the Food and Drug Administration met with OpenAI to discuss AI’s use by the agency. The meetings seem to be part of an effort by the FDA to use AI to speed up the approval process for drugs.
Why does it take a new drug over 10 years to reach the market? FDA commissioner Marty Makary on Wednesday . “Why aren’t we modernized with AI and others things?” We’ve just finished our first AI-assisted product review and that’s only the beginning.
Makary made the remarks following an annual meeting of American Hospital Association earlier in the week, during which he spoke about AI’s ability to help with the approval of new treatment for diabetes and certain cancer types. Makary did mention OpenAI in this initiative, but he did not specify it. Sources close to the project claim that a small team of OpenAI representatives met with the FDA, two associates of Elon Muskās Department of Government Efficiency on multiple occasions in recent weeks. The group has discussed the project cderGPT which is likely to stand for Center for Drug Evaluationregulates prescription and over-the-counter drugs in the US. Research GPT is also involved. Discussions were led by Jeremy Walsh, the FDA’s very first AI officer. No contract has yet been signed.
OpenAI refused to comment.
Walsh also met with Peter Bowman Davis, an undergraduate student on leave from Yale, who currently serves as acting chief AI officer for the Department of Health and Human Services to discuss the FDA AI ambitions. Politico was the first to report Bowman-Davis’s appointment. Andreessen Horowitz, part American Dynamism Team
Robert Califf, former FDA commissioner between 2016 and 2017 and again in 2022 until January, told me via email that the agency’s review team has been using AI since several years. “It’ll be interesting to learn which parts of the review have been ‘AI assisted,’ and what this means,” he says. “There has been a desire to reduce review times, and there is a general consensus that AI can help.”
Califf said before he left the FDA that the FDA was looking at the different ways AI could be utilized in internal operations. “Final approval reviews are only a small part of the opportunity,” he says.
It is clear that using AI to help with final drug reviews could compress a small portion of the notoriously lengthy drug-development timeline. The vast majority fail before they are ever reviewed by the FDA.
Rafael Rosengarten is the CEO of Genialis – a precision oncology firm – and a board member of the Alliance for AI in Healthcare. He is in favor of automating some tasks related to drug review, but believes there should be guidelines about what data is used to build AI models and the type of performance that is acceptable. “These machines are extremely adept at learning information. But they have to be taught in a way that they learn what we want them too,” he says.
According to him, AI could be used more quickly to tackle certain “low-hanging fruits” such as ensuring that applications are complete. He says that something as trivial as this could speed up the feedback given to submitters on the things that are needed to complete the application. More sophisticated uses will need to be tested and proven.
A former FDA employee who has used ChatGPT in a clinical setting says that the AI models’ tendency to create convincing information raises doubts about the reliability of a chatbot. “Who knows how robust this platform will be to perform these reviewers’ duties,” the ex employee says.
Currently, the FDA review process takes around a year. However, there are several mechanisms that can be used to speed up this timeline for promising drugs. Fast track is a designation for products that are designed to treat serious conditions and meet unmet medical needs. The FDA also has the breakthrough therapy designation. This was created in 2012 and allows them to give priority review to drug candidate that could provide a substantial advantage to patients over current treatment options.
Andrew Powaleny is a PhRMA spokesperson who says that it’s important to review medicines for safety and efficacy in a timely fashion so they can meet patient needs. “While AI is still in its infancy, harnessing it will require a thoughtful and patient-centered approach.
” The FDA has already begun to research the potential uses of AI. In December 2023, the FDA advertised a fellowship for a researcher to develop large language model for internal use. The fellowship description states that “During participation in the program, the fellow may engage in a variety of activities, including but not limited to applications of LLMs in precision medicine, drug-development and regulatory science.” Openai Announced in January a self hosted version of ChatGPT Gov. This chatbot is designed to comply with government regulation. The startup said it was also working to get FedRAMP moderate or high accreditations for ChatGPT enterprise, which would enable it to handle sensitive data from the government. FedRAMP is the federal government’s compliance program for assessing cloud products. A service cannot store federal data unless it has been authorized by this program.
Additional Reporting by Matt Giles.
