TLDR

FDA's December 2024 final guidance on Predetermined Change Control Plans (PCCPs) gives AI/ML medical device manufacturers a regulatory pathway to pre-specify modifications to adaptive algorithms without filing a new 510(k) for every update. A PCCP embeds a structured change control protocol directly into the original marketing submission, covering what will change (SaMD Pre-Specifications), how changes will be implemented and validated (Algorithm Change Protocol), and what impact on safety and effectiveness is expected. Under QMSR and ISO 13485:2016, PCCP management requires integration with design change control (ISO 13485 7.3.9), risk management (ISO 14971), and a full audit trail on every modification event. Quality teams that understand this framework before their first PCCP submission will avoid the most common errors that delay clearance and create post-market compliance gaps.

Artificial intelligence changes things. A traditional cleared medical device is static: the algorithm that was cleared is the algorithm on the market. An AI/ML-enabled device is different — the algorithm can learn, adapt, and improve based on new data. That adaptive capability is precisely what makes AI/ML devices valuable in clinical settings. It is also what makes them difficult to regulate under a traditional 510(k) framework, which requires a new submission every time a device undergoes a modification that could significantly affect safety or effectiveness.

FDA's Predetermined Change Control Plan (PCCP) framework exists to solve that problem. It allows manufacturers to define, in advance, what algorithm modifications they plan to make, how they will validate those modifications, and how they will ensure the changes remain within the bounds of the original clearance. FDA reviews and clears the PCCP as part of the marketing submission, and the manufacturer can then implement those pre-specified changes without a new 510(k) for each one.

Why PCCPs Exist: The Adaptive Algorithm Problem

Traditional 510(k) regulation assumes a fixed device. A manufacturer submits data demonstrating substantial equivalence to a predicate device, FDA clears the submission, and the cleared device goes to market. Any subsequent modification that could significantly affect safety or effectiveness requires a new submission.

AI/ML-enabled Software as a Medical Device (SaMD) breaks that model. An adaptive algorithm that improves its diagnostic accuracy on new patient data may perform a different function than it did at clearance — not because the manufacturer changed the algorithm code directly, but because the model's weights and parameters evolved through exposure to real-world data. Under traditional interpretation, each significant performance shift could require a new 510(k). That is not compatible with how these devices are designed to work.

FDA recognized this tension in its January 2021 AI/ML-Based SaMD Action Plan, which proposed the PCCP framework as a regulatory tool to enable safe iteration of adaptive algorithms. Congress subsequently codified the PCCP concept in the Food and Drug Omnibus Reform Act of 2022 (FDORA), which added Section 515C to the Federal Food, Drug, and Cosmetic Act. FDA's December 2024 final guidance, "Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions," provides the operational framework for implementing PCCPs in marketing submissions and quality systems.

What a PCCP Must Contain

FDA's December 2024 guidance establishes that a PCCP consists of two primary components and a supporting impact assessment.

Component 1: SaMD Pre-Specifications (SPS). The SPS describes what the manufacturer intends to change about the device. It defines the types of modifications that may be made, the intended scope of those changes, and what aspects of the device's intended use, inputs, or performance metrics may be affected. The SPS must be specific enough for FDA to evaluate the boundaries of the proposed changes, but flexible enough to accommodate the range of modifications that real-world performance data may generate.

Types of changes typically addressed in an SPS include: expansion of the device's patient population; modifications to input data types or acquisition parameters; changes to algorithm outputs or performance thresholds; and updates to the training or retraining methodology.

Component 2: Algorithm Change Protocol (ACP). The ACP describes how the manufacturer will implement, validate, and verify each type of modification specified in the SPS. The ACP must demonstrate that the manufacturer has a structured, repeatable process for collecting and managing the data used to train or retrain the algorithm; validating that the modified algorithm meets the performance specifications established at original clearance; verifying that the modification does not introduce new risks or degrade performance on the original intended population; and documenting the change event with a complete audit trail.

FDA expects the ACP to function as a living quality system document, not a one-time validation plan. Every modification event generates a new record in the ACP log, demonstrating that the pre-specified change protocol was followed.

