The integration of machine learning (ML) into medical devices is transforming the healthcare landscape, particularly in the area of precision medicine and personalized care. Technologies that were concepts for decades are now coming to the market with capabilities that exceed the imaginations of science fiction writers. From augmented intelligence to support informed clinical decisions through to processing extended datasets comprising of traditional inputs and currently untapped and individual information, the impact of these advancements promises to improve patient outcomes, streamline workflows for healthcare providers, and innovative solutions for complex medical challenges.
Loading...
Regulators have been quick to recognise the potential – and also the novel risks – of ML in medicine, bringing into question whether current regulations adequately address performance and safety of this technology. Although fundamentally these technologies are software and regulation of software as a medical device is established, the adaptability and autonomy of ML brings in risks such as unintentional bias and outcome drift, questioning long term performance and the consistency of the clinical benefit and utility as the algorithm develops; potentially improving outcomes, but also with the risk of degradation.
In 2021, regulators from three major jurisdictions –Canada, UK and US - joined forces to publish 10 guiding principles for ML practice for medical device development. Last month, the same team produced an update relating to guiding principles for transparency for machine learning-enabled medical devices. These new guidelines aim to standardize how critical information about machine learning-enabled medical devices (MLMDs) is communicated to ensure patient safety and effective clinical outcomes through clear dissemination of essential information that could impact risks and patient outcomes. This initiative marks a significant step toward a unified approach to transparency in MLMDs, providing guidance to developers on best transparency practices.
Transparency
The newly published document states key aspects of transparency are logic and explainability, which define the degree to which appropriate information about a machine learning-enabled medical device is clearly communicated to relevant audiences. For the FDA and its collaborators, the key aspects of effective transparency are:
- Communicating risks and patient outcomes.
- Considering specific needs of intended users and the context of the use of the information.
- Using the best media, timing, and strategies.
- Understanding users, their environments, and their workflows.
- Emphasizing "Human-Centered Design.
The guiding principles
The publication considers the following points crucial for ensuring a high degree of transparency in the functioning of medical devices utilizing AI:
Who - the audience involved with the device.
Transparency should address:
- Users of the device (e.g. health care professionals, patients, or caregivers)
- Those receiving care from the device, e.g. patients.
- Decision-makers supporting patient outcomes (support staff, administrators, etc.)
Why – the motivation for transparency.
Transparency is very important for:
- Patient-centered care, safety, and effectiveness of devices working with AI.
- Understanding complex information, identifying device risks and benefits, and ensuring safe and effective use.
- Helping in error detection, promotes health equity, maintains device safety, and fosters trust and confidence.
What – relevant information to be shared.
Information to consider should be:
- Medical purpose, function, and target conditions.
- Intended users, environments, and populations.
- Device workflow, inputs, outputs, and impact on healthcare decisions.
- Performance, benefits, risks, and risk management strategies.
- Device output logic, development details, and ongoing updates.
- Limitations, biases, and known gaps.
- Safety and effectiveness across the product lifecycle.
Where – placement of the information.
Information should be:
- Accessible via the user interface, including training, physical controls, display elements, packaging, labelling, and alarms.
- Optimized to the user considering adaptive information delivery through various modalities like audio, video, on-screen text, alerts, diagrams, and document libraries.
When – timing.
Timing should be strategic throughout the product lifecycle:
- When acquiring or implementing a device.
- When using the device.
- When the device is updated or new information is discovered.
- When there are high-risk steps / specific triggers.
How – the most appropriate method used.
To make information more accessible, developers should:
- Tailor information to suit the audience's needs.
- Organize content by importance to aid decision-making.
- Use plain language when clarity is paramount, or technical language for clinical specialists.
Impact on the Industry
These new principles set a standard for transparency in machine learning-enabled medical devices. By following these guidelines, developers can comply with regulatory standards and build trust with healthcare professionals and patients.
Get in Touch
Navigating the complexities of medical device certifications, especially with MLMDs, can be challenging. If you have questions about certifications for your machine learning-enabled medical device, contact us. We leverage quality and empower medical device manufacturers to reach compliance and elevate quality standards. For more insights and information on medical device certifications, reach out today to ensure your AI-driven medical devices meet the highest standards of transparency and safety.
DQS Newsletter
Stay informed and subscribe to our newsletter!
