What’s Covered?
The draft guidance issued in January 2025 builds on the FDA’s Total Product Lifecycle (TPLC) approach and offers structured recommendations for submitting AI-enabled medical devices, including those using machine learning models. It’s meant to help manufacturers understand what to include in marketing submissions (like 510(k), De Novo, PMA, HDE, or BLA) and to clarify expectations for the design, validation, and oversight of AI-driven software functions.
This is not a standalone compliance checklist—it complements existing FDA guidance (especially the “Premarket Software Guidance”) and other domain-specific guidance. The focus is on risk management, performance validation, fairness across demographics, and transparency over time, particularly for systems that evolve post-deployment.
Key content areas:
- Lifecycle Oversight (TPLC principles): Consistent with Good Machine Learning Practice (GMLP) and transparency goals, the document supports proactive management across the device’s full lifecycle, from development to post-market monitoring.
- Scope Clarification: Applies to all AI-enabled device software functions (AI-DSFs), including Software as a Medical Device (SaMD), Software in a Medical Device (SiMD), and device parts of combination products. AI-DSFs are defined as software functions implementing one or more mathematical models generating predictions or inferences.
- Submission Structure:
- Quality system documentation
- Device and model description
- Labeling and user interface design
- Risk assessment and cybersecurity
- Validation plans and performance benchmarks
- Ongoing device performance monitoring
- Summary tables and transparency tools (e.g., model cards)
- Fairness & Bias: Strong encouragement to assess model performance across subgroups (race, sex, age) and collect evidence showing consistent benefits to all relevant populations.
- Alignment with Consensus Standards: Leverages standards to improve documentation quality and consistency, referencing the FDA’s recognized standards database and encouraging declarations of conformity when applicable.
- Combination Products & Early Dialogue: Strongly recommends early interaction with the lead FDA division when combination products or novel techniques are involved.
- Appendices: Include design checklists, transparency tools, model cards, usability testing, and example 510(k) summaries to streamline preparation.
💡 Why it matters?
This guidance shows how regulators are catching up to AI’s complexity in medical devices. It reflects growing expectations around transparency, continuous validation, and equitable outcomes. For manufacturers, it offers predictability. For the public, it signals a shift toward more trustworthy oversight of AI in healthcare. For the governance community, it illustrates how traditional medical device regulation is adapting to AI’s dynamism.
What’s Missing?
Although detailed, the draft avoids strict requirements for adaptive or continuously learning systems—leaving ambiguity for models that update post-approval. There’s also no robust accountability framework for harm caused by biased or opaque systems. Moreover, while subgroup performance monitoring is encouraged, it stops short of prescribing threshold standards or consequences for systemic disparities. Guidance for post-deployment updates and real-time learning remains vague, especially in cases of autonomous behavior changes over time.
Also missing is a discussion on patient-facing transparency—how users (not just regulators) are informed about changes, limitations, or risks. While it nods to transparency tools, the specifics of how end-users are involved in oversight are underdeveloped.
Best For:
Regulatory teams and developers working on AI-based medical devices in the U.S. who need to prepare marketing submissions. Also relevant for AI governance professionals focused on medical AI, especially those working with SaMD or combination products.
Source Details:
Title: Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (Draft Guidance)
Issuing Agency: U.S. Food and Drug Administration (FDA)
Issued: January 7, 2025 (Draft for public comment)
Applies To: AI-enabled device software functions (AI-DSFs), including SaMD, SiMD, and AI-based device components in combination products
Centers Involved:
- Center for Devices and Radiological Health (CDRH)
- Center for Biologics Evaluation and Research (CBER)
- Center for Drug Evaluation and Research (CDER)
- Office of Combination Products
