Published October 2026
Two recent regulatory developments warrant attention from medical-device and IVD evidence teams: the UK government’s plan to recognize devices that meet EU MDR and IVDR requirements indefinitely, and FDA’s new hypothetical examples of predetermined change control plans for AI-enabled devices.
Both developments point to the same practical issue: manufacturers want to reuse sound evidence where they can, but they must also show that an evolving device remains safe and continues to perform as intended.
That issue reaches beyond market-access strategy. It shapes the way teams prepare and maintain performance evaluation reports (PERs), clinical evaluation reports (CERs), validation studies, state-of-the-art reviews, and post-market evidence.
The UK announcement strengthens the case for investing in EU evidence
On 21 October 2026, the MHRA published the government’s response to its consultation on indefinite recognition of CE-marked devices in Great Britain. The government intends to recognize devices compliant with the EU Medical Devices Regulation, Regulation (EU) 2017/745, and the EU In Vitro Diagnostic Medical Devices Regulation, Regulation (EU) 2017/746, indefinitely. It also intends to align Great Britain’s transitional arrangements for devices complying with the older Medical Devices Directive with EU timelines. The policy applies to England, Wales, and Scotland; Northern Ireland has a separate regulatory framework.
The distinction between a policy announcement and an implemented legal change matters. The government response says legislative changes will follow, subject to parliamentary approval, before the earliest relevant transition period ends on 30 June 2028. UKCA will remain a parallel route. Manufacturers supplying Great Britain will retain domestic obligations, including MHRA registration, post-market surveillance, and a UK Responsible Person where the manufacturer is based outside the UK.
For evidence teams, this gives an EU-focused program more durable value. A manufacturer serving both the EU and Great Britain may be able to use the same core body of clinical, literature, and post-market evidence across both markets. The caveat is important: teams should wait for the implementing legislation before revising formal compliance assumptions.
One sensible step now is to map the studies, claims, and evidence gaps that matter in more than one market. When intended purpose, device configuration, and patient population align, the same evidence may support several regulatory documents. Any local differences should still be called out directly in the relevant assessment.
FDA’s examples make prospective change planning more concrete
FDA’s Novel Medical Devices with Predetermined Change Control Plans: Hypothetical Case Examples describes two approaches to managing AI-device updates.
The sepsis-alert example allows specified hospital-level adaptations, including local model fine-tuning and alert-threshold adjustments. The adapted model is locked before validation, tested using independent data, and released through a controlled process. The arrhythmia example describes periodic retraining followed by automated validation. Deployment proceeds only when all predefined acceptance criteria are met; otherwise, the update is blocked and the manufacturer is alerted.
The examples are educational, not complete submission templates or new binding requirements. One detail is worth noting: although FDA calls the arrhythmia example “continuously learning,” the process it describes relies on defined retraining cycles and validation gates.
At the center of both scenarios is an FDA-authorized predetermined change control plan, or PCCP. A PCCP can cover specified future modifications without requiring a new marketing submission for every change, as long as each modification stays within the authorized plan and follows its methods. The plan ties together three things: the proposed modifications, the methods and acceptance criteria for evaluating them, and an assessment of their individual and cumulative effects.
FDA’s August 2025 final guidance, Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions, sets out the broader submission recommendations under docket FDA-2022-D-2628. The October examples show how that framework might operate in practice.
The evidence challenge starts before the next update
Planning works best when the proposed change is described precisely. “Improve performance” is too broad to guide a validation strategy. The description should identify what will change, how much variation is allowed, the clinical objective, and the evidence required before release.
Take a diagnostic algorithm with an adjustable decision threshold. Before testing an update, the team needs to decide what balance of false positives and false negatives is clinically acceptable, whether particular populations require separate analysis, and how uncertainty will affect the release decision. Those choices will vary with the device and its claims; they are planning questions, not universal requirements.
Dataset governance also needs early attention. FDA’s final AI PCCP guidance recommends separation of development and test data, representation of intended populations and environments, and controls against bias when test datasets are reused. Its appendix specifically asks how manufacturers will prevent repeated evaluations from revealing enough information about the test set to compromise independence.
In practice, that means naming who controls dataset access, recording how a test set may be reused, and deciding when a fresh dataset is necessary. The validation strategy has to remain credible after the first update—and after the fifth.
https://www.fda.gov/media/166704/download
Connecting change control with PER, CER, and SOTA work
Traceability is the thread that connects this work. Each clinical claim should point to its supporting evidence and identify the device version evaluated. When a change is proposed, that record makes it easier to see which conclusions still hold and which ones need another look.
For an IVD, it may help to consider scientific validity, analytical performance, and clinical performance separately. A software change could alter result interpretation without affecting other parts of the test system; a different change might call for substantial new analytical work. The evidence plan should make clear which situation applies and why.
Literature reviews are more useful when they capture details such as software version, decision threshold, reference standard, clinical setting, and population characteristics, when those details are reported. Without them, reviewers may struggle to judge whether published results apply to the current device.
A state-of-the-art update can also reveal whether an established performance benchmark is still clinically persuasive. The record should show how new evidence affected the benchmark and the benefit–risk assessment, including the reasons for keeping or changing acceptance criteria.
These recommendations concern evidence management, not automatic regulatory acceptance. An FDA-authorized PCCP does not establish EU or UK acceptance of a modification, and the UK recognition announcement does not authorize a particular AI update. A shared evidence program can reduce duplication, but it does not replace a jurisdiction-specific change assessment.
The clearest near-term deliverable is a record that connects each claim to the relevant device version, validation criteria, and post-market findings. Done well, that record makes evidence easier to reuse and gives reviewers a straightforward basis for deciding whether it still applies.
