A cutting-edge aspect of digital health is software as a medical device (SaMD) that uses artificial intelligence and machine learning to improve its performance based on real world use and experience. Until earlier this week, FDA has said very little about how to handle the challenge of regulating software for diagnosing or treating patient that learns and adapts in the field. That changed when the FDA Commissioner announced the release of a 20-page discussion paper outlining a potential framework for regulation.
The press release itself is relatively long and detailed, but the key elements of the framework are in the discussion paper. It is intended to elicit comments and feedback from interested parties. FDA even includes 18 focus questions, similar to what industry typically uses in presubmission packages for FDA.
Section I of the discussion paper provides discusses traditional medical device regulation and the challenged posed by Artificial Intelligence / Machine Learning‑based software (AI/ML‑based Software). Section II provides additional background about regulation of SaMD generally. Section III provides a typology of the kinds of modifications in the field that can occur with AI/ML‑based Software. Section IV is key, outlining a Total Product Life Cycle (TPLC) regulatory approach that grapples with the question of how postmarket evolution of AI/ML‑based Software that would ordinary require new premarket submissions can be effectively authorized by FDA in advance during the initial premarket review. In a nutshell, it appears that FDA aims to incorporate an envelope of permissible modifications in the field, provided there is sufficient characterization of how they will occur, how they will be controlled, and how patient risks will be monitored and managed.
That is something of an over‑simplification, but not to worry — we will provide a deeper dive into the discussion paper in the very near future!