The US Food and Drug Administration (FDA), along with Health Canada and the United Kingdom's Medicines and Healthcare Products Regulatory Agency, have identified ten guiding principles for development of safe and effective medical devices that use artificial intelligence and machine learning (AI/ML). AI/ML technologies have the potential to improve device performance by deriving insights from data generated through the delivery of health care in real-world use. The ten principles, called Good Machine Learning Practice (GMLP), are intended to be used to identify GMLP best practice and consensus standards.
The principles are:
- Multi-disciplinary expertise is leveraged throughout the total product life cycle.
- Good software engineering and security practices are implemented. The agencies explain this should include the fundamentals of good software engineering practices, data quality assurance, data management and cybersecurity.
- Clinical study participants and data sets are representative of the intended patient population. Data collection should ensure that relevant characteristics of the intended population (such as demographics) are represented in a sample of adequate size so that results can be generalized to the population.
- Training data sets are independent of test sets.
- Selected reference datasets are based upon best available methods.
- Model design is tailored to the available data and reflects the intended use of the device. Design should support the active mitigation of known risks, like overfitting, performance degradation, and security risks.
- Focus is placed on the performance of the human/AI team. Developers need to address human interpretability of the model outputs.
- Testing demonstrates device performance during clinically relevant conditions. Test plans are developed to generate clinically relevant device performance information independent of the training data set, considering the intended patient population, clinical environment and other factors.
- Users are provided clear, essential information. Users have access to information appropriate for the intended audience, such as health care providers or patients, and a means to communicate product concerns to the developer.
- Deployed models are monitored for performance and re-training risks are managed. Models must be monitored in real world use for potential improvement of safety and performance.