Regulatory Reforms Needed for AI Devices: Shuren
CDRH director Jeff Shuren acknowledged that new regulatory reforms from Congress will be needed for FDA and the medical device industry to experience transformative outcomes from the quickly developing artificial intelligence (AI) and software-driven medical device space. Speaking at an American Enterprise Institute panel on the future of AI, Shuren said his Center would welcome from Congress flexibilities similar to those offered under the emergency use authorization regulations that allowed the Center to efficiently get products on the market during the Covid-19 pandemic.
Shuren reiterated his preferred “voluntary alternative pathway” as the most viable for fitting AI into a regulatory pathway, which he discussed last month at an Alliance for Stronger FDA Webinar. He emphasized the importance of not changing the regulatory standard to market in the U.S. — reasonable assurance of safety and effectiveness should remain. AI developers could use this alternate pathway, and they won't lose the current options (510(k) or PMA) if they choose to switch back because they are more comfortable with the older pathways.
Shuren said during the Alliance Webinar that the new pathway would provide flexibility to fit the technology. “During Covid, what we could do with EUAs was to take this one authority and craft it much better to the technology itself … to allow us to do some things we couldn’t do before.” For example, he said that under the alternative approach, there could be cases where the capability of the manufacturer could be used in lieu of some of the requirements that today are looked at on a product-by-product basis. Shuren called this the “technology certification,” and once a firm is certified, they do not need to come back to FDA with tests within that particular category.
Additionally, Shuren said such an alternative pathway would look at mechanisms for leveraging better post-market data. There are things about quality systems that could be leveraged that could not be done with today’s framework, he added. “So essentially, it would provide general authority so we can build the pathways for the technology, which would be put out there in guidance, and then if you’re in that category, you could choose to go down this path, or stick with the traditional pathway.”
Shuren told the AEI audience that FDA should have the flexibility to design the pathway based upon the technology tailored to novel approaches and a better ability to leverage some of the post-market data and the make that a condition for staying on the market. He said it is absolutely important to have “good monitoring” in place, and that is also a needed safety feature.
In terms of current actions to help AI developers, Shuren told AEI that developing good machine learning practices or consensus standards will be important. “We've been working with other regulators in the International Medical Device Regulators Forum on AI to develop a report on basic terminology, and there’s a report that’s been issued out of there. And we are working with the UK and Canada on principles of good machine learning practices. Another “big piece” to help AI advance is “predetermined change control points,” Shuren said. This allows device makers to pre-identify changes they want to make, and outline a plan for how those will be verified and validated. As long as the sponsor sticks with the plan, it can make those changes without having to submit a new application, Shuren added. “So that’s a little bit more flexibility, and that’s within the authorities we can do today and how far we can kind of push the envelope.”
Looking to the future, Shuren said there are big infrastructure needs. “You need large well-curated data sets that are representative of the intended population that the technology is directed at,” he said. “You need that available so developers can design and evaluate responsibly and that is something that you’re going to need a government and community approach to deal with. There’s also a need for test pits to be able to do this in fairly safe and secure spaces… And there’s a need for standardized tools, particularly around evaluation and design, and for consensus-based standards."