FDA Paper on Artificial Intelligence in the Manufacture of Medicines

The US FDA recently published a discussion paper addressing artificial intelligence in the manufacturing of medicinal products. The FDA is considering the extent to which its risk-based regulation is applicable to AI technologies in the manufacturing of medicinal products.

The areas presented in the discussion paper focus on the manufacturing of medicinal products marketed under a New Drug Application (NDA), Abbreviated New Drug Application (ANDA) or Biologics License Application (BLA).

According to the FDA, AI offers many opportunities for the pharmaceutical industry, citing examples such as process optimisation and control, intelligent maintenance, and trend monitoring for continuous process improvement. In addition, AI can pave the way for the introduction of Industry 4.0, i.e. a networked and digitalised pharmaceutical value chain.

Five areas for the use of AI technology in the manufacture of medicinal products

In detail, the document lists five areas where AI will play a role and mentions points to consider.

Cloud applications can impact oversight of pharmaceutical production data and records. Data integrity and data quality must be ensured here. While the FDA allows the use of third parties within GMP manufacturing (with appropriate oversight by the manufacturer), it sees potential issues with existing quality agreements between the pharmaceutical manufacturer and a third party (e.g. for cloud data management). These may have gaps in risk management that arise from the monitoring and control of manufacturing processes by an AI.

The Internet of Things (IOT) may greatly increase the amount of data generated in pharmaceutical manufacturing, impacting existing data management practices. The increase in data may impact both the frequency and type of data. There are GMP requirements for the data and metadata to be stored for each batch of a manufactured medicinal product. However, if the amount of raw data collected during the manufacturing process increases significantly, it may be necessary to balance data integrity and retention with the logistics of data management.

Applicants for a marketing authorisation may need clarity on whether and how the application of AI in pharmaceutical manufacturing is subject to regulatory oversight.

AI could be used in various manufacturing processes, such as monitoring and maintenance of manufacturing equipment, continuous improvement, supply chain logistics and characterisation of raw materials. Applicants need to understand how AI can be used in manufacturing processes that are subject to regulatory GMP oversight.

Standards may be needed for the development and validation of AI models used for process control and to support release testing. AI could also be used to support real-time release testing. However, there are few industry standards or FDA guidances for the development and validation of such models.

Challenges in the use of AI for the authorities

Continuously learning AI systems that adapt to real-time data may also pose a challenge for regulatory oversight.

Currently, models used in manufacturing (e.g. in-process controls, real-time release testing) are developed, validated, implemented and adapted as needed through a change control process within the pharmaceutical quality system. AI models can involve continuous learning, where the model evolves over time as new information becomes available. Thus, if the model changes by itself, it may be unclear when or to what extent a change notification to the agency is required.

The FDA seeks feedback on the following questions or discussion points:

• What types of AI applications can you envision for pharmaceutical manufacturing? 
• Are there additional aspects not mentioned in the discussion paper that should be considered by FDA? 
• Would guidelines in the area of AI be beneficial in pharmaceutical manufacturing? If so, which ones? 
• What are the necessary elements for a manufacturer to implement AI-based models in GMP manufacturing? 
• What are common practices for validating and maintaining a self-learning AI? 
• What mechanisms are required to manage the data used to build AI models for pharmaceutical manufacturing? 
• Are there other aspects of implementing AI-based models in pharmaceutical manufacturing for which further guidance would be helpful? 
• Are there aspects of the application of AI in pharmaceutical manufacturing that are not covered in this document but that FDA should consider?

On the FDA website you can find the Discussion Paper: Artificial Intelligence in Drug Manufacturing, Notice; Request for Information and Comments.