Postmarket Responsibilities for Medical AI: Part III — FDA Should Not Expand Its Oversight of Commercialized AI

Towards Data Science

Bradley Merrill Thompson, Member of the Firm in the Health Care & Life Sciences practice, in the firm’s Washington, DC, office, authored “FDA Should Not Expand Its Oversight of Commercialized AI,” part three of the three-part series, “Post-Market Responsibilities for Medical AI” in Towards Data Science.

Following is an excerpt:

In Part I of this series, I outlined FDA’s existing statutory authority to oversee the commercialization of medical devices. That authority is substantial, and has served the agency well for decades. It is agnostic to the particular form of technology, and is readily adaptable to the needs of AI. But FDA wants more when it comes to medical AI. FDA wants to be constantly informed of the performance of medical AI in the marketplace.

That is not only unnecessary; it would be positively detrimental to innovation and the availability of important, potentially life-saving technology. To be clear, I’m not suggesting that we simply trust companies to do the right thing, defined as whatever each company decides in its own mind is the right thing. Instead, I’m suggesting that the principles laid out in Part II be embedded in Good Machine Learning Practice regulations that FDA would promulgate. If those requirements for postmarket vigilance by companies are clear, the vast majority of companies developing medical AI will follow them.

What I’m challenging is the notion that in addition to clarifying the regulatory requirements that medical AI companies are expected to follow as they engage in postmarket vigilance, that we also increase the reporting obligations on those companies. It’s the increase in reporting to which I am objecting. Neither Congress nor FDA should change the law to (1) require that companies share more information postmarket with FDA, (2) allow FDA to oversee day-to-day decision-making in the commercialization phase of the product lifecycle, or (3) give FDA the ability to conduct virtual inspections.

In this, Part III, I will outline these three primary arguments against expanding FDA reporting obligations for commercialized, medical AI. …

  1. FDA Should Not Increase the Reporting Obligations of Companies
  2. FDA Should Not Involve Itself in the Day-To-Day Decision-Making in the Commercialization of New AI Products
  3. FDA Should Not Bypass On-Site Physical Inspections

Related reading:

“Part I – FDA’s Plan to Increase Its Oversight,” by Bradley Merrill Thompson.

“Part II — What Industry Should Do in Commercializing Medical AI,” by Bradley Merrill Thompson.