UKE: AI writes electronic doctor's letters
AI is to help with writing doctors' letters. The University Medical Center Hamburg-Eppendorf is testing a language model trained with its own data.
(Image: Indypendenz/Shutterstock.com)
The University Medical Center Hamburg-Eppendorf (UKE) has gone live with the AI application "ARGO" to support the creation of doctor's letters. With the help of a Large Language Model (LLM), which was trained with data from the UKE, doctors are to be relieved of the burden of documentation at the end of an inpatient stay.
ARGO creates a draft of a doctor's final report based on the patient information collected, in which the reason for admission, course of treatment and decisions are explained. The generated text is then checked, adapted and approved by the treating physicians.
AI model trained on UKE data
According to the UKE, the AI model was developed by the non-profit UKE subsidiary "Innovative Digitale Medizin" (IDM gGmbH). "Since the language model is trained on patient cases from the UKE and the development is linked to high quality assurance standards, ARGO is very precise," explains IDM Managing Director Dr. Nils Schweingruber.
Videos by heise
Following a successful test phase, ARGO is now being gradually introduced in the UKE clinics. According to IDM, the application will also be made available to other clinics and research institutions in Germany in the future.
Challenges of AI in the medical field
Other research institutions are also pursuing similar approaches: The Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) , for example, is developing a "doctor's letter generator", which is due to be launched by the end of 2024. This is also intended to create AI-supported discharge letters.
However, there are also data protection hurdles here. In an interview with heise online, lawyer Philipp MĂĽller-Peltzer pointed out that it must be ensured that patient data is processed in accordance with European standards and not simply used to train algorithms.
Doctors must also check whether important content is missing or contains errors that could lead to misunderstandings or treatment errors. The responsibility always remains with the doctor, as the AI primarily has a supporting function. Incorrect decisions that could be considered treatment errors could have consequences under liability law.
AI should relieve the burden of documentation
Documentation in hospitals presents doctors in hospitals and medical practices with major challenges. This is why Health Minister Karl Lauterbach has repeatedly held out the prospect of AI easing the burden. At present, however, it often fails due to malfunctions in the telematics infrastructure (TI), the healthcare network, and errors in connected hardware and software, which is why calls for a stable infrastructure are growing louder.
Since March, practices have been obliged to have up-to-date and KBV-certified software for the electronic doctor's letter in order to avoid a reduction in the TI flat rate. This is also the case for other applications such as e-prescriptions.
In February, the National Association of Statutory Health Insurance Physicians (KBV) had unsuccessfully requested an extension of the deadline for the e-physician letter from the Federal Ministry of Health. At that time, not all practice management systems were ready for the e-prescription. For this reason, software manufacturers should be more strongly obliged in future to provide suitable products in good time. The BMG sees one step in making Gematik its digital agency once again and giving it more powers, including the possibility of sanctions. The planned launch of the electronic patient file 3.0, the "ePA for all", is expected to be more difficult than the e-prescription.
(mack)
