Pathway card
From a locked model to planned change: documenting model updates of a medical device
How can an AI-enabled medical device change its model without every update becoming a new regulatory question?
A pathway card shows where a practice starts, where it can get to, what does not change on the way, and the steps from one to the other. The path from before to after
The short answer
By describing the planned changes, the method to make and validate them and their impact before the device is authorised: in the USA as a predetermined change control plan (PCCP), in the EU as changes the provider pre-determines and documents in the technical file at the initial conformity assessment (AI Act Article 43(4)). Changes outside that plan remain changes that need their own assessment.
For: Regulatory affairs, software and documentation teams of AI-enabled medical devices
Key points
- FDA’s final PCCP guidance (December 2024, revised August 2025) asks for a description of modifications, a modification protocol and an impact assessment.
- The MDR has no PCCP; the AI Act exempts pre-determined, documented changes from substantial modification, the MDR change rules still apply.
- Changes must stay within the intended use or intended purpose.
- Every change under a plan still goes through design controls, verification and validation of the whole device.
- Model, software and device version must stay traceable to each other.
Before and after
Before
Locked model, changes case by case
- The model is frozen at authorisation.
- Each retraining raises the question of a new submission or assessment.
- The technical file explains each change afterwards.
After
Planned change within documented bounds
- The planned changes, the method and the impact are described before authorisation.
- Changes inside the plan follow its protocol and are recorded.
- Model, software and device versions are linked in the file.
Side by side
Each row names how far the control is as of the review date.
| Aspect | Before | After | Maturity |
|---|---|---|---|
| Planned model updates (US) | Each significant change raises the question of a new 510(k), De Novo or PMA supplement. | Changes inside the authorised PCCP are made under its protocol without a new submission. | Building block available FDA final guidance, revised August 2025 |
| Planned model updates (EU) | Every change is checked against the notified body’s change rules. | Changes pre-determined and documented at the initial assessment are not a substantial modification under the AI Act. | Building block available AI Act Art. 43(4); MDR change rules still apply |
| Version traceability | Model files are versioned in the data science tools only. | Model, software and device versions are linked in configuration management and the technical file. | Product-specific Tool support varies |
| Lifecycle recommendations (US) | No single FDA document on the whole AI device lifecycle. | The FDA lifecycle draft guidance describes documentation across the lifecycle. | Draft or proposal Draft guidance of January 7, 2025 |
Applicable law
The after rests on these rules and guidance documents (status as of November 10, 2026):
2024-12-04FDA final guidance on PCCPs for AI-enabled device software functions (revised August 18, 2025)Show the details
The law
FDA final guidance on PCCPs for AI-enabled device software functions (revised August 18, 2025)
Section 515C FD&C Act; recommendations on description of modifications, modification protocol and impact assessment.
What it means here
In the USA, a PCCP authorised with the device lets planned changes be made without a new submission.
2028-08-02AI Act Article 43(4), applicable to devices from August 2, 2028 (Regulation (EU) 2026/1744)Show the details
The law
AI Act Article 43(4), applicable to devices from August 2, 2028 (Regulation (EU) 2026/1744)
Pre-determined changes of a high-risk AI system that continues to learn, documented in Annex IV point 2(f), are not a substantial modification.
What it means here
In the EU, pre-determined changes must be in the technical file at the initial assessment; MDR change rules apply alongside.
Guidance is not law; the FDA lifecycle guidance of January 2025 is still a draft.
The path from before to after
Each step names who does it and what it leaves as evidence.
