Morrison & Foerster LLP

01/21/2025 | News release | Distributed by Public on 01/21/2025 17:06

FDA Draft Guidance on Artificial Intelligence-Enabled Device Software Functions

This post is part of MoFo's 2025 Intersection of AI and Life Sciences blog series. In this blog series, we explore how artificial intelligence is revolutionizing research, innovation, and patient care in the life sciences. Stay tuned for expert insights regarding the impact of AI on intellectual property, licensing, contracts, regulatory policy, enforcement, privacy, and venture markets in life sciences.

On January 6, 2025, FDA released a draft guidance entitled Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. In accordance with FDA's total product life cycle (TPLC) approach to the oversight of medical devices, the draft guidance outlines marketing submission content recommendations for devices that include at least one artificial intelligence (AI)-enabled device software function, as well as recommendations for the design, development, deployment, and maintenance of such devices. The guidance also provides a recommended approach to transparency and bias for these devices, which FDA encourages developers to incorporate from device design through decommission.

The guidance encourages early engagement with FDA, especially when the device involves new and emerging technology or when device validation involves novel methods.

Comments and suggestions regarding the draft guidance must be submitted by April 7, 2025. If you would like assistance in making a comment or suggestion, please reach out to the FDA + Healthcare Regulatory and Compliance team at Morrison Foerster.

Information That Sponsors Should Include in a Submission for an AI-Enabled Device[1]

Device Description

The device description should provide information that helps FDA understand the characteristics of the AI-enabled devices, such as device inputs and device outputs, how AI is used to achieve the device's intended use, the level and type of training that intended users have and/or will receive, and any calibration and/or configuration procedures that must be regularly performed by users. Additionally, sponsors should include a description of all configurable elements (i.e., alert thresholds), as well as whether the users who make configuration decisions require any qualifications or training.

User Interface Information

Sponsors should include information about the user interface and device workflow, including how information is presented to users. This can be conveyed through different methods, such as photographs, a written description, example reports, and recorded video.

Labeling Information

An AI-enabled device's labeling should explain many pieces of information to the user at an appropriate reading level, such as how AI is used to achieve a device's intended use, what the model output means, the degree of automation that the device exhibits, and installation and implementation instructions. The labeling should also describe the model architecture, model development data (including demographic distributions), performance data (including primary endpoints of the validation study), device performance metrics across important subgroups, performance monitoring, and known limitations.

Risk Management File

Sponsors should provide, as part of a "Risk Management File," a risk management plan, including a risk assessment.

Training and Testing Data Information

Sponsors should include in a marketing submission information on data collection, development and test data independence, reference standards, and representativeness. The draft provides a template for organizing this information that lists specific descriptions and information that should be provided. If there are any differences in the data management between the development phase and the validation phase, the sponsor should explain the differences and justify them.

Model Description and Development

Sponsors should include descriptions of each model used as part of AI-enabled devices, including any quality control criteria. Additionally, sponsors should include a description of how models were trained and an explanation of any pre-trained models that were used.

Validation Testing

Sponsors should include validation testing that provides objective information to characterize the model's performance. The draft discusses both performance validation and human factors validation (or, alternatively, an evaluation of usability; see Appendix D). Sponsors should include information regarding all study protocols, including study design and analysis details, as well as study results.

Performance Monitoring Plans

If a sponsor elects to employ proactive performance monitoring as a means of risk control and to provide safety and effectiveness assurances, the sponsor should include information regarding their performance monitoring plans.

Cybersecurity

Sponsors should include cybersecurity controls and security risk management relevant to the AI components or features.

Public Submission Summary

For most marketing authorization decisions, sponsors are required to provide public submission summaries. To support transparency, sponsors of AI-enabled devices should include in this summary specific information about device characteristics, such as descriptions of the statistical confidence level of predictions and descriptions of the development and validation datasets. Sponsors are encouraged to use model cards to organize this information (see Appendix E).

Transparency

Appendix B contains recommendations for developing a transparent device around users, starting at the design phase of the TPLC. Reflecting the 2024 Joint Guiding Principles, the draft guidance instructs sponsors to take a holistic approach to transparency, emphasizing that transparency should be considered during device design and throughout the full continuum of implementation. The draft includes examples of relevant considerations. Appendices E and F include examples of how to communicate information about a device's AI model and the limitations of that device, based on FDA's "user-centered research."

Performance Validation Considerations

Appendix C lists recommendations for clinical performance validation, such as pre-specifying study protocols and statistical analysis plans, as well as providing examples of when to conduct a precision study. This section also calls attention to scenarios leading to potential bias.

[1] This section highlights examples of information that sponsors should include in a submission for an AI-Enabled Device. For a complete list, see U.S. Food & Drug Admin., Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations: Draft Guidance for Industry and Food and Drug Administration Staff (2025).