Further to the guiding principles on the use of artificial intelligence (AI) and machine learning (ML) technologies jointly published by the U.S. Food and Drug Administration (FDA), Health Canada, and the United Kingdom’s Medicines and Healthcare products Regulatory Agency (i.e., the guiding principles on good ML practice for medical device development (Good ML Principles) issued in 2021 and the guiding principles on predetermined change control plans for ML-enabled medical devices in 2023),1 the agencies recently released another set of guiding principles to specifically promote transparency for ML-enabled medical devices (MLMDs).2 The Transparency Principles provide considerations for companies, such as MLMD manufacturers and developers, to adopt and implement good transparency practices throughout their devices’ product lifecycles—considerations that can enable the safe and effective use of MLMDs in an era marked by the proliferation of ML technologies. The Transparency Principles reflect the FDA’s ongoing, international efforts to help “optimize human-centered transparency by supporting predictability and harmonization across jurisdictions,” which, in turn, can impact the adoption of these devices and facilitate innovation in AI/ML technologies in the healthcare industry.3
The Transparency Principles emphasize the effective communication of relevant information about a MLMD to users, such as the intended use(s), device development, device performance, and method for reaching the output or result, or basis for a decision or action (i.e., the “logic”).4 They consider:
The Transparency Principles expand upon the Good ML Principles, which provide a broad foundation for developing safe, effective, and high-quality medical devices that use AI/ML technologies.6 The Transparency Principles especially offer further clarity for two Good ML Principles: i) focus on the performance of the human-AI team (Good ML Principle 7); and ii) provide users clear, essential information (Good ML Principle 9).7
The Good ML Principles and Transparency Principles are summarized in Table 1 and Table 2, respectively.
Table 1. Good ML Principles8
Principle No. | Principle |
Principle 1 | Multi-disciplinary expertise is leveraged throughout the total product lifecycle |
Principle 2 | Good software engineering and security practices are implemented |
Principle 3 | Clinical study participants and data sets are representative of the intended patient population |
Principle 4 | Training data sets are independent of test sets |
Principle 5 | Selected reference datasets are based upon best available methods |
Principle 6 | Model design is tailored to the available data and reflects the intended use of the device |
Principle 7 | Focus is placed on the performance of the human-AI team |
Principle 8 | Testing demonstrates device performance during clinically relevant conditions |
Principle 9 | Users are provided clear, essential information |
Principle 10 | Deployed models are monitored for performance and re-training risks are managed |
Table 2. Transparency Principles9
Guiding Principle | Description |
Who | Transparency is relevant to all parties involved in a patient’s healthcare, including those intended to:
|
Why | Transparency supports:
|
What | The type of relevant information depends on the benefits and risks of each MLMD, as well as the needs of the intended users. Relevant information may include:
|
Where | Information about the device should be accessible through the user interface (e.g., training, physical controls, display elements, packaging, labeling, and alarms), including the software user interface (e.g., on-screen instructions, and warnings). Maximizing the utility of the software interface can:
|
When | The timing of communication depends on the stage of the product lifecycle, such as:
|
How | The communication of information requires a holistic understanding of the users, use environments, and workflows. This may be addressed by applying human-centered design principles, which address the whole user experience and involve relevant parties throughout design and development. |
The FDA is inviting public comments to the Transparency Principles through the public docket (FDA-2019-N-1185).
Contact Us
For questions regarding regulatory strategy and recent developments in FDA regulation, including the FDA’s guiding principles and industry guidance for AI/ML-enabled medical devices, please contact Eva Yin, Jonathan Trinh, or any member of Wilson Sonsini’s FDA regulatory, healthcare, and consumer products practice.
[1] See U.S. Food and Drug Admin et al., Good Machine Learning Practice for Medical Device Development: Guiding Principles (Oct. 2021), https://www.fda.gov/media/153486/download (hereinafter “Good ML Principles”); U.S. Food and Drug Admin et al., Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices: Guiding Principles (Oct. 2023), https://www.fda.gov/media/173206/download?attachment.
[2] See U.S. Food and Drug Admin et al., Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles (June 2024), https://www.fda.gov/media/179269/download?attachment (hereinafter “Transparency Principles”); U.S. Food and Drug Admin., Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles (updated June 13, 2024), https://www.fda.gov/medical-devices/software-medical-device-samd/transparency-machine-learning-enabled-medical-devices-guiding-principles.
[3] U.S. Food and Drug Admin., Press Release, CDRH Issues Guiding Principles for Transparency of Machine Learning-Enabled Medical Devices (June 13, 2024), https://www.fda.gov/medical-devices/medical-devices-news-and-events/cdrh-issues-guiding-principles-transparency-machine-learning-enabled-medical-devices.
[4] Transparency Principles, at 1.
[6] Id. at 1; see Good ML Principles, at 1; U.S. Food and Drug Admin., Good Machine Learning Practice for Medical Device Development: Guiding Principles (updated Oct. 27, 2021), https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles.