AI and ML in Medical Devices: The FDA has issued guidance documents that clarify the agency’s current thinking on the regulation of AI/ML technologies, aiming to create a flexible regulatory framework for these innovations.

EU Regulation (EU) 2017/745: Applicable in the European Union, this regulation emphasizes the accountability and traceability of AI systems used in medical devices and pharmaceuticals, offering a similar context as that of the FDA.

Documentation Requirements

When implementing AI systems within GxP environments, specific documentation practices must be followed to ensure compliance and facilitate inspections. These include:

• Validation Documentation: AI systems must be thoroughly validated to ensure their intended use aligns with regulatory requirements. This includes predefined validation protocols detailing the verification of AI outputs and performance criteria.
• Change Control Records: Any modifications to the AI system must be documented through a robust change control process. This should include the rationale for changes and impact assessments on GxP activities.
• Training Records: Staff involved in operating AI systems should receive proper training. Documentation of training sessions, materials used, and participant lists should be maintained for traceability.

Review and Approval Flow

The regulatory review and approval flow for AI interventions in GxP systems generally adhere to traditional pathways but require consideration of unique AI attributes:

1. Pre-Submission Preparation: Sponsors must conduct thorough due diligence and compile documentation evidencing compliance with regulatory expectations prior to submitting any application.
2. Agency Engagement: Engaging with the FDA or EMA early in the development phase can provide insights into agency expectations and alignment on the proposed use of AI technologies.
3. Submission and Review: After submission, the agency will conduct a review, potentially raising questions about data integrity, rationale for AI incorporation, and risk evaluation associated with ML algorithms.
4. Approval Decision: Based on the review findings, agencies will grant approval or require additional data. Key response submissions should address any deficiencies identified during the review.

Common Deficiencies and How to Avoid Them

Investing in AI systems without proper groundwork can lead to common deficiencies during regulatory reviews. Below are notable pitfalls and recommendations to avoid them:

• Lack of Clear Justification: Provide a robust justification for the use of AI in critical operations rather than traditional methods. Clearly articulate the benefits and specific GxP applications, along with scientific evidence supporting these claims.
• Poor Validation Practices: Ensure validation efforts are comprehensive and follow strict protocols. Include stakeholder input and assign clear responsibilities to facilitate accountability throughout the validation steps.
• Inadequate Data Management: Maintain a strong data governance framework for AI outputs, including methodologies for data collection, processing, and evaluation. This ensures transparency and supports regulatory scrutiny.

RA-Specific Decision Points

Regulatory Affairs professionals must evaluate specific decision points when considering the use of AI within GxP systems. These points include:

Variation vs. New Application

Determining whether to submit a variation or a new application is critical when introducing AI technology into existing workflows:

• Evaluate the extent of changes: Minor enhancements that optimize existing processes may only require a variation. However, significant modifications that fundamentally alter product safety or efficacy may necessitate a new application.
• Documentation of impact: Providing data to support your decision is essential. Perform a risk-benefit analysis to outline the implications of the use of AI on quality, safety, or efficacy.

Justification for Bridging Data

When leveraging AI in GxP settings, determining the necessity for bridging data is essential:

• Consider whether existing data can support AI outcomes. If new algorithms produce results that deviate significantly, additional bridging data may be required to justify compliance with existing standards.
• Anticipate agency scrutiny: Prepare for potential questions regarding the validity of the bridging data and how it correlates with previous clinical or analytical outcomes.

Conclusion

Utilizing AI and ML in GxP quality systems offers promising advancements for the pharmaceutical and biotech industries. However, it introduces complexity in regulatory compliance. By understanding regulatory expectations and documenting compliance efforts effectively, RA professionals can navigate the challenges posed by these cutting-edge technologies. A proactive approach in addressing FDA expectations and continuous engagement with regulatory agencies not only simplifies approvals but also fosters a culture of innovation within organizations.

For further guidance, relevant regulations, and detailed information regarding AI applications in GxP contexts, it is advisable to consult the FDA Guidance on AI and ML in Medical Devices.