Impact Assessment. The impact assessment evaluates how the modifications described in the SPS, implemented under the ACP, are expected to affect the safety and effectiveness of the device. It must address: the effect on device performance across the intended population; changes to known risks and risk controls; and any new risks introduced by the proposed modifications. The impact assessment connects the PCCP to the device's overall risk management file.

What FDA Will and Will Not Accept

FDA accepts specificity. The strongest PCCPs define clear quantitative boundaries for each modification type. Instead of "the algorithm may be updated to improve sensitivity," a well-formed SPS states: "sensitivity to the target condition may increase from the cleared baseline of 87% to a maximum of 95%, measured on a population dataset conforming to the specifications in Appendix A, without a corresponding reduction in specificity below 82%." FDA needs numbers it can hold the manufacturer accountable to.

FDA accepts structured validation frameworks. An ACP built around ISO 14971 risk analysis, statistically powered validation studies, and defined acceptance criteria is substantially more likely to clear than one that describes validation in general terms. The ACP should reference the statistical methodology, the dataset composition requirements, and the threshold at which performance changes require escalation to full design change review.

FDA will not accept unlimited scope. A PCCP that attempts to cover every possible modification to the algorithm is not a PCCP — it is an open-ended request for modification authority. FDA expects manufacturers to define meaningful boundaries. Modifications outside those boundaries require a new submission.

FDA will not accept disconnected quality systems. A PCCP that describes a modification process separate from the manufacturer's existing design change control procedures will be questioned. The PCCP should reference and integrate with the QMS change control process, not operate in parallel to it.

Embedding PCCP Management in QMS Change Control Under QMSR and ISO 13485

Under QMSR and ISO 13485:2016, every design change must go through a formal design change control process under ISO 13485 Section 7.3.9. That requirement does not disappear because a modification is covered by a PCCP. The PCCP defines which modifications can proceed through an expedited change control pathway — one that does not require a new 510(k) — without eliminating the change control obligation.

Practical QMS integration for PCCP management requires the following structural elements:

PCCP Change Event Record. Every time the manufacturer implements a modification covered by the PCCP, a formal change event record must be generated. The record must document: the nature of the modification (with reference to the applicable SPS category); the dataset used for retraining or update; the validation and verification activities performed; the ACP step-by-step compliance log; the impact assessment update; the approval signatures of responsible personnel; and the effective date. This is the PCCP audit trail that FDA will request during a QMSR inspection or PMA supplement review.

Risk Management File Integration. Each PCCP modification event must feed back into the risk management file. The impact assessment update should compare the post-modification risk profile against the pre-modification baseline. If the modification changes the probability or severity of any existing risk, or introduces a new hazardous situation, the risk management file must be updated before the modification is released. This is precisely the integration that CP 7382.850 inspectors evaluate under ISO 13485 Clause 7.1.

Design Change Control Linkage. The PCCP change event record should link to the active design change control record in the QMS. Under QMSR inspection Model 2, FDA will review design and development records including design changes. A modification covered by a PCCP that cannot be traced to a corresponding design change control record creates a documentation gap.

Out-of-Scope Escalation Protocol. The ACP must include a documented protocol for what happens when a performance observation or data finding suggests a modification that falls outside the PCCP's SPS boundaries. That escalation must route to a formal design change control review and determine whether a new submission is required. This protocol should be documented in the QMS as a standing procedure.

Training and Competency Records. Personnel responsible for executing PCCP modification events must be trained on the ACP procedures and on the regulatory significance of the PCCP boundary conditions. Training records for PCCP-related activities are reviewable by FDA under the Management Oversight QMS Area.

Writing SaMD Change Protocols That FDA Will Accept

The Algorithm Change Protocol is the document FDA reviewers scrutinize most closely in a PCCP submission.

The ACP should begin with a clear statement of the algorithm's current architecture, inputs, outputs, and performance baseline at the time of the original marketing submission. This creates the reference state against which all future modifications are evaluated.

Each SPS modification category should have its own ACP section describing the specific validation approach for that type of change. An input expansion change requires a different validation approach than a retraining event on expanded population data. The ACP should not treat all changes with a single generic validation protocol.

