The UK’s Medicines and Healthcare products Regulatory Authority (MHRA), the US Food and Drug Administration (FDA) and Health Canada have recently published a joint statement identifying ten guiding principles to help inform the development of Good Machine Learning Practice (GMLP). The purpose of these principles is to “help promote safe, effective, and high quality medical devices that use artificial intelligence and machine learning (AI/ML)”.
The development and use of medical devices that use AI and ML has grown considerably over the last few years and will continue to do so. It has been recognised that such technologies have the potential to transform the way in which healthcare is deployed globally, through the analyse of vast amounts of real-world data from which software algorithms can learn and improve. However, as these technologies become more complex and nuanced in their application, this brings into question how they should be overseen and regulated. Crucially, it must be ensured that such devices are safe and beneficial to those who use them, whilst recognising associated risks and limitations.
The guidance confirms that the ten principles are intended to lay the foundation for developing GMLP that “addresses the unique nature of these products” and to “help cultivate future growth in this rapidly progressing field”.
A summary of the guiding principles are as follows:
- Leveraging multi-disciplinary expertise throughout the product life cycle through in-depth understanding of a model’s intended integration into clinical workflow and the intended benefits and any associated risks.
- Implementing good software engineering and security practices, including methodical risk management and design process, risk management decisions and rationale, and ensuring data authenticity and integrity.
- Ensuring clinical study participants and data sets are representative of the intended patient population so that results can be reasonably generalised to the population of interest.
- Utilising training data sets that are independent of test sets, to ensure all potential sources of dependence are addressed to assure independence.
- Basing selected reference datasets upon best available methods to ensure that clinically relevant and well characterised data is collected and that limitations of the reference are understood.
- Tailoring model design to the available data to reflect the intended use of the device and to ensure it supports the active mitigation of known risks.
- Placing focus of the performance of the Human-AI Team, to ensure human factors and interpretability of outputs are addressed, rather than focusing on model performance in isolation.
- Demonstrating device performance during clinically relevant conditions through testing, to generate clinically relevant device performance information independently of the training dataset.
- Providing clear, essential information to users, which is contextually relevant and appropriate to the intended audience; this includes the intended use of the product, indications for use, acceptable inputs and known limitations.
- Monitoring and management of deployed models for performance and re-training risks in “real world” use, with a focus on maintained or improved safety and performance.
The use of AI and ML in medical devices and how this should be effectively governed will no doubt remain a top priority for healthcare regulators. It is clear that the principles should be used as a starting point, intended to lead to further and more far-reaching global collaborative initiatives. The guidance indicates that the authorities have sought to identify areas through which the International Medical Device Regulators Forum (IMDRF), international standards organisations and other collaborative bodies could advance GMLP. It is hoped this will enhance collaboration through research, educational tools and resources, consensus standards and even influence regulatory policies and guidelines.