By Ismail Amin
This article was originally published by The Amin Law Group on March 6, 2023.
On March 1, 2023, the FDA’s Center for Drug Evaluation and Research (CDER) issued a discussion paper on the role of Artificial Intelligence (AI) in the context of drug manufacturing and the implications for “measurement, modeling, and control technologies used in pharmaceutical manufacturing” (FDA, 2023).
The discussion paper referenced Current Good Manufacturing Practice(s) (CGMP) regulations and the potential “ambiguities” and discrepancies that may arise when implementing AI for new drug products, under New Drug Application (NDA); Abbreviated New Drug Application(s) (ANDA); & Biologics License Application(s) (BLA). Interestingly, the paper acknowledged gaps in the current regulatory framework considering the rapid advancements in AI. It’s certainly inevitable AI will be implemented to varying degrees to speed up the manufacturing process, but it must be done so both safely and efficiently. Recent news has exuberantly highlighted the promise of AI, but it has also exposed its shortcomings.
To wit, the context of the CDER discussion paper was to invite industry discussion into the following areas:
- Cloud applications and their oversight of pharmaceutical manufacturing data and records.
Here, the emphasis relates to data integrity and quality with respect to third-party data management systems (located near the manufacturing site or far away). More specifically, while the current regulations allow manufacturers to utilize third parties for certain CGMP data management functions, existing agreements may have gaps “with respect to managing the risks of AI in the context of manufacturing monitoring and control.” Also, what about the cybersecurity implications of having third-party AI programs aggregating, controlling, and storing sensitive data? These implications must be addressed by manufacturers and third-party software firms alike.
- Digitization of Manufacturing controls and data management issues.
In the context of the Internet of Things (IoT), there are data management concerns, such as how to store excessive data. Where to store data? And which data gets (safely) discarded? Of course, all of this must be done while adhering to international, federal, and state privacy laws.
- The Application of AI “in manufacturing operations that are subject to regulatory oversight”.
With respect to manufacturing operations, NDA, ANDA, and BLA applicants will need to appropriately discern which parts require regulatory oversight. For instance, how would supply chains or raw materials be implicated by the implementation of AI during the manufacturing processes?
- The need for standards “for developing and validating AI models used for process control and to support release testing”.
In this regard, manufacturers would need to gain “clarity regarding how the potential to transfer learning from one AI model to another can be factored into model development and validation.” While AI’s benefits are substantial to the manufacturing process (supporting analytical procedures for product testing; supporting real-time release testing; and predicting in-process quality attributes), there are unknown variables that must be addressed. For instance, how does a manufacturer ensure that a software program is appropriately applying one input for one product, into the next?
- “Continuously learning AI systems that adapt to real-time data may challenge regulatory assessment and oversight.”
Finally, how is an ever-evolving AI model ever “considered an established condition of a process”? Here, the CDER highlights concerns with FDA examinations and verifications of “model lifecycle strategy” and “continuously updated AI control models” during site inspections. In other words, it’s difficult to measure a moving target.
AI’s proliferation into the pharmaceutical and biologics manufacturing world was inevitable. The implications for bringing drugs and biologics to market much quicker and with measured scalability are profound. However, the counterweight of product safety, data integrity, and manufacturing reliability cannot be diminished.