Short answer
Under the EU MDR, a change to a certified AI/ML device needs Notified Body approval when it could affect safety and performance or the conditions of use (Annex IX, section 4.10, for Annex IX certificates); your Notified Body’s change procedure sets out how to notify. Retraining that keeps validated performance within pre-specified thresholds on representative data is easier to defend as non-significant, while changes that shift performance metrics, patient populations or subgroup outputs usually need notification. The MDR has no equivalent of the FDA’s PCCP, which authorises a planned change envelope in advance. Singapore’s HSA handles machine-learning updates through the GN-21 flowcharts and its optional Change Management Program.
Why change control is the central AI/ML compliance problem
Hardware medical devices do not spontaneously change their behaviour after certification. AI/ML SaMD does — or at least, it can. Model retraining on new data, drift correction, architecture updates, new training datasets, and threshold adjustments all have the potential to alter the clinical output of a certified device. Regulators wrote change control frameworks before AI was commonplace, and the criteria do not always map cleanly onto software.
The practical consequence is that AI/ML SaMD teams face a recurring compliance question with every release cycle: is this change significant enough to require regulatory notification — and if so, to whom, in which market, with what documentation? Getting this wrong in either direction is costly: notifying unnecessarily delays releases; failing to notify when required is a post-market non-conformity waiting to be found in the next surveillance audit.
The EU MDR framework: the annexes, your NB — and where MDCG 2020-3 fits
For devices certified under the MDR, the duty to tell the Notified Body about changes comes from the conformity assessment annexes. For software certified under Annex IX, the manufacturer must inform the NB of any plan for substantial changes to the quality management system or the device range covered (Section 2.4), and changes to an approved device require NB approval where they could affect the safety and performance of the device or the conditions prescribed for its use (Section 4.10). Each NB translates this into its own change notification procedure, which forms part of your agreement with it.
MDCG 2020-3 Rev.1 (May 2023) — the guidance most often quoted on "significant changes" — has a narrower scope: it explains what counts as a significant change in design or intended purpose for legacy devices still placed on the market under MDD/AIMDD certificates (MDR Article 120). It states that it does not apply to MDR-certified devices. Its flowcharts, including Chart C on software changes, are nevertheless a useful reference for structuring your own significance criteria — but the examples were drafted largely with conventional devices in mind.
Applying these criteria to AI/ML changes
For AI/ML SaMD, the key question is whether a change could affect the clinical performance — the accuracy, sensitivity, specificity, or clinical decision output — of the device, or the conditions of its use, beyond what was validated and certified. The following categories require particular attention:
- Intended use or indications: Any extension to a new condition, patient population, anatomical site, or clinical setting is significant by definition. Adding a new diagnostic target to an imaging AI — even if the underlying architecture is unchanged — is a new intended purpose.
- Algorithm changes affecting clinical output: Changes to model weights, architecture, inference thresholds, or post-processing logic that produce measurably different outputs for the same inputs require assessment. The test is whether the clinical output could differ — not whether it typically does in practice.
- Training data scope: Retraining on data from a substantially different patient population (different demographics, different scanner types, different clinical settings) is a significant change if the original validation was not designed to cover that population.
- Input data format or type: Changes to what the model receives as input — new imaging modalities, additional laboratory parameters, different data preprocessing — affect what the algorithm is doing and require re-evaluation.
- Output interpretation: Changes to how clinical users interact with outputs — new risk scores, changed threshold displays, different flagging logic — may not change the model but can change clinical use patterns.
The decision tree: does this change require NB notification?
Comparison: EU MDR vs HSA CMP vs FDA PCCP
The three major frameworks for AI/ML SaMD change control share the same underlying concern — ensuring that post-market changes do not degrade clinical safety or performance — but differ significantly in architecture and in how much they can accommodate the continuous improvement reality of AI products.
