AI transparency in medical devices is achievable and essential, according to new guidance jointly sponsored by the FDA, Health Canada and the UK Medicines and Healthcare products Regulatory Agency.
The document, published on June 13, updates and expands on the principles outlined by the three bodies in 2021.
The guidelines define transparency in “MLMDs” – machine learning medical devices – as “the extent to which relevant information about an MLMD … is clearly communicated to relevant audiences.”
The guidelines call on developers and marketers of AI in healthcare to fully consider the impact of their technology on patient safety and, in turn, population health. They include six questions to help conduct these pre-screenings.
Q: For whom is transparency in the MLMD relevant?
A. Transparency is relevant for those who use the device, such as healthcare professionals, patients and caregivers, as well as for those who receive healthcare using the device, such as patients, the new directive states. More:
It is also relevant to other parties, including those who make decisions about the device to support patient outcomes, such as support staff, administrators, payers and governing bodies.’
Q: Why is transparency critical to patient-centered care and the safety and effectiveness of a device?
A. The transparent and consistent presentation of information, including known information gaps, can have many benefits. First, transparency “promotes fluency and efficiency in the use of MLMDs.”
In addition, transparency can increase trust in technology and promote the adoption and access to useful technologies.
Q: What information should be shared?
A. The type of information that may be shared will vary across MLMD ranges, the new policy says, with appropriateness depending on the benefits and risks of each MLMD and the needs of the intended users.
It is good practice to provide information that enhances the understanding of the device and its intended use. A clear and accurate description of a device generally includes information about its medical purpose and function, the diseases or conditions it is intended to treat, and the intended users, environments of use and target audiences.
Q: Where is the best place to place this information?
A. Ideally, through the user interface. This should include “all elements of the device that the user sees, hears, and touches.” Examples of the user interface include training, physical controls, displays, packaging, labeling, and alarms.
A good practice is to optimize the use of the software’s interface so that the information it conveys is tailored to the user. User needs can be met using a variety of modalities, including audio, video, on-screen text, alerts, charts, software security measures, and document libraries.
Q: At what point in the implementation of the MLMD should the communication target?
A. “Considering information needs at every stage of the entire product lifecycle can support successful transparency,” the document says. “It can also be helpful to provide timely notifications when the device is updated or changed, or when new information about the device is discovered.”
It may be appropriate to provide targeted information, such as on-screen instructions or warnings, at a specific stage of the workflow, such as high-risk steps, and at specific triggers, such as when certain input or output functions are present.
Q: What is the best way to support the long-term transparency of a particular MLMD?
A. Communicating information about MLMDs requires a holistic understanding of users, environments and workflows. These can be addressed by applying human-centered design principles, the guide says.
Transparency-supporting information is most accessible and usable when it contains the level of detail appropriate for the target audience. In some cases, simple language may be appropriate for understanding and usability. In other cases, technical language may be relevant for specialised clinical users.
Read the entire document.
