AI in drug manufacturing: Stakeholders call for harmonization, further guidance

A discussion paper released by the US Food and Drug Administration (FDA) on incorporating artificial intelligence (AI) and advanced manufacturing techniques into the pharmaceutical manufacturing regulatory framework has yielded numerous and wide-ranging comments from stakeholders detailing their current and intended uses for the technology.

FDA asked for more information on the types of AI applications used, the elements of AI-based models in a current good manufacturing practice (CGMP) environment, practices for validating and maintaining self-learning AI models, appropriate AI data management and what guidance for AI stakeholders would like to see, as well as other aspects of AI in pharmaceutical manufacturing not listed in the discussion paper. (RELATED: FDA seeks feedback on artificial intelligence in drug manufacturing, Regulatory Focus 1 March 2023)

Stakeholders offered a wide range of applications for AI technology in the pharmaceutical manufacturing regulatory framework, including Chemistry, Manufacturing, and Controls (CMC) development, process control and development, process design and scale, autonomous systems for drug manufacturing, producing planning, management of assets, quality control, quality assurance, modeling for stability and shelf-life, management of documents, training, supply chain optimization, sustainability, labor and end-to-end operations visibility, among others.

“It is key to distinguish between applications where AI is being used to define the operational space, for example, ‘This is what it looks like when the process runs correctly’ and where AI is being used for control within the operational space, for example, ‘This lot was produced by a process that ran correctly,’” the National Institute for Innovation in Manufacturing Biopharmaceuticals (NIIMBL) wrote in their comment. “In addition to its role in pharmaceutical development and manufacturing, AI will likely play a role in other parts of the product lifecycle, such as supply and distribution networks to encourage end-to-end quality-by-design approaches.”

Commenters were also interested in seeing the agency release guidance on a number of different AI-related topics such as data systems and architecture, data collection and preparation, model development, machine learning algorithms, maintenance, validation, documentation, training, validation and verification of algorithms, technical transfer, acceptable lifestyle strategy for models, and relationships with third party providers of AI.

In their comment, the Association for Accessible Medicines (AAM) said additional guidance on AI use in drug manufacturing is a “critical need” and the agency “should determine the areas where AI is most likely to make the greatest contribution in manufacturing and focus on clarifying the regulatory requirements applicable to AI in those areas.”

Harmonization efforts

Members of the Connected Health Initiative (CHI) encouraged FDA’s Center for Drug Evaluation and Research (CDER) in their comment to ensure guidance aligns with what has already been published by the Center for Devices and Radiological Health (CDRH). Other stakeholders offered similar calls for harmonization with existing organizations developing AI-related documents such as the International Council on Harmonisation (ICH).

“We strongly support the use of global harmonization and clarity through existing venues such as ICH. Additional guidance on the use of AI in drug manufacturing would be helpful, but USFDA guidance in isolation within a global marketplace might add unintended complexity for manufacturers,” NIIMBL wrote.

The International Society for Pharmaceutical Engineering (ISPE) agreed in their comment that a harmonization approach was appropriate with AI, particularly when aligning with entities such as the European Commission Proposal for a Regulation of the European Parliament and of the Council (Artificial Intelligence Act) and the European Union Aviation Safety Agency.

“Aligned regulatory expectations and vocabulary would enable regulatory convergence and support multinational companies to apply AI technology, thereby increasing operating efficiencies and reducing the cost of goods,” ISPE wrote. “Engagement with industry professional societies, like ISPE, could enable faster and more complete guideline development.”

Changes to regulatory framework

In their comment, the Biotechnology Innovation Organization (BIO) said existing CGMP guidance for models and frameworks need to be updated with clarification on whether the guidance applies to AI models. For AI models and machine learning (ML), BIO noted they would welcome guidance on AI and ML intended use, acceptability, tuning, and other parameters.

Some of the more public discussions concerning AI and ML may not apply to manufacturing, they explained. “The publicly visible regulatory discussion on the use of AI and ML in drug development is often focused on applications for ‘big data’ such as real-world evidence and clinical trial data modeling, where specific consideration is given to challenges like data ownership, privacy, and, ethics which are not as prevalent in the AI application in manufacturing,” they said. “Considering this, a distinction of regulatory guidance for manufacturing and CGMP versus such applications would be valuable.”

The Pharmaceutical Research and Manufacturers of America (PhRMA) requested more information from FDA in their comment outlining regulatory submission documentation for AI in manufacturing as well as the need for proportional documentation of “explainability and transparency of AI models.”

“PhRMA strongly supports a risk-based approach to the type and amount of information needed, including clear delineations between what is expected in a regulatory submission and what would need to be available for an FDA inspection,” they said.

ISPE agreed, noting “the basic principles and regulatory framework for quality oversight should not be changed but may be fulfilled through industry-accepted risk-based approaches.”

AAM pointed out there is also an opportunity for the development of separate regulatory standards or a certification program for third-party developers and suppliers of AI technology “to provide additional assurance of adequate oversight and data safety and security.”

In general, “clarifying how existing regulatory requirements will be applied to AI is crucial to facilitating its adoption by the generic drug industry,” AAM said. “Generic pharmaceutical companies operate in a highly competitive market with low profit margins on individual products. Therefore, the benefits of investing in new technology must be compelling and offer benefits across product lines.”

In their comment, Amazon said FDA’s decision to allow third party providers for CGMPs is welcome, but asked for more information on the issue of manufacturing monitoring and control gaps raised in the discussion paper.

“FDA states in the white paper that these gaps could ‘lead to challenges in ensuring that the third-party creates and updates AI software with appropriate safeguards for data safety and security,’” Amazon noted. “Additional clarity from FDA on the types of controls and documentation for these controls would help alleviate delays or confusion as part of FDA inspections.”

Emerging technologies

PhRMA expressed their belief that FDA should be “technology-agnostic” when developing guidance for AI. “[G]iven the pace of technological change in this space, any guidance should be flexible and the Agency should carefully consider the content that should be specified in guidance versus the content that could be written in more easily updated documents, such as Q&As, reflection papers, and points-to-consider documents,” they said.

The CHI noted FDA’s emerging technology discussions should be done the frame of improved responsible deployment and streamlined processes. Citing the example of potential issues with security and control gaps between manufacturers and third parties providing cloud data management as “speculative assertions,” CHI said cloud computing “enables secure, ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction.”

“In simple terms, cloud computing allows organizations to leverage servers and access computer system resources—such as computing power, storage, and network power—to meet their changing technology needs and are increasingly relied upon throughout the healthcare ecosystem,” they said. “The capabilities of cloud computing are necessary tools for advancing FDA’s interests and goals in pharmaceutical manufacturing data and recordkeeping. It is crucial that FDA’s discussion of cloud computing in the Discussion Paper be revised reflect its value and utilities.”

In response to cloud computer references made in the discussion paper, Amazon said “the agency should take a broader approach to accelerate adoption of innovative technologies that can improve product quality, facilitate innovation, and reduce burden in meeting regulatory requirements.” The company said guidance from FDA on cloud-based technologies as they relate to manufacturers, producer reviewers, and FDA investigators would be welcome and “provide explicit examples on ways cloud services can help manufacturers meet and exceed regulatory requirements, such as document retention and access controls under Part 11, CGMPs, and Good Laboratory Practices.”

AI discussion paper