Software & AI
Good Machine Learning Practice(GMLP)
In one line
Guiding principles for the development of AI/ML-enabled medical devices.
Definition
Good Machine Learning Practice (GMLP) describes guiding principles - jointly issued by FDA, Health Canada, and the UK MHRA - for developing AI/ML-enabled medical devices that are safe, effective, and high-quality.
The 10 principles, published in October 2021, cover multidisciplinary expertise across the lifecycle, good software engineering and security, representative data, independent training and test sets, a focus on the human-AI team, clear user information and monitoring of deployed models. The same three regulators later published related principles on PCCPs and transparency.
Why it matters
GMLP principles inform FDA review expectations, including data management, model validation, and human-AI team performance.
Common pitfalls
- •A common pitfall is failing to establish a robust data management plan that addresses data quality, integrity, and representativeness, potentially leading to biased or inaccurate model performance.
- •Misinterpreting GMLP as a static checklist rather than a set of guiding principles can lead to superficial compliance without true assurance of device safety and effectiveness.
- •Underestimating the importance of real-world performance monitoring and continuous learning for adaptive AI/ML models can result in undetected performance degradation over time.
- •Neglecting to involve multi-disciplinary teams including clinicians, data scientists, and regulatory experts throughout the device development process can lead to overlooked risks and inefficiencies.
- •Failing to adequately document the rationale for model choices, data curation, and validation strategies can hinder regulatory review and post-market surveillance.
Sources