For Regulatory Affairs Managers: Building a Predetermined Change Control Plan (PCCP) for an AI-Enabled Surgical Device

A Predetermined Change Control Plan (PCCP) is the part of a marketing submission where you describe, in advance, the AI model modifications you intend to make and how you will validate them. Once FDA authorizes it with the device, changes implemented consistently with the plan do not need a new 510(k) or PMA supplement. FDA's final PCCP guidance says a PCCP covers modifications that would otherwise require a new marketing submission. FDA's draft AI lifecycle guidance (January 2025, not final) adds recommendations on transparency and bias, including checking whether a device benefits all relevant demographic groups similarly. Sources: the FDA documents indexed in our FDA surgical robotics and SaMD base.

What a PCCP is, and its legal basis

Section 3308 of the Food and Drug Omnibus Reform Act of 2022 (FDORA), enacted December 29, 2022, added section 515C to the FD&C Act. Under 515C, a PMA supplement or new 510(k) is not required for a change that is consistent with a PCCP that FDA has approved or cleared. FDA's guidance Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions is final. It was originally issued on December 4, 2024, and the version in the base was issued on August 18, 2025.

The guidance applies to AI-enabled device software functions (AI-DSFs) that the manufacturer intends to modify over time. That includes modifications made manually, modifications implemented automatically by software (“continuous learning”), or both. FDA states it will consider PCCPs for automatic modifications but flags the added complexity and suggests clear boundaries or guardrails for the range of automatic updates.

What a PCCP can and cannot cover

A PCCP covers modifications that generally would otherwise require a new marketing submission, meaning those that could significantly affect, or otherwise affect, safety or effectiveness. Changes that would not need a submission are outside its scope; you document them under your quality system instead. FDA recommends a limited number of specific modifications that can be verified and validated. Types that may be appropriate include:

• Performance: improvements from retraining the model on new data within the intended use population, from the same type and range of input signal;
• Inputs and compatibility: new sources of the same input signal type (for example other makes or models of an acquisition system), limited new input types, or updated compatible hardware, operating systems or cloud infrastructure;
• Use and performance: for example, authorization for a specific subpopulation within the originally indicated population after retraining on a larger dataset for it.

The limits are firm. Modifications must keep the device within its intended use and, for 510(k) devices, keep it substantially equivalent to the predicate. FDA believes they should generally stay within the indications for use as well, and encourages a Q-Submission before proposing any indications change.

The three components FDA expects

The PCCP goes in the submission as a standalone section titled “Predetermined Change Control Plan”, with a version number. It is flagged in the cover letter and referenced from the device description, labeling and the safety or substantial equivalence sections. The authorization letter references the PCCP by title and version.

Getting the PCCP authorized and using it

1. Choose a submission type that ends in an FDA decision. The guidance lists the original PMA, modular PMA, 180-day, panel-track and certain real-time PMA supplements; traditional and abbreviated 510(k)s; and the original De Novo request. FDA gives feedback in a Pre-Submission but does not authorize a PCCP there.
2. Plan the labeling. Users should be told that the device incorporates machine learning and has an authorized PCCP, since software updates may change performance, inputs or use. As changes ship, the labeling should describe them and how users are informed.
3. Implement exactly as authorized. A change is consistent with the PCCP only if it is listed in the Description of Modifications and carried out under the Modification Protocol. Document it in your quality system; no new submission is needed.
4. Treat deviations as ordinary changes. If a change is not in the plan, or departs from the protocol, assess it under the usual device-modification rules. Changing the PCCP itself generally requires a new marketing submission.
5. Mind the predicate effect. For a later 510(k) that uses your device as a predicate, the comparison is to the version cleared before PCCP changes, until a subsequent clearance covers the modified device.

Bias, transparency and monitoring: the draft AI lifecycle guidance

FDA's Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, issued January 7, 2025, is still a draft. Once finalized it would represent FDA's current thinking. It defines AI bias as “a potential tendency to produce incorrect results in a systematic, but sometimes unforeseeable way” that can affect all or part of the intended use population. It recommends addressing transparency and bias across the total product lifecycle, “including by collecting evidence to evaluate whether a device benefits all relevant demographic groups (e.g., race, ethnicity, sex, and age) similarly”.

• Address representativeness in data collection for development, testing and monitoring.
• Report performance across important subgroups; the draft names patient characteristics (sex, age, race, ethnicity, disease severity), geographic sites and data collection equipment.
• Keep test data independent of training data and sequestered from model developers.
• Consider a postmarket performance monitoring plan, since data drift can degrade performance. The draft notes such a plan may become a special control in a De Novo.

These recommendations dovetail with the PCCP. The Impact Assessment must discuss unintended bias for each modification, and the Modification Protocol's data management section is where representativeness is demonstrated. Given the draft status, treat them as direction of travel and confirm the current version on fda.gov.

To pressure-test a draft plan, ask the base questions such as “We want to update our AI model after launch without filing a brand-new FDA submission every time; is there a way to plan for that upfront?” The answer quotes the PCCP guidance directly.

Sources: FDA final PCCP guidance for AI-DSFs and FDA draft AI-enabled device software functions guidance. This article explains FDA's recommendations; it is not regulatory advice for your device.