Experts advise against prescriptive FDA policies for regulating AI

BALTIMORE – Experts said that any future framework for regulating artificial intelligence (AI) by the US Food and Drug Administration (FDA) should avoid being overly prescriptive. They also urged greater collaboration between industry partners for developing best practices for using AI.

Pharmaceutical and medical device industries need to be very specific in what they hope to achieve with AI, said experts at the RAPS Regulatory Intelligence Conference on 7 June.

The panel was moderated by Chris Whalley, director, regulatory intelligence at Pfizer, and featured commentary from Bradley Thompson, RAC, who leads Epstein, Becker & Green’s AI practice, Sam Kay, vice president for pharmaceuticals at Basil Systems, Gopal Abbineni, director of regulatory strategy for medical devices at Bayer, Elizabeth Rosenkrands Lange, senior director and head of US global regulatory for EMD Serono/Merck.

Whalley asked the panelists to describe how they use AI tools at their companies.

Abbineni responded saying that Bayer is using AI both as part of the medical device software and as part of their efforts in the regulatory intelligence, including developing its own generative AI tool called MyGenAssist.

Rosenkrands Lange added that Merck has built its own AI tools. “We are also doing a lot of pilots and we are also using our own tools. Our conclusion is that there are limitations, and we need to cooperate with vendors.”

When asked whether he’s seen any interesting use cases for AI, Thompson said it is difficult to pick one individual case, but from his own experience, he said that "there is so much text data in FDA 510(k) summaries that are in PDF form but there is also a lot of data in the postmarketing area. There are over a million MDRs filed in a given year and most of them are in a text field, and ther is information in all those text field that can be mined."

The panelists also discussed whether they are optimistic or pessimistic about current potential of AI tools.

Kay said he is a pessimist. “I do not think that it hits the threshold of carrying out the activities in a thorough way.” He said there are incremental benefits to be gained by using AI. One learning however, is that companies should not invest huge sums of money in AI without a specific purpose. When they do so, he said that “I see them failing in a month or two.”

Thompson said that AI can help companies reduce the amount of time it takes to develop new products; however, he cautioned that using AI “is a first node. There’s need for more layers.”

Whalley also asked the panelists what is missing from the policy conversations with FDA. The agency is expected to release a new guidance on AI in drug development later this year.

Abbineni said that one area that should be discussed is whether using AI can reduce the need for additional animal studies. “Are we asking this question of FDA? Are we really handing focus on the low-hanging fruit?” he asked.

Rosenkrands Lange observed that "what is missing,,,,is more case studies shared among industry, that is what I'd like to see."

For his part, Kay urged regulators not to adopt a “prescriptive” AI regulatory framework for pharmaceuticals. He said that “prescriptive is restrictive, you do not want a tight framework.” He added that “a continuing dialogue and a loose framework of principles that are open for interpretation” is what’s needed.

RAPS Regulatory Intelligence Conference

Experts advise against prescriptive FDA policies for regulating AI

Related topics

Recommended content

Welcome to the new RAPS Digital Experience