FDA Issues Long-Awaited Action Plan for Artificial Intelligence/Machine Learning-Based Software As a Medical Device

On January 12, 2021, the FDA issued its long-awaited action plan concerning the regulation of artificial intelligence (AI) and machine learning (ML)-based Software As a Medical Device (SaMD).  The plan comes on the heels of an April 2019 FDA white paper, which provided an initial proposed regulatory framework for SaMD, as well as an open comment period in which the FDA solicited stakeholder feedback.  The action plan outlines five primary goals and commitments to advance the FDA’s interest in facilitating the innovation of SaMD while developing proper oversight, as follows.

First, the FDA commits to issue draft guidance on the Predetermined Change Control Plan first described in its initial white paper.  One of the key challenges identified by the FDA in that paper was the regulation of AI/ML-based products, which may autonomously learn and evolve over time with use – in some cases, potentially rendering them unrecognizable from the product initially approved by the FDA.  Consequently, the FDA proposed that manufacturers provide a Predetermined Change Control Plan to specify precisely what aspects of the device would be changed by ML, and how the prescribed algorithm would implement any changes.  The action plan takes this a step further, indicating the FDA has gathered sufficient information to publish its first draft guidance on what specific information should be included in the Predetermined Change Control Plan.

Second, the FDA commits to encourage and harmonize the development of Good Machine Learning Practice (GMLP).  The FDA’s initial white paper described GMLP as a “set of AI/ML best practices . . . akin to good software engineering practices or quality system practices” that might be prescribed for all manufacturers as a kind of baseline methodology.  The FDA has now further committed to the concept of GMLP, including by partnering with standard-setting bodies and collaborative communities worldwide in order to harmonize these separate standardization efforts into a single set of guidelines.  The action plan makes clear GMLP will be a significant part of whatever regulatory framework is ultimately adopted by the FDA.

Third, the FDA commits to holding a public workshop on how device labeling supports transparency to users and enhances consumer trust in AI/ML-based products.  Another concern identified in the FDA’s initial white paper was the potential lack of transparency in the way AI/ML algorithms function, which could, in turn diminish consumer confidence in SaMD.  To address this concern, the action plan indicates the FDA will hold a public workshop to discuss and identify “types of information that FDA would recommend a manufacturer include in the labeling” of its SaMD to support transparency and increase consumer confidence.

Fourth, the FDA indicates it will provide support for efforts to develop a methodology to evaluate and improve ML algorithms in order to identify and eliminate potential biases.  Such a methodology is warranted “[b]ecause AI/ML systems are developed and trained using data from historical data sets” and thus are “prone to mirroring biases present in the data” including those based on race, ethnicity, and/or socio-economic status.  In an effort to foster the elimination of such biases, the action plan confirms that the FDA will support regulatory science research efforts, including through collaborations with AI researchers at academic institutions around the country.

Finally, the FDA’s action plan commits to the development of a Real-World Performance monitoring program, through which stakeholders can work with the FDA to employ SaMD in real-world settings.  The goal of this program is to determine what kind of metrics and performance evaluations should be collected in order to evaluate how SaMD are performing in a real-world setting,  The FDA has also expressed, both in its initial white paper and in the action plan, that it will incorporate this kind of Real-World Performance monitoring into the final regulatory framework it adopts.  Thus this collaborative venture will directly influence the kinds of monitoring future SaMD will be subject to.

While a final framework for the regulation of AI/ML-based SaMD may still be years off, the FDA’s latest action plan constitutes an unambiguous recognition that SaMD is here to stay.  Moreover, the action plan reflects a clear commitment to collaborate with stakeholders to ensure that the proper balance between innovation and regulation is achieved.  With this publication, the FDA has highlighted some of the key issues that manufacturers will need to consider as they approach the research and development of SaMD.  These include, for example, adopting a clear and transparent framework to explain how algorithms function, as well as the development and implementation of protocols to identify and minimize the impact of biases on ML algorithms.  For many manufacturers, these are entirely new considerations, which will necessitate a robust approach to ensure regulatory compliance.

© 2009- Duane Morris LLP. Duane Morris is a registered service mark of Duane Morris LLP.

The opinions expressed on this blog are those of the author and are not to be construed as legal advice.

Proudly powered by WordPress