The Food and Drug Administration on Tuesday released a five-pronged plan to underpin its approach to regulating medical software with artificial intelligence or machine learning components, including plans to issue new guidance for how developers modify such software.
The document represents the first so-called “action plan” that the FDA has released on the topic. Agency officials plan to update its approach as technology and research advance.
“To stay current and address patient safety and improve access to these promising technologies, we anticipate that this action plan will continue to evolve over time,” said Bakul Patel, director of the Digital Health Center of Excellence in the FDA’s Center for Devices and Radiological Health, in a statement.
The FDA launched the center of excellence in September 2020 to coordinate efforts to revamp how the agency regulates digital health technologies.
The agency in its action plan solidified plans to issue a draft guidance document that would ask developers to submit a “predetermined change control plan” with their premarket submissions for AI- and machine learning-based medical devices. The FDA first suggested including such information in premarket submissions in a discussion paper the agency released and solicited public feedback on in 2019.
The agency plans to release the draft guidance later this year.
Developers submitting predetermined change control plans would outline what types of modifications or upgrades they intend to make to the software once it’s already in use, as well as how they plan to deploy those updates while managing risks to patients.
In 2019, when the FDA released the discussion paper on AI in medical devices, then-FDA Commissioner Scott Gottlieb highlighted AI’s need for regular updates as a challenge for regulating the technology.
The agency’s typical approach to regulating software in medical devices requires any changes or upgrades to software to receive separate FDA approval. That doesn’t work for AI products, Gottlieb said, since part of the supposed benefit of AI is that it continually adapts in response to new data.
Outlining types of modifications that a product plans to incorporate would help the FDA review and anticipate such changes.
The FDA in its action plan outlines a handful of other steps it plans to take as part of its approach to regulating AI and machine learning, which span premarket to postmarket oversight—like developing methods to evaluate and improve algorithms, encouraging real-world pilots that monitor outcomes, and promoting patient-centered practices, such as transparency into how AI devices work.
FDA officials said the agency will continue to seek feedback on its approach to regulating devices that use AI and machine learning.