Industry Offers Input on AI in Drug Manufacturing
Pharmaceutical Research and Manufacturers of America (PhRMA) says artificial intelligence (AI) presents many opportunities for industry and it supports FDA’s application of a “risk-based regulatory framework to the use of AI technologies in drug manufacturing.” In just-posted comments to the agency about a recent discussion paper on AI’s use in pharmaceutical manufacturing, PhRMA recommends that FDA use “clear and consistent terminology” when describing when AI in manufacturing comes within the scope of FDA’s regulatory purview.
FDA’s discussion paper (see earlier story) is based on issues raised in a 2021 National Academies of Sciences, Engineering, and Medicine report titled “Innovation in Pharmaceutical Manufacturing on the Horizon: Technical Challenges, Regulatory Issues, and Recommendations.” The innovations in the report “could have implications for measurement, modeling, and control technologies used in pharmaceutical manufacturing,” the agency says. “AI may play a significant role in monitoring and controlling advanced manufacturing processes.”
PhRMA also urges the agency to coordinate with other regulators under international harmonization activities. “A harmonized approach across global regulators would best help foster innovation in this space by minimizing duplicative, overlapping, or inconsistent approaches to regulation,” the group says. The comments also seek FDA flexibility in guidance. “We note that AI is an agile area with new developments occurring rapidly, and the timeline for keeping regulatory guidance current will not likely maintain that pace. Accordingly, we recommend that FDA carefully consider the content that should be specified in guidance versus the content that could be written elsewhere and in more easily updated resources, such as Q&As, FDA Web sites, white papers, bullet-formatted guidances, and points-to-consider documents.”
Regarding data input, PhRMA says AI models are only as good as the data used to train them. “The data input into any AI model is critical for multiple reasons, including in evaluating the algorithms, providing transparency on how the model functions, and understanding how a given model may change/update,” it says. The group also says it is important to understand FDA’s thinking around data delineation, and how data may be used to test the algorithms.
Additionally, model-validation requirements for implementing AI applications should depend on intended uses, PhRMA says. “A risk-based approach to validation requirements can cover varied uses, but it may currently be too early to develop validation criteria for universal usage,” it says. “For example, applications of AI in drug manufacturing that have a direct impact on product quality would have stricter validation requirements. Agency guidance should consider that requirements for AI usage will likely vary based on the application and the associated risk.”
The Biotechnology Innovation Organization (BIO) also submitted comments and recommends that CDER continue to closely coordinate with CDRH’s Digital Center of Excellence on AI and other related digital technology initiatives to better align the two centers.
BIO strongly recommends the use of protocols (e.g., post-approval change management protocols) as a mechanism to submit, review, and maintain AI models. The group says increased frequency in reporting model changes will likely lead to many deviations and ultimately restrict the use of the models. Additionally, BIO recommends that the agency consider using platform technology master files for third parties to maintain certain proprietary information or processes.