In 2024, the MHRA, U.S. FDA, and Health Canada published guiding principles for transparency for machine learning-enabled medical devices (MLMDs). These principles build upon the joint Good Machine Learning Practice (GMLP) principles — especially:
- GMLP Principle 7: focus is placed on the performance of the human–AI team
- GMLP Principle 9: users are provided clear, essential information
While these principles promote transparency for MLMDs, transparency is good practice to consider for all medical devices. See the IMDRF GMLP learning path on this site for the full ten principles; this article focuses on the transparency framework.
Defining transparency
In this document:
- Transparency describes the degree to which appropriate information about an MLMD — including its intended use, development, performance, and, when available, logic — is clearly communicated to relevant audiences
- Logic refers to information about how an output or result was reached or the basis for a decision or action
- Explainability refers to the degree to which this logic can be explained in a way a person can understand
Logic and explainability are aspects of transparency.
Effective transparency ensures information that could impact risks and patient outcomes is communicated; considers what the intended user needs and the context of use; uses appropriate media, timing, and strategies; and relies on a holistic understanding of users, environments, and workflows.
Human-centred design
Human-centred design is an iterative process that addresses the whole user experience and involves relevant parties throughout design and development. This approach can help develop MLMDs with a high degree of transparency, validate transparency, and ensure users have all device-related information they need.
Human-centred design means taking a perspective that is iterative, addresses the whole user experience, and involves relevant parties throughout design and development — with responsive and iterative design, validation, monitoring, and communications.
For Hong Kong clinicians, this means transparency materials should be tested with users like you — not only written for regulators or engineers.
The six-element framework
The guiding principles consider who, why, what, where, when, and how:
Who: relevant audiences
Transparency is relevant to those who use the device (healthcare professionals, patients, caregivers), those who receive healthcare with the device (patients), and additional parties who make decisions about the device to support patient outcomes (support staff, administrators, payers, governing bodies).
Why: motivation
Transparency is essential to patient-centred care and for the safety and effectiveness of a device. It helps parties identify and evaluate risks and benefits, detect and investigate errors or performance decline, promote health equity through bias identification, and build fluency, trust, and confidence in MLMD use.
A clear intended use supports understanding whether the device is intended to inform or replace the judgment of a healthcare provider.
What: relevant information
Appropriate information varies by MLMD but good practice includes:
- medical purpose, diseases or conditions addressed, intended users, environments, and target populations
- how the device fits in the healthcare workflow, including intended inputs, outputs, and impact on clinical decisions
- device performance, benefits, risks, and risk management activities (e.g. bias management)
- the logic of the model, when available and understandable
- limitations: known biases, failure modes, confidence intervals, data gaps, contraindications
- how safety and effectiveness are maintained across the lifecycle — including monitoring, change management, and local validation
This aligns with GMLP Principle 9 — users receive clear, essential information — and supports GMLP Principle 7 by enabling assessment of human–AI team performance in real workflow.
Where: placement of information
Device information is available through the user interface — all elements the user interacts with (display, packaging, labelling, alarms, training, controls). Optimise the software user interface so information is responsive, personalised, adaptive, and available through multiple modalities (text, audio, video, alerts, diagrams, safeguards, document libraries).
When: timing of communication
Consider information needs at each stage of the total product lifecycle. Provide timely notifications when the device is updated or new information is discovered. Targeted information — on-screen instructions or warnings — may be appropriate at specific workflow stages or upon specific triggers.
How: methods to support transparency
Apply human-centred design principles: provide the appropriate level of detail for the intended audience, arrange content to support informed decisions, and use plain language or technical language as appropriate for the user group.
Practical questions for Hong Kong healthcare settings
Before adopting an MLMD, ask:
- Who was transparency designed for — clinicians in your specialty and setting?
- What limitations, biases, and data gaps are disclosed for your patient population?
- Where do warnings and performance information appear in the workflow you will use?
- When will you be notified of model updates or newly discovered risks?
- Does documentation support human–AI team performance (GMLP 7), not just algorithm accuracy?