The integration of artificial intelligence (AI) and machine learning (ML) in healthcare has transformed how medical technology is developed, evaluated, and deployed. This innovation calls for adherence to rigorous quality standards to ensure safety, efficacy, and compliance. While traditional Good Practice (GxP) guidelines—such as Good Manufacturing Practice (GMP), Good Clinical Practice (GCP), and Good Laboratory Practice (GLP)—provide a robust framework for product development and quality assurance, they do not explicitly encompass AI/ML-based technologies. However, the principles of Good Machine Learning Practice (GMLP) serve as an extension of GxP principles tailored for the AI/ML landscape. Regulatory authorities, including the U.S. Food and Drug Administration (FDA) and the International Medical Device Regulators Forum (IMDRF), have adopted these principles, originally compiled by a collaborative, and the GMLP terminology to guide the development and oversight of AI/ML-enabled medical devices.
Understanding GxP in Healthcare
GxP is a collection of quality guidelines and regulations designed to ensure that products are safe, meet quality standards, and adhere to regulatory compliance throughout their lifecycle. Core components include:
- Good Manufacturing Practice (GMP): Focuses on manufacturing processes to ensure products are consistently produced and controlled according to quality standards.
- Good Clinical Practice (GCP): Guides ethical and scientific standards for designing, conducting, and reporting clinical trials involving human subjects.
- Good Laboratory Practice (GLP): Addresses non-clinical laboratory studies to ensure the integrity and quality of safety data.
These frameworks prioritize product safety, efficacy, and data integrity—principles that are increasingly vital as healthcare integrates advanced technologies like AI/ML.
Introducing Good Machine Learning Practice (GMLP)
The GMLP principles were developed to guide the development, deployment, and maintenance of AI/ML-based medical devices. These guidelines focus on ensuring that machine learning systems are safe, effective, and aligned with user needs throughout the product lifecycle. Key GMLP principles include:
- Multi-Disciplinary Expertise Throughout the Total Product Lifecycle: Collaboration across clinical, technical, and regulatory teams ensures that AI/ML models are designed with efficacy as well as patient safety and pharmacovigilance in mind.
- Good Software Engineering and Security Practices: Adopting robust software development methods and cybersecurity measures safeguards model integrity and data privacy.
- Representative and Relevant Data: Collecting diverse, high-quality data reduces bias and enhances the model’s generalizability across patient populations.
- Separation of Training and Test Data: Preventing data leakage by maintaining strict boundaries between training and testing datasets ensures objective model evaluation.
- Model Design Aligned with Intended Use: Designing models tailored to their intended clinical application optimizes safety and performance.
- Focus on Human-AI Team Performance: Ensuring seamless integration between human users and AI systems maximizes the effectiveness of clinical workflows.
- Testing Under Real-World Conditions: Validating models in real-world clinical environments confirms their practical reliability and safety.
- Transparency and User Information: Providing clear documentation about model functionality and limitations supports informed decision-making by users.
- Monitoring and Maintenance: Continuous post-market monitoring ensures models maintain performance and adapt to new data or clinical insights.
Incorporating the GMLP Framework
The GMLP includes detailed considerations across three critical domains:
- Planning and Architecture: Defining the AI application’s scope, assessing whether it makes autonomous decisions, and ensuring responsible data handling.
- Regulatory Compliance: Documenting device descriptions, design justifications, and comprehensive software testing procedures to maintain traceability between design, risk management, and validation.
- Lifecycle Management: Evaluating how AI systems evolve over time, managing system retirement, and implementing robust backup mechanisms.
Bridging GxP and GMLP: A Unified Approach to Quality in AI/ML
Though GMLP is not formally endorsed by the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH), its principles naturally align with GxP standards. Both frameworks emphasize:
- Risk-Based Approaches: Identifying and mitigating risks throughout the product lifecycle aligns with both GxP and GMLP.
- Data Integrity: Ensuring the accuracy, consistency, and reliability of data is foundational to both sets of guidelines.
- Documentation and Traceability: Comprehensive records of design, development, and testing processes are essential for regulatory compliance and quality assurance.
- Continuous Monitoring and Improvement: Ongoing performance evaluation and updates align with GxP’s lifecycle approach to quality management.
By integrating GMLP principles with traditional GxP frameworks, healthcare organizations can better navigate the regulatory landscape, fostering the development of safe, effective, and reliable AI/ML medical technologies.
Conclusion
The convergence of GxP and GMLP creates a comprehensive quality framework that supports the safe and effective integration of AI/ML technologies in healthcare. While GMLP is not yet an ICH-endorsed standard, its alignment with GxP principles ensures that AI/ML-enabled medical devices meet the high standards of safety, efficacy, and regulatory compliance essential in the healthcare industry.
Interested in how FDA is looking at AI-enabled tech as a regulator? Check out our related blog: Engaging with the FDA on AI in Clinical Trials: Beyond Traditional Meetings
References
- Healthcare Products Collaborative. Good Machine Learning Practice (GMLP) Framework. https://healthcareproducts.org/ai/aighi/gmlp/
- U.S. Food and Drug Administration (FDA). Good Machine Learning Practice for Medical Device Development. https://www.fda.gov/
- International Medical Device Regulators Forum (IMDRF). AI/ML in Medical Devices. https://www.imdrf.org/
