ARTICLE SUMMARY:
Did FDA clear a product that incorporates a large language model? Excerpted from Pathways’ Picks July 1: FDA Clears an LLM? Asia Updates, Ireland’s Elevation, and More.
A December 510(k) clearance of medical management software for people with type 2 diabetes may be the first-ever FDA authorization of a product that incorporates a large language model (LLM), though more details need to be confirmed.
Palo Alto-based digital health firm UpDoc issued a press release June 25 announcing “FDA clearance for the first Software as a Medical Device that uses patient-facing large language models.” The system, the company says, includes an LLM that interacts with patients via voice or text to help them adjust insulin dose and diet in response to changes in glucose levels and symptoms, and it also triggers appropriate follow-up tests, updates the patient’s medical record, and advises medical attention if necessary. Summary documents on FDA’s website for the December 510(k) clearance of the system describe an UpDoc Agent “Conversation Service” along with patient and provider portals, though they don’t specifically mention AI or an LLM.
Multiple digital health regulatory experts (for instance, Yujan Shrestha, MD, and Hugh Harvey, MD) who reviewed the press release, FDA documents, and a prior study from UpDoc founders publicly concluded that the system likely relies on a standard insulin dosing calculator, which patients and clinicians routinely use to adjust doses within a clinician-established protocol. The LLM, they believe, is overlayed to serve as a “conversational shell” to help inform, respond to, and prompt the patient, but not serve any role in informing clinical decisions outside of the outputs from the calculator.
Even that limited use case is a big deal, some experts tell Market Pathways, noting that it appears to be the first LLM of any type to get through FDA and the US agency has been very cautious about the technology due to concerns about AI hallucinations or other unpredictable risks. “Anything LLM-based touching FDA at all is no small feat,” Acacia Parks, PhD, a regulatory consultant specializing in digital health from Avania, said in an interview. “Anybody who's been in the trenches trying to have these conversations [with FDA] knows how much is involved.”
Continue reading Pathways’ Picks here.