In short
- A predetermined change control plan (PCCP) describes the planned modifications to an AI-enabled device, the protocol for developing, validating and implementing them, and an assessment of their impact, and FDA reviews it as part of a 510(k), De Novo or PMA.
- Modifications made as the authorized PCCP specifies do not need a new marketing submission, while changes outside the plan, or not carried out as its protocol requires, generally do.
- FDA's final guidance, originally issued on December 4, 2024 and reissued on August 18, 2025, recommends a limited number of specific modifications that can be verified and validated and that keep the device within its intended use.
A PCCP for an AI medical device is a plan, authorized together with the device, that lists the specific model changes you intend to make and exactly how you will develop, test and roll out each one. Section 515C of the Federal Food, Drug, and Cosmetic Act, added by the Food and Drug Omnibus Reform Act enacted on December 29, 2022, lets FDA approve or clear such a plan. Changes made consistent with it do not need a new 510(k) or a PMA supplement. FDA's final guidance on PCCPs for AI-enabled device software functions was originally issued on December 4, 2024, and the current version is dated August 18, 2025.
A PCCP lets a model improve without a new submission for every retraining, while FDA still reviews, before any change ships, what the changes will be and how you will prove each one is safe.
What does section 515C allow?
- For devices approved through PMA, a supplemental application is not required for a change consistent with an approved PCCP.
- For devices cleared through 510(k), a new premarket notification is not required for a change consistent with a cleared PCCP.
- The plan is authorized only if the device remains safe and effective without the change, and for 510(k) devices, remains substantially equivalent to its predicate.
- FDA may require a plan to include labeling for safe and effective use as the device changes, notification requirements if the device does not function as intended, and performance requirements for changes made under the plan.
- A changed version of a device cannot serve as a predicate. Only the version cleared or approved before the changes can, although FDA's guidance adds that once a later 510(k) clears a device that includes the implemented modifications, that device becomes an eligible predicate.
Which submissions can carry a PCCP?
The guidance lists original PMA applications, modular PMAs where the PCCP is a module, 180-day and panel-track PMA supplements, real-time PMA supplements for minor changes, Traditional and Abbreviated 510(k)s, and original De Novo requests. A PCCP is authorized only as part of a marketing authorization. FDA encourages a Pre-Submission to get feedback on a proposed plan, but it does not authorize a PCCP in a Pre-Submission.
In the submission, the PCCP should be a standalone section with a title and version number, mentioned in the cover letter and listed in the table of contents as Predetermined Change Control Plan. Once authorized, the plan's title and version are referenced in the device's letter of authorization.
What goes in a PCCP?
Three parts: a Description of Modifications, a Modification Protocol and an Impact Assessment. FDA treats them as interrelated. The description says what will change, the protocol says how each change will be made and proven, and the impact assessment ties them together by weighing benefits and risks.
Description of Modifications
List each planned modification and the rationale for it. FDA recommends a limited number of modifications that are specific and can be verified and validated, each linked to a performance evaluation activity in the protocol. The description should also say:
- Whether each change is implemented automatically by software, manually with human review and decision, or a mix.
- Whether it applies the same way to all devices on the market, or differs by clinical site or patient, and if so, which local factors justify it.
- How often updates are expected, from annually or less, which leaves the device mostly locked, to continuous adaptation during use.
The guidance names three types of modifications that may fit: changes to quantitative performance specifications, such as improvements from retraining on new data within the intended use population and the same type and range of input signal; changes to inputs and compatibility, such as new makes or models of the same acquisition system, limited new inputs, compatible hardware, updated operating systems or updated cloud infrastructure; and certain changes to use and performance, such as authorization for a subpopulation within the original population after retraining on more data. Modifications must keep the device within its intended use, and FDA generally expects them to stay within its indications for use too.
Modification Protocol
The protocol is the step-by-step method for each change, with pre-defined acceptance criteria. FDA identifies four components, and a traceability table can show which method in each component supports each modification.
