Summary
In October 2021, the FDA, Health Canada, and the UK’s MHRA jointly published 10 guiding principles for Good Machine Learning Practice (GMLP) to support safe, effective, and high-quality AI- and machine learning-enabled medical devices. This article walks through all 10 principles and explains why ethics, data integrity, and model transparency form the infrastructure for responsible devices — and how USDM Life Sciences helps you operationalize them.
Machine learning in healthcare presents unprecedented opportunities for diagnosis, treatment planning, and patient care. However, medical device development in particular demands rigorous compliance with regulatory standards to ensure safety, reliability, and effectiveness.
In October 2021, the U.S. Food and Drug Administration (FDA), Health Canada, and the United Kingdom’s Medicines and Healthcare products Regulatory Agency (MHRA) identified 10 guiding principles to help promote safe, effective, and high-quality medical devices that use artificial intelligence (AI) and machine learning.
The Vision for GMLP Guiding Principles
We already see AI and machine learning transforming healthcare. Brilliant insights are derived from an unprecedented data revolution in life sciences and software algorithms are trained on real-world use cases.
The 10 guiding principles for GMLP lay the foundation for medical device development. They also identify areas where the International Medical Device Regulators Forum (IMDRF), international standards organizations, and other collaborative bodies could advance GMLP.
GMLP guiding principles help medical device developers to:
- Adopt good practices that have been proven in other sectors
- Tailor practices from other sectors and apply them to medical technology and the healthcare sector
- Create new practices specific to medical technology and the healthcare sector
Why it matters: GMLP is not a single regulation you can certify against — it is a shared set of expectations that spans engineering, data science, clinical, and quality teams. Treating these 10 principles as design inputs from day one is far cheaper than retrofitting them after a model is in production.
So what are the 10 guiding principles for GMLP? Let’s take a look.
- Multi-disciplinary expertise is leveraged throughout the total product life cycle. Cultivate an in-depth understanding of the desired benefits and associated patient risks to help ensure that ML-enabled medical devices are safe and effective. Aligning engineering, clinical, quality, and data science roles — the kind of cross-functional model behind a modern agentic AI team — keeps these perspectives connected across the lifecycle.
- Good software engineering and security practices are implemented. Incorporate good software engineering practices, data quality assurance, data management, and robust cybersecurity practices into model design.
- Clinical study participants and datasets are representative of the intended patient population. Ensure that relevant characteristics of the intended patient population, use, and measurement inputs are sufficiently represented in an adequate sample size so that results can be reasonably generalized to the population of interest.
- Training datasets are independent of test sets. Consider and address all potential sources of dependence, including patient, data acquisition, and site factors.
- Selected reference datasets are based upon best available methods. Ensure that clinically relevant and well characterized data are collected and that the limitations of the reference are understood. When reference data originates with vendors or external partners, evaluating that provenance is part of third-party risk management.
- Model design is tailored to the available data and reflects the intended use of the device. Understand the clinical benefits and risks related to the product to derive clinically meaningful performance goals for testing and to confirm that the product can safely and effectively achieve its intended use.
- Focus is placed on the performance of the Human-AI team. Keep a human in the loop so that human interpretability of model outputs are addressed, rather than just the performance of the model in isolation.
- Testing demonstrates device performance during clinically relevant conditions. Generate device performance information independently of the training dataset and consider the intended patient population, important subgroups, the Human-AI team, measurement inputs, and potential confounding factors. A risk-based testing strategy informed by Computer Software Assurance (CSA) helps focus verification effort where patient risk is highest.
- Users are provided clear, essential information. Ensure that the information is appropriate for the intended audience, such as healthcare providers or patients. Make users aware of device modifications, updates from real-world performance monitoring, and how to communicate product concerns to the developer.
- Deployed models are monitored for performance and re-training risks are managed. Monitor deployed models during real-world use (post-market surveillance) and, when models are trained after deployment, have appropriate controls in place to manage risks.