| Dimension | EU MDR (+ AI Act from 2028) | Singapore HSA (GN-21 R7, GN-37) | FDA (PCCP framework) |
|---|---|---|---|
| Framework type | Reactive — assess each change against documented criteria and your NB's procedure | Categorised change notification; optional Change Management Program (CMP) with pre-specified changes | Prospective — change plan authorised as part of the marketing submission |
| AI-specific guidance | No MDR guidance dedicated to AI changes (MDCG 2020-3 covers legacy devices only); AI Act Article 43(4) recognises pre-determined changes for high-risk AI medical devices from 2 August 2028 | Flowchart 2.5 in GN-21 for machine learning-enabled devices; GN-37 R1 on the CMP for SaMD incl. ML-enabled SaMD; GL-04 R4 (Dec 2025) | Final guidance on PCCPs for AI-enabled device software functions (first issued Dec 2024; current version dated Aug 2025) |
| Who approves changes | Notified Body (changes that could affect safety and performance or conditions of use); manufacturer's documented assessment (other changes) | HSA before implementation (Technical, Review and Administrative changes); Notification changes can be implemented once HSA acknowledges the submission; changes within an approved CMP via the manufacturer's QMS | FDA authorises the PCCP; modifications implemented according to it need no new marketing submission |
| Algorithm retrain | Needs NB approval if it could affect safety and performance — e.g. performance outside validated bounds or a new population | Category determined via Flowchart 2.5; can be pre-specified under an approved CMP | Permitted without new submission if within the authorised PCCP |
| Speed for compliant updates | Depends on NB review time for each notified change | Defined categories make the route predictable; CMP allows pre-approved updates to be implemented directly | Fast — changes within PCCP scope need no new submission |
| Documentation burden | High — every change requires documented significance assessment regardless of outcome | Moderate — tier determines required documentation | Front-loaded — PCCP preparation is intensive; ongoing changes are light |
Malaysia is moving the same way: MDA/GD/0072, which applies once MeDC@St 3.0 launches, has a software flowchart and a PCCP approved at registration. See our Malaysia change management guide.
What this means for EU-first AI/ML SaMD companies
The absence of a PCCP equivalent in the MDR is a genuine operational constraint for AI/ML SaMD companies with continuous improvement cycles. Every algorithm change that could affect safety and performance — even a well-validated, clearly beneficial retrain — needs NB approval, with associated delays. The MDR today has no formal mechanism to pre-approve a defined set of future changes the way FDA allows.
Two developments will change this picture. The AI Act (Article 43(4)) provides that changes to a high-risk AI system that the provider pre-determined at the initial conformity assessment, and described in the technical documentation (Annex IV, point 2(f)), are not a substantial modification; for AI medical devices this applies from 2 August 2028. And the Commission's MDR/IVDR proposal COM(2025) 1023 would sort post-certification changes into tiers and let the NB and manufacturer agree a predetermined change control plan where appropriate — but that is a proposal still under negotiation, not law.
The practical responses available to EU-certified AI/ML companies today are:
- Pre-specify performance thresholds in the original technical file: In our experience, this is the single most important step. Thresholds documented and assessed at certification give future retraining a pre-agreed baseline instead of a case-by-case argument with the NB.
- Design your change control SOP around documented significance criteria: Base them on the MDR annex wording and your NB's procedure, using MDCG 2020-3's flowcharts as a structure. A repeatable assessment — run before every release — produces the audit trail that demonstrates compliant change management during NB surveillance audits.
- Batch non-significant changes: Minor updates that are individually non-significant can be implemented without NB approval. Accumulating multiple small improvements before a version release can reduce regulatory overhead.
- Engage the NB early on borderline changes: Many NBs offer a structured dialogue or pre-notification query process. This avoids surprise objections and can speed up review when notification is required.
The change control record: what to document
Whether a change is assessed as significant or not, the change control record must be sufficient to demonstrate the assessment was performed rigorously. For AI/ML changes under EU MDR, the record should include the items below. If you need someone to run these assessments release by release, this is typical work for part-time QA/RA support after launch.
- Change description: Precise specification of what changed — model architecture, training data, hyperparameters, inference logic, post-processing, thresholds. Version control references (git commits, dataset versions) should be linked.
- Significance assessment: Documented evaluation against each criterion in your change control SOP (for example structured along the MDCG 2020-3 flowcharts) with explicit conclusion — significant or not significant — and the justification for that conclusion.
- Performance comparison: Pre- and post-change performance metrics on the validation dataset, compared against pre-specified acceptance thresholds. If thresholds were not pre-specified, explain why and document the alternative basis for assessment.
- Risk assessment update: ISO 14971 risk assessment reviewed and updated if the change introduces new hazards or modifies existing risk controls. Reference the specific risk file revision.
- IEC 62304 records: Software change request, change impact analysis, and verification records per the IEC 62304 change management requirements for your software safety class.
- NB notification: If significant — copy of notification submitted to NB, NB acknowledgement, and outcome of NB review. If not significant — explicit statement that NB notification was not required and why.
- AI Act (from 2 August 2028): For AI medical devices, also record whether the change falls within the pre-determined changes described in the technical documentation (Annex IV, point 2(f)) or is a substantial modification under Article 43(4).
- ↗ MDCG 2020-3 Rev.1 (May 2023) — Guidance on significant changes (legacy devices, Article 120 MDR) — European Commission MDCG
- ↗ HSA GN-21 R7 — Change Notification for Registered Medical Devices; GN-37 R1 — Change Management Program for SaMD; GL-04 R4 — Software Medical Devices — Health Sciences Authority Singapore
- ↗ FDA — Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions — FDA final guidance (current version August 2025)
- ↗ EU AI Act, Article 43 — Conformity assessment (pre-determined changes, Article 43(4)) — European Commission AI Act Service Desk