The U.S. Food and Drug Administration (FDA), Health Canada, and the UK’s Medicines and Healthcare products Regulatory Agency (MHRA) have jointly identified ten guiding principles that can inform the development of Good Machine Learning Practice (GMLP).
These guiding principles are intended to help promote safe, effective and high-quality medical devices that use artificial intelligence and machine learning (AI/ML).
There is a great deal of interest with these technologies in the medical field, especially with the design and operation of medical devices. This is to the extent that regulatory guidance is required and the opportunity has arisen for a transatlantic protocol to be fashioned between three national regulatory agencies.
Artificial intelligence and machine learning technologies have the potential to transform healthcare. This includes processes to derive new insights from the vast amount of data generated during the delivery of healthcare. This type of processing is centred on software algorithms that have the capability to learn from real-world use.
The types of data captured can sometimes be utilised to improve a product’s performance. However, given that valuable patient data is being used and where errors with development could lead to patient harm, it is important that manufacturers are ‘guided’ towards adopting best practices.
The practices include ensuing that the manufacturer has an in-depth understanding of a model’s intended integration into clinical workflow. This means that the desired benefits and associated patient risks must be identified upfront.
It is also important that good software engineering practices, data quality assurance, data management, and robust cybersecurity practices are each in place. To ensure these practices are embedded the use of risk management from the design process is required. The model design should be suited to the available data and support the active mitigation of known risks, like overfitting, performance degradation, and security risks.
With the clinical study, protocols should ensure that the relevant characteristics of the intended patient population (for example, in terms of age, gender, sex, race, and ethnicity), use, and measurement inputs are sufficiently represented in a sample of adequate size. Through this, the results can be reasonably generalised to the population of interest.
For effective machine learning, effective training data sets need to be selected and maintained. Also required are reference datasets based on clinically relevant and well characterised data.
Ensuing that the intended user of the medical device can actually use the device safely. As an example, where the model has a “human in the loop,” human factors considerations and the human interpretability of the model outputs need to be addressed with emphasis on the performance of the Human-AI team, rather than just the performance of the model in isolation.
These types of considerations can lead to better designed medical devices, and ones that are safer for patients.