The Three Pillars Beneath GMLP
The 10 principles describe what to do. Three cross-cutting disciplines determine whether you can do it responsibly and repeatably:
- Ethics — patient safety, privacy, and bias mitigation as non-negotiable design constraints.
- Data integrity — quality, accuracy, and representativeness of the data that trains and validates the model.
- Model transparency & explainability — outputs clinicians can interpret, trust, and act on.
Three More Facets of Machine Learning for Medical Devices
While the 10 guiding principles lay the foundation for GMLP and medical device development, ethics, data integrity, and model transparency are the infrastructure for responsible and reliable medical devices.
Ethical Considerations
Ethics are the ultimate guide in developing and deploying machine learning in medical devices. Developers must prioritize patient safety and privacy; therefore, ethical machine learning practices require transparent data usage and patient data confidentiality. They also help to mitigate biases in machine learning models to prevent unequal treatment outcomes across various patient demographics. Embedding these expectations into formal AI governance and compliance processes turns good intentions into auditable controls.
Ethics, data integrity, and model transparency are the infrastructure for responsible and reliable medical devices.
Data Integrity
The quality, accuracy, and representativeness of the data used to train machine learning models influences its reliability and validity. Comprehensive and diverse datasets that accurately reflect a target population consist of data collected from a wide range of ages, genetic backgrounds, and health conditions. Data curation and preprocessing remove inaccuracies and ensure the data is relevant to the medical conditions the device addresses.
Model Transparency and Explainability
Understanding the decision-making process for medical device development is important for clinical acceptance. Model transparency and explainability ensure that healthcare professionals are able to interpret and trust the outputs of devices powered by machine learning. Techniques like model simplification help to demystify complex algorithms and make them more accessible to clinicians and patients.
How USDM Can Help
Data is the lifeblood of AI, but people are the weak link in responsible AI and machine learning. USDM Life Sciences provides the training and expertise and helps your life sciences organization establish a data governance framework for the integrity and security of your data. If you are early in your journey, an AI readiness assessment is a practical first step toward applying GMLP with confidence.
To learn more about integrating advanced technologies into medical device development, contact us today. Our industry experts will help your organization realize the benefits of machine learning and its responsible use and develop an AI governance framework that will work for your organization.
FAQ: Good Machine Learning Practice (GMLP) for Medical Devices
What is Good Machine Learning Practice (GMLP)?
GMLP is a set of 10 guiding principles published in October 2021 by the U.S. FDA, Health Canada, and the UK’s MHRA to help promote safe, effective, and high-quality medical devices that use artificial intelligence and machine learning. The principles span the total product life cycle, from multi-disciplinary design through deployment monitoring.
Who created the GMLP guiding principles?
The 10 guiding principles were jointly identified by the U.S. Food and Drug Administration (FDA), Health Canada, and the United Kingdom’s Medicines and Healthcare products Regulatory Agency (MHRA). They also point to areas where bodies such as the International Medical Device Regulators Forum (IMDRF) and international standards organizations could advance GMLP further.
How do GMLP principles address bias and patient safety?
Several principles directly target representativeness and safety: ensuring datasets reflect the intended patient population, keeping a human in the loop on the Human-AI team, and monitoring deployed models for performance drift. Ethical practice reinforces this by mitigating biases so treatment outcomes do not vary unfairly across patient demographics.
Why are ethics, data integrity, and model transparency considered separate from the 10 principles?
The 10 principles define what to do across the lifecycle, while ethics, data integrity, and model transparency are the cross-cutting infrastructure that makes responsible execution possible. Without trustworthy data, interpretable models, and clear ethical guardrails, even well-followed principles can produce unreliable or unsafe devices.
How does USDM help with GMLP and AI governance?
USDM Life Sciences provides training and expertise to help organizations establish a data governance framework for the integrity and security of their data, and to build an AI governance framework suited to their needs. Engagements can begin with an AI readiness assessment and extend into responsible AI and regulatory compliance for medical device development.
Ready to apply GMLP to your devices? USDM’s industry experts can help you operationalize the 10 guiding principles, strengthen data integrity, and stand up an AI governance framework that fits your organization. Contact us today to get started.