In April 2019, the Food and Drug Administration (FDA) published a proposed regulatory framework for evaluating modifications for artificial intelligence/machine learning (AI/ML) software as a medical device (SaMD). AI/ML technologies in SaMD present unique issues, because the technology is intended to learn from real-world feedback and incorporate improvements into the SaMD algorithms. Since publishing the proposed framework, the FDA utilized workshops, publications and other means to obtain stakeholder feedback from device manufacturers and the public. The recently published action plan describes the agency’s intended actions incorporating this feedback.
The FDA’s 2019 proposal incorporated a “Predetermined Change Control Plan”, in which the manufacturer would identify what aspects of the SaMD the manufacturer intends to change through machine learning (the SaMD Pre-Specifications, SPS), and how the algorithm will learn and change while maintaining safety (the Algorithm Change Protocol, ACP). The agency received numerous comments on the SPS and ACP, as well as suggestions for performing a focused review of the Predetermined Change Control Plan. The FDA intends to publish draft guidance on the content of the SPS and ACP, identification of types of modifications to AI/ML devices, and process for focused review. The goal is to publish the draft guidance in 2021.
Other activities included in the action plan are continued efforts to encourage harmonization of Good Machine Learning Practice developments, which will be pursued in collaboration with the FDA’s Medical Device Cybersecurity Program; convening a public workshop on how device labeling can clearly convey to users (including patients) how the device’s algorithms will be modified over time; supporting regulatory science methods to improve ML algorithms, including identification of bias; and collaborate with stakeholders on real-world performance monitoring pilots.