, compared with 97% for the free version of ChatGPT and 87% for the subscription version of the AI program.

"This is a great proof of concept and a great way to show the potential of this use," Costiens said.

Costiens said that based on these results, follow-up clinical trials are necessary to evaluate the potential real-world impact of adding AI consultations to emergency care.

"Someone needs to really start comparing doctors who use AI in their daily practice with doctors who don't and see if that affects the time patients spend in the emergency room, the correct diagnosis, how long it takes to get treated. Make the right decision diagnosis," Costiens said.

Costiens came up with the idea for the study after hearing doctors discuss the difficulties of diagnosing patients with complex health problems.

"I used ChatGPT to input some general symptoms and medications of the patient, and the AI was still able to give the correct diagnosis." Costiens said that ChatGPT even interpreted the results of the diagnosis as a correlation between the two medications the patient was taking. effect.

Doctors are already supported by many different forms of computer statistical programs and scoring systems for screening lab results, and Costiens says AI can play a similar role.

"I just think this is another potential tool to support physician decision-making and improve and speed up diagnosis," Costiens said.

He added that if an AI program is designed specifically for health care, its ability to assess patients may be greater.

"If you look at medical information today, there's a lot of literature out there. There's more information out there than any doctor can take in. I think this is a tool to learn all of it," he said.

“When we see future medical-focused language models being trained on all these medical articles, I think the future is bright,” Costiens said.

Jessica., president of the Association of Emergency Medicine Resident Physicians. Adkins. Dr. Jessica Adkins Murphy agreed that this showed the potential of AI to improve emergency care.

"It has a lot of potential and possibility to radically streamline a lot of our processes and identify the most obvious few diagnoses that might explain a patient's condition," Murphy said. "I think this study absolutely confirms that. That's it. It creates a good starting point."

However, Murphy said patients often have overlapping health issues and complex medical histories, which can be difficult for AI programs like ChatGPT to analyze.

In addition, patients often experience symptoms that are unlikely to be serious health threats, such as abdominal pain from a heart attack, Murphy said.

"I just don't think this study fully reflects the complexity of our patients," Murphy said. So while I'm excited to see what AI can do, I don't think medicine is about just choosing the five most obvious options for patients. ” (Editor’s note “The five most obvious choices” refers to selecting the five most common or most straightforward options from among multiple possible options based on a set of related conditions or symptoms in a certain research or treatment method)

Both Costiens and Murphy said the best way AI can help in the short term is by assisting doctors with the paperwork burden.

Murphy said: "Currently, the most direct and effective effect is that AI can improve the efficiency of some very boring parts of medical care, such as chart work and note writing." "For example, there are now AI applications that can listen to you talking to patients and simultaneously record you. Take notes and basically complete the patient’s medical history and physical examination.”

These AI applications can also produce basic diagnoses, but Murphy's colleagues told her that they always have to add or remove AI-generated diagnoses from the completed chart annotations.

"I'm very optimistic that it will make us more efficient and allow us to reach more patients because we spend less time on such tasks, but it won't really replace physician judgment on complex cases," he said. Murphy said.

Costiens envisions a future in which AI programs will be integrated into healthcare systems' electronic medical records, allowing them to access doctor notes, lab data, imaging results and patient histories.

"Based on that information, (artificial intelligence) can write a letter or take a note, or suggest what follow-up tests should be done or what diagnosis is most likely," Costiens said.

Dr. Steven Brooks, chief of emergency medicine at Cleveland Clinic Akron General, agreed.

"The use of electronic medical records provides a way for standard information collection. Electronic medical records can be combined with AI to help emergency physicians quickly and efficiently make differential diagnoses and then order appropriate testing," Brooks said.

He added: "As the U.S. population ages, health care services grow, and the demand for more doctors becomes greater, AI has the ability to continue to develop and help doctors improve efficiency and quality of medical services."

However, privacy remains an issue.

Costiens said that the operating environment of AI should be limited to within the health care system and should not rely on cloud platforms.

Another concern is that healthcare workers should use AI programs approved by regulatory agencies such as the U.S. Food and Drug Administration (FDA).

"You should consider that if you use this software to affect patient care, you are using a software program as a medical device and it was not designed to be a medical tool," Costiens said.

Costiens recently reported the findings in the Annals of Emergency Medicine and at the annual meeting of the European Society of Emergency Medicine in Barcelona, Spain.

(Compiled by: Zhong Daren)

(Editor: Jiang Qiming)

(Source of the article: First published by Xinsancai)