The field of artificial intelligence (AI) is stretching the boundaries of conventional med-tech regulation, and several regulatory agencies are working to cut that Gordian knot. Marc Lamoureaux, director of digital health at Health Canada’s (HC) medical device directorate, said on a Sept. 21 webinar that legislation passed in 2019 gives the agency a “regulatory sandbox” in which to experiment with AI regulation, a mechanism he said may bring these algorithms to market much more rapidly than would otherwise be the case.
The FDA was one of the first regulatory entities that took up the AI question, publishing a discussion paper intended to draw feedback. Bakul Patel, director for digital health policy at the FDA’s Center for Devices and Radiological Health, said on the Sept. 21 webinar sponsored by the Advanced Medical Technology Association that the feedback was more extensive than anticipated, and included responses from some parties that ordinarily have little to say about FDA regulation.
However, HC’s search for regulatory flexibility is not the only such conceptualization. Jeff Shuren, director of the FDA’s Center for Devices and Radiological Health, gave voice to a similar sentiment last year. Shuren said the notion of “regulatory Legos” might offer the agency the required flexibility to meet the demands of emerging and yet-to-emerge technologies that are not adequately addressed by the existing regulation. Thanks to the difficulties posed by AI and software as a medical device, “we have to rethink the whole thing,” Shuren said at the time.
No ‘tidal wave’ of AI filings at Health Canada ... yet
Lamoureaux, who co-chairs a committee at the International Medical Device Regulators Forum (IMDRF) that is looking into AI regulation, is involved in some related work at the International Electrotechnical Commission as well. “There isn’t really a tidal wave of medical device applications” for AI despite rumors to that effect, he said, although there has been a discernible uptick in such filings in recent months. Most of those applications are tied to medical imaging, and Lamoureaux said the agency expects the volume of such filings to continue to escalate.
Despite the novelty of AI and machine learning, the agency is well prepared to deal with the technical questions surrounding AI, Lamoureaux said. HC has an advisory committee to assist on these and other issues, and while the newly formed AI working group at IMDRF has lost some momentum due to the COVID-19 pandemic, he said a guidance for locked algorithms is in process “and will be coming shortly.”
However, HC has not granted marketing authorization for a legitimately adaptive algorithm to date, and Lamoureaux said HC sees the FDA pre-certification mechanism for software as a medical device as instructive. The Canadian statute requires that significant device modifications must be taken up as a new regulatory filing, something he said is a barrier for innovation in this space. This is a barrier for regulators as well, Lamoureaux said, because “it inundates us with decisions” about how to regulate these algorithms.
More importantly, this is also a barrier for patients, Lamoureaux continued, adding that an economic plan for Canada published last year includes some details regarding modernization of the HC regulatory framework. This consist in part of “regulatory sandboxes,” which are “a controlled safe space” for experimentation with novel products without immediately invoking the standard regulatory apparatus. In the underlying legislation, which was passed last year, “AI and medical products were specifically called out,” he said.
In June 2019, the legislation called for advanced therapeutic products (ATPs) to be dealt with in the regulatory sandbox. This new regulatory mechanism can include novel products and classes of products, and Lamoureaux said that as a result, “we can implement these relatively flexible and agile models and iterate them very quickly.” This would allow HC to tweak its regulatory framework to fit the needs of a technology without requiring yet another statutory authorization from the Parliament of Canada.
Once a novel regulatory mechanism for a category of products is fully developed, HC will move it back into the standard regulatory framework, but the agency is also developing a concierge service in relation to this new pathway. This will provide a single point of contact for developers of these products, such as tailored advice for regulatory requirements and for passage through any challenges that await at other agencies involved in the device’s path to market.
“We definitely have the lens to harmonize with other jurisdictions,” Lamoureaux said, adding that the flexibility “can allow us to pivot, depending on where the international focus is.”
Patel: FDA cannot regulate AI via single guidance
Patel said the discussion paper reinforces the total product life cycle (TPLC) approach, adding that the response to the paper was encouraging. He said the questions revolving around AI regulation cannot be adequately addressed via a single guidance, but noted that standards, including international standards, will play a significant role in the FDA approach.
At present, the FDA is looking at two or three concepts in the discussion paper that lend themselves to action in the short term. Included in this is how to regulate the adaptation of an algorithm over time, but another point of focus in the near term is the conditions under which a developer must return to the FDA for a follow-on regulatory filing.
Another area that requires more exploration is the measure of an algorithm’s real-world performance. The FDA will obtain data from the sponsor, but there is a question of whether or how much of this information will be disclosed to the public and to other regulatory entities. The patient element of this dynamic is the subject of a patient engagement advisory hearing scheduled for Oct. 22, 2020, Patel said.