FDA releases framework for reviewing AI-based medical devices

In a first step to develop a set of AI-specific rules and regulations, the FDA released a discussion paper April 3 detailing how it plans to vet and approve AI medical devices without compromising quality or patient safety.

AI-based medical devices are growing in popularity and scope, Commissioner Scott Gottlieb, MD, said in a statement, but it’s difficult for an agency that has been vetting “locked” algorithms for years to develop a framework for reviewing technology that’s constantly adapting. Most AI devices the FDA has authorized to date involve static technology that’s modified by the manufacturer at intervals, rather than tech that continually evolves based on what it’s learning in real-time.

“We are exploring a framework that would allow for modifications to algorithms to be made from real-world learning and adaptation, while still ensuring safety and effectiveness of the software as a medical device is maintained,” Gottlieb said. “A new approach to these technologies would address the need for the algorithms to learn and adapt when used in the real world.”

That approach would be a more tailored fit than the FDA’s existing regulatory paradigm for software as a medical device. For more traditional devices, when modifications are made that might significantly alter the safety or effectiveness of the technology, a sponsor has to make a submission demonstrating the safety and effectiveness of the modifications. Officials can’t take the same approach with AI, since the software is ever-changing.

According to the FDA’s paper, the first step in developing a framework for regulating AI-based devices is to outline information specific to devices the FDA might require for premarket review, including an algorithm’s performance, the manufacturer’s plan for modifications and the ability of the manufacturer to manage and control the risks of those modifications. They might require manufacturers to submit what’s known as a predetermined change control plan—an outline of any anticipated modifications based on an algorithm’s retraining and update strategy, as well as a plan of action for completing those modifications in a safe, controlled manner.

“Consistent with our existing quality systems regulation, the agency expects software developers to have an established quality system that is geared toward developing, delivering and maintaining high-quality products throughout the lifecycle that conforms to the agency’s standards and regulations,” Gottlieb said.

The FDA said its main goal in developing this new framework is to assure that ongoing algorithm changes follow pre-specified performance objectives and change control plans, use a validation process that ensures improvements to the performance and safety of the AI software and include real-world performance monitoring.

“We’re exploring this approach because we believe that it will enable beneficial and innovative artificial intelligence software to come to market while still ensuring the device’s benefits continue to outweigh its risks,” Gottlieb said.

“Artificial intelligence has helped transform industries like finance and manufacturing, and I’m confident that these technologies will have a profound and positive impact on healthcare. I can envision a world where, one day, artificial intelligence can help detect and treat challenging health problems, for example by recognizing the signs of disease well in advance of what we can do today.”

The FDA is asking for feedback from experts and stakeholders in the medical space to flesh out their rudimentary framework. Next steps will include issuing draft guidance informed by that feedback.