Bound the change
Name which model changes are planned (e.g. retraining on new data of the same population) and which are out of scope.- Who
- Regulatory affairs with product management
- Evidence
- Description of modifications
Write the protocol
Describe data management, retraining, performance evaluation and the update procedure, with acceptance criteria.- Who
- Data science and software engineering
- Evidence
- Modification protocol
Assess the impact
Assess benefits and risks of each planned change in the risk management file.- Who
- Risk management
- Evidence
- Impact assessment, updated risk file
Put it in the file
Record the plan in the technical documentation (EU: Annex IV point 2(f)) and the labelling notes the device has a plan.- Who
- Technical writers
- Evidence
- File sections, labelling text
Change under the plan
Implement each change through design controls and verify and validate the whole device.- Who
- Engineering with quality
- Evidence
- Change record with model, software and device version
Limits and residual risk
A plan only covers what it names. Unforeseen changes, a new population or a new intended purpose need their own assessment. Status of the guidance documents as of November 10, 2026; check the current versions.
The context
Why model changes are a documentation problem
A model that is retrained on new data is a different model. For a medical device that raises a question at each update: is this a change that needs a new submission or a new assessment? Without a plan, the answer is found case by case — and the technical file has to tell the story afterwards.
The US route: a predetermined change control plan
The FDA's final guidance on PCCPs for AI-enabled device software functions (December 4, 2024, revised August 18, 2025) describes three parts: a description of modifications, a modification protocol (data, retraining, performance evaluation, update procedures) and an impact assessment. Once authorised with the device in a 510(k), De Novo or PMA, changes inside the plan are made without a new marketing submission. The guiding principles of FDA, Health Canada and the MHRA (October 2023) add: focused and bounded, risk-based, evidence-based, transparent, lifecycle perspective.
The EU route: pre-determined changes
The MDR and IVDR have no PCCP instrument; changes are assessed under the notified body's change procedures. The AI Act adds Article 43(4): for high-risk AI systems that continue to learn, changes the provider pre-determined at the initial conformity assessment and recorded in the technical documentation (Annex IV point 2(f)) are not a substantial modification. MDCG 2025-6 stresses that substantial modification is an AI Act concept of its own. Both checks run in parallel.
What the file must hold
The path below lists what each step leaves in the file. The core is a link that most files lack: model version → software version → device version, each with its validation evidence (see the card on model, software and device versions).
Questions readers ask next
- Is there a PCCP in the EU?
- Not under that name. The AI Act lets pre-determined changes documented at the initial conformity assessment escape the substantial-modification rule (Article 43(4)); the MDR and IVDR change rules of the notified body still apply.
- Does a PCCP remove design controls?
- No. Each change made under an authorised PCCP still goes through design controls, including verification and validation of the whole device.
- Can a PCCP change the intended use?
- No. Planned changes have to stay within the intended use of the authorised device.
Sources
- Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (final guidance, revised August 18, 2025) — U.S. Food and Drug Administration, 4 December 2024Final guidance: Guidance published in its final form by an authority or an official body. Guidance explains the law; it is not law itself.
- Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices: Guiding Principles — FDA, Health Canada and MHRA, 24 October 2023Final guidance: Guidance published in its final form by an authority or an official body. Guidance explains the law; it is not law itself.
- Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (draft guidance) — U.S. Food and Drug Administration, 7 January 2025Draft guidance: A draft for consultation. It may still change and must not be treated as the final text.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act) — EUR-Lex, 12 July 2024Law in force: Binding legislation that applies as of the review date.
- MDCG 2025-6 / AIB 2025-1: FAQ on the interplay between the MDR/IVDR and the AI Act — Medical Device Coordination Group and Artificial Intelligence Board, 19 June 2025Final guidance: Guidance published in its final form by an authority or an official body. Guidance explains the law; it is not law itself.
- Regulation (EU) 2017/745 on medical devices (MDR) — EUR-Lex, 5 May 2017Law in force: Binding legislation that applies as of the review date.
Review log and changes
Every context card is checked against its sources before it is published, and again whenever it changes; the date under the byline is the last review. Corrections (something was wrong) and additions (something was missing) are logged below with date and time (Berlin time). Typos, formatting and link fixes are not listed.
Reviewed
No corrections or additions since publication.