Good Machine Learning Practice: FDA issues Discussion Paper on AI/ML in Drug Development

The FDA recently issued three Documents / Discussion paper on the use of Artificial Intelligence / Machine Learning (AI/ML) in "Drug Manufacturing", "Change Control for Medical Devices" and "Drug & Biological Development".

AI/ML in Drug and Biological Product Development

According to the agency, AI/ML is increasingly integrated in areas where FDA is actively engaged, including Digital Health Technologies (DHTs), and Real-World Data (RWD) analytics. The published discussion paper is intended to initiate communication surrounding AI/ML with stakeholders, including industry and academia, to promote mutual learning and discussion. It will complement and provide future guidance on AI/ML in drug development. Comments can be sent to the FDA until 9 August 2023.

In particular, the FDA wants to receive feedback on the key areas in the context of AI/ML. These areas are: 

• Human-led governance, accountability, and transparency
• Quality, reliability and representativeness of data
• Model development, performance, monitoring 
• Verification and validation of AI/ML

FDA's Questions

• What are examples of current tools, processes, approaches, and best practices being used?
• What practices and documentation are being used to record data source selection?
• What approaches are being used to document the assessment of uncertainty in model predictions?
• How is uncertainty being communicated?
• What methods and standards should be developed to help support the assessment of uncertainty?

Good Machine Learning Practice (GMLP) and FDA's Experience with AI/ML in Drug Development

AI/ML is increasingly integrated in areas where FDA is actively engaged, including clinical trial design, DHTs, and RWD analytics. Over the last few years, the agency has seen a rapid growth in the number of submissions that reference AI/ML. These submissions include a range of therapeutic areas, and the uses of AI/ML within the submissions cover many different areas of the drug development process (e.g. from drug discovery and clinical trial enrichment to endpoint assessment and postmarket safety surveillance). Inclusion of AI/ML in the clinical development/research phase represents the most common stage for AI/ML uses in submissions.

One of the ways FDA has been supporting the development of innovative and robust AI/ML is through the establishment of the CDER AI Steering Committee (AISC), which coordinates efforts around AI/ML uses across therapeutic development. In addition, the agency is developing a framework for AI/ML-based devices, including predetermined change control plans for devices incorporating AI/ML, as well as a foundation for Good Machine Learning Practice (GMLP) for medical device development.

For more information on the draft guidance and how to comment, please see the Federal Register notice and visit the Artificial Intelligence and Machine Learning for Drug Development webpage.