As artificial intelligence (AI) continues to expand into all sects of technology, pharmaceutical partners may be wondering how such a relatively new and unregulated technology can be used in a highly regulated environment to enhance patient safety and product access. AI has already been proven to drive efficiency across pre-commercial organizations in early drug discovery due to its lower regulatory barrier, but using AI-enabled tools in GxP-regulated environments introduces additional complexity. Below, we discuss considerations for leveraging AI in GxP-regulated environments, including use cases as well as the need to first establish a solid AI strategy.  

What’s the FDA’s current stance on AI tools? 

As organizations in all industries increasingly rely on AI and machine learning to optimize operations, the FDA, in particular, is now leveraging the tool to further drive efficiency within the drug approval process. 

The FDA is in the process of establishing guidance documents related to the use of AI in the pre-commercial phase to support regulatory decision-making. This means the document does not address the use of these tools, however, in the early drug discovery or commercial phases.  

The FDA has included a risk-based credibility assessment framework in the draft guidance to support the creation and validation of artificial intelligence tools:  

1. Define the question of interest addressed by the AI model 
2. Define context of use 
3. Assess the AI model risk 
4. Develop a plan to establish AI model credibility within the context of use 
5. Execute the plan 
6. Document the results of the credibility assessment plan and discuss deviations 
7. Determine adequacy of the AI model for the context of use 

These steps ensure the reliability of the product and quality of the training data, which should be considered when creating any AI-enabled tool. 

It’s important to note that FDA guidance documents serve as recommendations rather than requirements. Companies should consider these guidance documents as requirements to validate that tools in use promote safety and efficacy of the product development and manufacturing processes.  

The FDA will also likely produce requirements surrounding the topics discussed in guidance documents. By adhering to current guidance, there’s a lower risk of remediation required for the tool itself (or products created or validated using the tool) upon publication of future regulations. 

The FDA published a discussion paper in March of 2023 regarding the use of AI in the drug manufacturing process, likely indicating future guidance regarding the subject. The paper references areas of consideration and associated regulations that may guide pharmaceutical companies during their AI journey. 

Additionally, the FDA has implemented their own artificial intelligence tool, called Elsa, to aid their employees. This will potentially lead to a shortened scientific review timeline and faster patient access. 

As more AI tools are implemented across highly regulated environments, it’s important to note that all tools still require human oversight. AI shouldn’t replace knowledgable individuals in completing regulated tasks. For example, the FDA’s tool may summarize adverse events data, but this output must still be reviewed by scientists to properly evaluate and verify the safety of a product. 

Opportunities for AI in a Regulated Environment 

If appropriate measures are taken, AI has the potential to transform commercial pharmaceutical organizations by performing repetitive tasks that enhance productivity. 

For example, a chatbot could be used to review drafted Standard Operating Procedures (SOPs), Work Instructions (WIs), Change Controls, and other business documentation against a set of FDA regulations to ensure compliance, and indicate where necessary items are missing. The documents would be drafted, reviewed, and routed by employees, emphasizing the necessity for human interactions with the AI tool. 

Another use case of AI is to develop a Certificate of Analysis (CoA)  for production batches. The large in-process and release testing datasets can be quickly analyzed by an AI tool. The tool can notify manufacturers of attributes out of specification and prompt a Corrective Action or Preventive Action (CAPA) workflow directly in a Quality Management System (QMS).  

Additionally, the tool can create trending reports by aggregating manufacturing data from multiple batches. Each of these documents produced by AI require human review prior to official sign-off to ensure specification compliance before patients receive a drug. 

AI can also be used to optimize manufacturing operations, such as through its use with digital twins, a virtual model of a manufacturing system, as well as to simulate planned system maintenance, reducing production downtime. Organizations can analyze the results of proposed manufacturing changes, identify risks, and prepare a change impact assessment prior to making any investment. 

Thinking Through Your AI Strategy  

When preparing to implement AI into your GxP environment, it’s necessary to evaluate the training data, use cases, and risks associated with the tool. Additionally, proper governance of the tool may need to be established to further ensure security and compliance. For example, a company may consider implementing firewalls that prevent widespread access and ensure proper training, promoting security. Additionally, the tool must be compliant with data integrity regulations and promote data traceability. Finally, the tool must be periodically audited to ensure consistent quality. 

To prepare for this institutional investment, an implementation roadmap may be necessary to best consider compliance, governance, and training data. Contact Clarkston’s quality and compliance experts to develop your GxP-compliant AI strategy.