FDA Seeks Feedback on AI in Drug Development, Manufacturing

The FDA is working to adapt its regulatory regimes to the use of artificial intelligence (AI) across the agency’s product centers, a response to the frequent use of AI for a variety of regulated products. The FDA’s drug center recently released two discussion papers on AI, one each for drug development and drug manufacturing, feedback from which will help set the agency’s expectations for such uses over the next few years.

In a May 10 statement, the FDA stated that AI software with machine learning (ML) capabilities is within the scope of both discussion papers, and that digital replicas of human physiology are already being used to model a medical intervention prior to human clinical studies. In 2021, more than 100 applications to the FDA’s centers for pharmaceuticals and biotech therapies included AI/ML components, but the opacity of these products is a source of concern.

Other sources of concern for the FDA in the context of AI and ML are bias and cybersecurity, all of which drive the need for a risk-based approach to the regulation of AI and ML in drug development and manufacturing. Patrizia Cavazzoni, director of the FDA’s Center for Drug Evaluation and Research (CDER), said the agency will convene a workshop that will draw on the expertise of drug and biotech development experts to determine how AI and ML can be used in these endeavors while remaining mindful of the associated challenges.

The FDA’s discussion paper for drug manufacturing discusses the use of AI and ML in the context of advanced manufacturing technologies, but also poses the question of how the current Good Manufacturing Practices (cGMP) might be applied to AI and ML. The application types that are included in this discussion paper are new drug applications, abbreviated new drug applications and biologics license applications.

Among the questions the FDA would like to explore are how AI can be used to design and scale up drug manufacturing processes and to develop and manage advanced process controls. The use of AI/ML in process controls requires the addition of sensors in the production process, but this paper states that several manufacturers have already reported the incorporation of these technologies in their production lines.

Other uses of AI/ML include process monitoring and fault detection, and trend monitoring for consumer complaints and deviation reports. The FDA sees the use of cloud applications to manage the drug manufacturing process as potentially problematic because of the possibility of data management lapses. The use of cloud computing may also introduce new complexities to the FDA’s facility inspection process.

The discussion paper for drug development highlights the role that AI and ML can play in identification of a target for a drug therapy and to identify any potential drug-target interactions in the context of drug repurposing. AI and ML can also be used in the clinical research phase of drug development by assisting with the recruitment of potential clinical study participants and the selection and stratification of those who are eventually enrolled in a study.

CDER has established a steering committee for the use of AI and ML in drug and biologics development, which coordinates these efforts between CDER and the Center for Biologics Evaluation and Research (CBER). The paper reviewed recent developments at the Center for Devices and Radiological Health (CDRH), including the recent draft guidance for predetermined change control plans (PCCP), which we previously highlighted.

The three key areas for consideration in discussion paper start with human-led governance, accountability and transparency, which the FDA said may require additional regulatory clarity via guidance. The second key area is the quality, reliability and representativeness of data, which also encodes the question of the replicability of the results of studies using the AI or ML software. The third area is model development and validation, which poses the question of how to determine whether a data source used in model development is fit for the purpose of the software.

The FDA will hold the workshop to discuss these questions Sept. 26-27, 2023. Registration information can be found here.