Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry and Other Interested Parties; Draft Guidance

A professor of practice in Regulatory Sciences and AI at Northeastern University provides constructive feedback on the draft guidance. The commenter suggests refining the definition of AI, proposing a new framework called "AI-Enabled Ecosystems for Therapeutics" (Ai2ET), and recommending the inclusion of a decision tree to help navigate regulatory uncertainty.

The EU Innovative Health Initiative (IHI) VICT3R project, a public-private consortium, supports the draft guidance and recommends expanding the AI definition to include Virtual Control Groups (VCGs). They also suggest clarifying the definition of "credibility" using established frameworks and providing more diverse case studies and examples.

The Animal Health Institute (AHI), a national trade association representing animal health product manufacturers, supports the proposed guidance on using AI in regulatory decision-making. They advocate for a risk-based approach and provide specific recommendations to enhance the guidance, including clarifying data requirements, expanding the credibility framework to include various model types (like LLMs and regression models), and incorporating a structured model development plan.

The National Association of Manufacturers (NAM) supports the draft guidance, praising its risk-based, technology-neutral approach that avoids creating new, burdensome regulations. They recommend specific improvements, such as clarifying "human in the loop" dynamics, defining AI terminology, and leveraging NIST and international standards to manage risk and ensure developer-deployer accountability.

IQVIA, a global provider of advanced analytics and clinical research services, supports the FDA's draft guidance on using AI for regulatory decision-making. They recommend enhancing the guidance by emphasizing real-world testing, incorporating life-cycle monitoring into the core framework, and including specific considerations for generative AI models.

The Connected Health Initiative (CHI) supports the FDA's draft guidance on AI for drug and biological products, particularly its risk-based approach and total product lifecycle perspective. They suggest prioritizing performance testing over training data transparency and urge the FDA to align its policies with existing CDRH guidance and the organization's own health AI principles.

PhRMA, representing the country’s leading biopharmaceutical research companies, supports the FDA's Draft Guidance on using AI for regulatory decision-making. They praise the risk-based "Credibility Framework" but request further clarity on specific definitions, risk assessment criteria, and the handling of third-party and foundation models.

The Centre of Excellence for Regulation Science and Innovation (CERSI) for the Digital Transformation of Medicines Development and Manufacturing provides feedback on the draft guidance, generally supporting its direction while advocating for more specific definitions, global harmonization, and a risk-based approach to minimize the need for individual consultations for low-risk AI models. They specifically recommend establishing a pilot program to gather more diverse industry examples and ensuring alignment with existing chemometric and statistical modeling frameworks.

ForHumanity, a non-profit organization focused on AI risk mitigation, supports the FDA's proactive approach to providing guidance on AI in drug regulatory decision-making. They argue that while the draft provides a sound foundation, it needs more specific, actionable requirements regarding bias mitigation, human intervenability, transparency, and the management of adaptive AI systems.

Global Key Solutions Corp. provides detailed feedback on the draft guidance, suggesting specific terminology changes and clarifications to ensure AI's role is accurately represented as "assisting" rather than "supporting" decisions. They also advocate for clearer definitions regarding human oversight, life cycle maintenance, and the role of the Quality Unit in overseeing AI systems.