- Data management practices: how new data will be collected, annotated, curated, stored, retained and controlled; how the reference standard is set; how training, tuning and test data are kept apart; and how access to test data is blocked during training and tuning.
- Retraining practices: which processing steps may change, the rationale for any architecture changes, and the triggers for retraining, such as a volume of new data or observed drift.
- Performance evaluation: study design, metrics, acceptance criteria and statistical tests for each change, comparing the modified device with the original version and with the last modified version. If a change fails its criteria in a way that cannot be resolved, the failure is recorded and the change is not implemented.
- Update procedures: how the update is deployed, how users are told, how labeling is updated, how legacy users are affected, and how real-world performance will be monitored after the change.
Impact Assessment
- Compare the device with each modification implemented individually against the device with none.
- Discuss the benefits and risks of each modification, including risks of harm and unintended bias.
- Explain how the protocol's verification and validation continue to assure safety and effectiveness.
- Explain how implementing one modification affects another.
- Describe the cumulative impact of implementing all of them.
What happens after FDA authorizes the plan?
When a modification is listed in the Description of Modifications and carried out exactly as the Modification Protocol says, no new submission is needed. You implement it and document the change and its analysis in your quality system. If the change is not in the plan, you evaluate it like any other device change. If it is in the plan but was not carried out as the protocol requires, a new marketing submission is likely needed before you implement it.
FDA treats not following an authorized PCCP as a deviation, and says deviations would generally cause the device to be adulterated and misbranded. Changing the plan itself, whether the description, the protocol or the impact assessment, generally needs a new marketing submission, although for 510(k) devices a Special 510(k) may be appropriate in some cases.
Labeling carries the plan to users. FDA recommends that labeling state that the device has an authorized PCCP and, as changes ship, describe what was implemented, the data behind it, how it was implemented and how users are informed, for example through updated instructions or a version history. Labeling should not describe modifications that have not been implemented in the version users have. Public summaries, such as the 510(k) summary, should also describe the plan.
How should engineering build for a PCCP?
A PCCP works only if the team follows it to the letter on every change and keeps the records, and much of that depends on how the engineering pipeline is built.
- Version every dataset, label set and reference standard, and keep the sequestered test set behind access controls that training jobs cannot cross.
- Turn the acceptance criteria into an automated evaluation that runs the same protocol on every candidate model and compares it with the original and the last released version.
- Make the release pipeline refuse to ship a model that fails its criteria, and record the failure.
- Tag each release with the PCCP modification it implements, so the trace from plan to evidence to deployed version is a lookup.
- Generate the labeling changes and version history from the same release records.
- Monitor performance in production against the plan's monitoring section, with a route back into risk management.
- Treat platform changes such as operating system or cloud updates as possible plan items, since FDA lists compatibility changes among the modification types.
What mistakes should you avoid?
- Writing a broad plan in the hope of covering everything. FDA recommends a limited number of specific, verifiable modifications.
- Planning a new intended use. Modifications must stay within it.
- Letting test data leak into training or tuning across versions.
- Forgetting the quality system. FDA expects changes under a PCCP to be implemented and documented in it. The guidance notes that FDA's amended quality system rule, which incorporates ISO 13485:2016 by reference, took effect on February 2, 2026.
- Leaving the plan out of the device description, labeling and substantial equivalence discussion, where FDA expects it to appear.
Planning an AI device that will keep learning?
We build medical device software under an IEC 62304 process and produce the documentation an FDA submission needs, including the data, evaluation and release records a PCCP depends on. The manufacturer writes the plan with its regulatory team and makes the submission. Onix Systems has built software since 2000 and has 370+ engineers. Talk to one of our engineers about your model update pipeline, or start with our free five-day technical audit, which ends with a written report.
Sources
- FDA: Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions (guidance page)
- FDA: PCCP for AI-Enabled Device Software Functions, final guidance PDF (August 18, 2025)
- 21 U.S.C. 360e-4: predetermined change control plans for devices, FD&C Act section 515C (Cornell LII)
- FDA: Content of Premarket Submissions for Device Software Functions (June 2023)