AI in medical devices · Medium · 13 min lesson

FDA approach to AI-enabled devices: submission content, PCCP and transparency

FDA draft guidance on AI-enabled device software functions (January 2025), final PCCP guidance (December 2024), section 515C of the FD&C Act, transparency principles (June 2024)

What you will learn

Map the content FDA recommends in a marketing submission for an AI-enabled device software function and state the status of each guidance document.
Describe the three elements of a predetermined change control plan and decide whether a planned model change fits one.
Apply the transparency principles to labeling, the user interface and the public summary, and spot gaps in a draft record.

Key takeaways

FDA uses a total product life cycle approach for AI-enabled devices; the January 2025 lifecycle guidance is a draft and nonbinding, the December 2024 PCCP guidance is final.
The draft recommends submission content in order: device description, user interface and labeling, risk assessment, data management, model description, validation, monitoring, cybersecurity and a public summary.
A PCCP rests on section 515C of the FD&C Act and has three elements: description of modifications, modification protocol and impact assessment.
A PCCP is optional and covers only a limited number of specific, verifiable modifications; anything outside it still needs the normal change decision.
Transparency means telling the right audience what the device does, how well, for whom and what changes, through labeling, the interface and the public summary.

Free practice question

Halcyon Imaging is preparing a marketing submission for TriageCT. A colleague proposes to wait until the lifecycle guidance for AI-enabled device software functions is final before organizing any evidence, because it was issued as a draft in January 2025.

How should the team treat the draft guidance?

Use it as FDA's current thinking, cite it as a draft and record reasons for any deviation.

Correct. The draft is nonbinding yet shows what reviewers expect, so planning to it while stating its status is sound under the lifecycle approach in section III.

Ignore it until final, since drafts have no value.

Not quite. Waiting loses time because the draft section list from V to XIII already shows the evidence reviewers expect.

Treat it as binding law that overrides the quality system.

Not quite. A draft is marked not for implementation and is nonbinding, so it cannot override ISO 13485 processes or regulation.

Use only the appendices and skip the main sections.

Not quite. The appendices support the main sections V to XIII and do not replace them, so skipping the body leaves the submission incomplete. This follows the FDA guidance and section 515C of the FD&C Act.

Key terms

AI-DSF: A device software function that implements one or more AI models to achieve its intended purpose, as the FDA draft guidance uses the term.
PCCP: A plan, authorized with the marketing submission under FD&C Act section 515C, describing planned modifications and how they will be validated and implemented.
total product life cycle: FDA's approach of expecting evidence and controls from design through decommissioning, not only at the time of submission.
model card: A structured summary of a model's technical characteristics, data, performance and limits, offered as a template in the FDA draft guidance appendices.
data drift: A change over time in the data a model sees or in its performance, which the draft guidance asks sponsors to address.
GMLP: Ten guiding principles for good machine learning practice identified in 2021 by FDA, Health Canada and the MHRA.

Official sources

Go further

The full lesson has narrated slides, more scenario questions with explained answers and a simulation.

Open the Academy Watch on YouTube

Illustrative example. Educational only, not legal advice.