The MHRA has jointly published new transparency guiding principles for machine learning medical devices (MLMDs).

The latest guidance builds on previous principles for good machine learning practice for medical device development, which the MHRA published with international regulators the FDA and Health Canada in 2021. 

The 2021 guidelines were designed to advance the safety and effectiveness of AI and ML medical devices across the product lifecycle. The new guiding principles are focused on transparency in MLMDs, setting out a series of principles for developers and manufacturers to ensure there is an adequate level of transparency for patients and healthcare professionals when using MLMDs.

Treading the balance between medical devices and less-regulated digital health products is well-known to manufacturers and regulators. AI and ML products present additional challenges in this area, particularly around intended use, change and transparency.

The new guiding principles present a level of best practice for MLMDs, which developers will need to consider when preparing their market access and regulatory strategy for AI/ML products, whether they believe their product is a medical device or not. 

What is “transparency” under the new guidance?

For the purposes of the guidelines, the term ‘transparency’ refers to the way in which details about a MLMD are communicated to the relevant audiences. Transparency is based on the concepts of ‘logic’ and ‘explainability’. Logic is how the MLMD reaches its output. Explainability is the degree to which the logic can be explained to a user.

The MHRA also describe the concept of ‘human-centred design’ as a key aspect of transparency. Human-centred design addresses the user experience, involving the relevant parties in design and development stages.

What are the transparency guidelines?

The guiding principles are designed to be considerations in achieving sound transparency procedures. The guidelines follow a ‘who, why, what, where, when, how’ approach:

• Who (relevant audiences): The concept of transparency is applicable to those who will be using the device (caregivers and healthcare professions), those who receive healthcare with the device (patients), as well as third parties involved in decision-making on a device (such as governing bodies or regulators). 
• Why (motivation): The motivation behind effective transparency is that it prioritises clear communication with users about a MLMD, so that care is patient-centred and risks are managed effectively. Patients can then in turn make informed decisions as part of an equitable healthcare system.
• What (relevant information): The type of information to be shared around MLMDs should include, but is not limited to, a description of the use, performance, benefit, risks, development, safety and limitations across the device’s lifecycle. 
• Where (placement of information): Information being effectively placed will improve the responsiveness, personalisation and user-friendliness of the ML-enabled device. The guidance advises optimizing the software user interface to ensure that the conveyed information is responsive to the user’s needs.
• When (timing): Transparency is crucial at various stages of the use lifecycle of a device, and communication timings should be considered on a case-by-case basis. The guidance suggests providing timely notifications about device updates or new information, along with targeted on-screen instructions or warnings triggered by specific events.
• How (methods used to support transparency): Approaches such as human-centred design will be crucial for transparency. This approach will involve the relevant parties throughout the design and development process; addresses the whole user experience; and will involve tailoring the level of detail in communication to the intended audience, selecting either plain language or technical terminology based on their needs. 

If you would like to discuss any issues relating to digital health products and medical device regulation or market access, please contact a member of our Health Tech team. 

This article was written by Rory Trust, Annalise Slocock and Abigail Cropper.