artificial inteligence
Europe approves an LLM-based clinical device
STAT's Brittany Trang writes: Earlier this year, a generative AI clinical decision support tool was approved as a class IIb medical device under Europe's CE marking. Called Prof. Valmed (short for "validated medical information"), the tool uses GPT-4o to generate medical answers from a restricted database of clinical guidelines, regulatory documents, quality handbooks, patient handbooks, Cochrane reviews, and publicly available PubMed documents.
The testing for the regulatory approval included 900 questions across different specialties, with the first endpoint for a "safety" index and the secondary endpoint for a response quality index, or how satisfied users were with the answer. The tool is approved for use by medical workers such as nurses and as a medical education tool (Prof. Valmed CEO Vera Rödel mentioned that the company was sponsored by a pharma company who planned to use the tool for its employees' medical education), but is also supposed to support clinical diagnosis and treatment.
Rödel and her co-founder, Heinz Wiendl, a doctor, appeared on a webinar this morning to talk about the tool. Questions about how the team tested for hallucinations, controlled for automation bias, and how it compares to tools like OpenEvidence were not answered. And just last fall, the FDA took a very conservative approach to the idea of approving generative AI devices, with an advisory committee cautioning that the technology is very different than traditional medical devices and would perhaps require totally different approaches to regulation.
No comments