heidelberg: “Could you please describe how long this cough has been going on and whether there are any accompanying symptoms such as fever, shortness of breath, or chest pain?” What sounds like a classic question from a doctor was asked in this case by an artificial intelligence (AI) system.
For example, AI has long been able to detect tumors in X-rays. However, development has progressed to the point where it is now possible to communicate with patients.
Researchers primarily from Germany’s Heidelberg University Hospital and the Else Kroner Fresenius Center for Digital Health at the Dresden University of Technology, in collaboration with a team at Google, have developed two independent AI models for patient management, covering everything from diagnosis to treatment recommendations. Both are currently published in scientific journals nature.
The MIRA (Medical Intelligence for Reasoning and Action) and AMIE (Articulate Medical Intelligence Explorer) models are far more advanced than most existing AI tools in healthcare.
Unlike many AI applications in the medical field, which are limited to specific tasks such as detecting tumors with X-rays, both models independently collect a patient’s medical history through chat, order diagnostic tests such as microbiological tests, and create treatment plans, including prescriptions for specific drugs.
Factors such as clinical guidelines, professional literature, and potential interactions between medications the patient is taking are also considered.
According to the study authors, Google’s system provided very detailed and practical instructions rather than vague recommendations. In tests, both models proved to be better and more accurate than human doctors in some cases, the authors said.
Our team points to issues such as staff shortages and a lack of continuity during treatment across multiple appointments. Experts in Heidelberg and Dresden describe the model as a “co-pilot” for doctors, giving them more time to care for patients.
Regarding a wide range of tasks in clinical patient management, Nature says, “If AI agents can perform such tasks and provide effective management reasoning, they could assist physicians in their daily tasks and potentially address physician shortages in some parts of the world.”
Not ready for real world use
Neither group says their models are ready for real-world use. Both tested their applications using AI-generated patient simulations based on real data, but these simulations had pitfalls. A virtual patient reacts and reacts very differently than a real person arriving at the hospital with an acute condition.
Researchers said MIRA generally recommended appropriate, evidence-based treatments. However, the recommendations were not 100% reliable.
“These results demonstrate the potential that AI agents can bring to medicine,” said Uwe Platzbecker, medical director of Dresden University Hospital. He said a key question for future developments is “how can such innovations be safely and transparently integrated into clinical practice for the benefit of patients?”
Kerstin Denecke, a researcher at the Bern University of Applied Sciences, identified current obstacles, including the current state of data in the medical field, regulatory approval processes, clear accountability, and the need for representative studies on the risks of such systems.
A patient-centered digital health expert said, “Clinical decisions require more than just adhering to guidelines.” What is needed, she said, is understanding each patient’s individual situation.
Robert Ranisch, professor of medical ethics at Germany’s Potsdam University, called the MIRA study “an exciting and methodologically well-designed contribution.”
There are many possibilities, but there are also concerns.
However, he said the study looked at the performance of the AI agent under laboratory conditions. As with many AI studies, a key question was whether the technology would work in everyday clinical settings. “This is exactly where many promising AI systems fall short so far,” he said. With real patients, clinicians, incomplete data and various IT systems working, they quickly reached their limits, he added.
Reinhard Busse, head of healthcare management at the Technical University of Berlin, said the ability of AI agents to map clinical processes and diagnostic and treatment decisions in a structured way does not in itself mean they will deliver better care or lower costs.
Meanwhile, German medical professionals have also expressed concerns that the human touch will be lost when AI-based assistance systems are introduced.
Researchers need to investigate whether interpersonal and emotional aspects are being relegated to the background, for example by supporting human communication or replacing them with technological voice assistance systems such as chatbots, the German Federal Medical Association has previously said.
Eugen Blisch, director of the German Patient Protection Foundation, also emphasized that despite the time savings brought about by AI, “the conversation between doctor and patient will continue to play an essential role for patients.” This is especially important for older people, he said.
At the same time, he warned against increasing dependence on non-European companies. “The future potential of analytics and communications cannot depend on tech billionaires and their political entanglements.” – dpa