The ACP should define the statistical minimum for each validation dataset, including sample size, demographic representation requirements, prevalence assumptions, and confidence intervals for performance claims. FDA's AI/ML-Based SaMD Action Plan and the December 2024 final guidance both reference the importance of statistically rigorous validation dataset specifications in ACP submissions.

The ACP should include a change log template — a standardized form that personnel complete for each modification event, capturing all required data points in a consistent format. This template becomes the PCCP change event record, and its consistency across modification events makes FDA review of the post-market modification history substantially more efficient.

The Intersection of PCCPs and QMSR Inspection Risk

Post-market PCCP compliance is an active inspection risk area under CP 7382.850.

The Design and Development QMS Area covers design inputs, design verification, design validation, software validation, and design transfer. PCCP modification events involve all of these elements. A modification implemented under the ACP is a design change, and design changes require documented verification and validation regardless of whether they require a new 510(k).

The Change Control QMS Area covers product and process changes. Every PCCP modification event is a product change. If the change control record does not exist or is incomplete, investigators will note the gap.

The Measurement, Analysis, and Improvement Area covers complaint handling and feedback. For AI/ML devices, postmarket performance data — including any complaint or adverse event associated with the algorithm's clinical output — must feed into the risk management file and into the PCCP's ongoing performance evaluation process.

The Management Oversight Area covers risk management file completeness and the medical device file. The PCCP itself is part of the device's regulatory history, and the modification event log is part of its design history. Both must be current and consistent.

How Cloudtheapp Supports PCCP Management

Cloudtheapp is an AI-powered, no-code Quality Management System validated for FDA 21 CFR Part 820 (QMSR), ISO 13485:2016, ISO 9001, and ISO 22001. For manufacturers managing AI/ML device updates under a PCCP, the platform provides the infrastructure to execute every element of the ACP as a documented, auditable, QMS-integrated workflow.

The Change Management module manages PCCP modification events as structured change control records linked to the ACP procedure, with fields for modification category, validation dataset reference, performance data, impact assessment, and approval signatures. Every change event produces a complete record with a full audit trail.

The Risk Management module maintains the device risk file as a live document updated with each PCCP modification event. When a modification changes the device's risk profile, the risk management file is updated through a structured workflow, timestamped and attributed to responsible personnel.

The Document Control module manages the ACP document itself — version control, approved revisions, and distribution to trained personnel — ensuring that the ACP in use is always the current approved version.

The Learning module manages training records for personnel responsible for executing PCCP modification events, ensuring competency is documented before personnel are authorized to execute ACP procedures.

The Audit Trail capability spans every module. Every record carries a complete history meeting 21 CFR Part 11 requirements for electronic records and electronic signatures, producing exactly the documentation chain FDA investigators expect when reviewing a post-market PCCP modification history under QMSR.

Ready to build the QMS infrastructure that supports PCCP management, AI/ML change control, and continuous QMSR inspection readiness? Request a demo of Cloudtheapp and speak with a quality management specialist about your specific SaMD compliance environment.

Frequently Asked Questions

What is a Predetermined Change Control Plan (PCCP)?
A PCCP is a document submitted to FDA as part of a marketing submission that specifies in advance what modifications a manufacturer plans to make to an AI/ML medical device, how those modifications will be implemented and validated, and what impact they are expected to have on safety and effectiveness. FDA-approved PCCPs allow manufacturers to implement the pre-specified modifications without a new marketing submission for each change.

What FDA guidance covers PCCPs?
FDA's December 2024 final guidance, "Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions," is the primary document. The PCCP concept was addressed in FDA's January 2021 AI/ML-Based SaMD Action Plan and codified by the Food and Drug Omnibus Reform Act of 2022 (FDORA).

Do all AI/ML medical devices need a PCCP?
No. A PCCP is a voluntary regulatory tool. Manufacturers who plan to file a new submission for every modification do not need one. However, for devices with adaptive algorithms designed to improve with new data, a PCCP is typically the more efficient regulatory strategy.

What is the difference between SaMD Pre-Specifications and an Algorithm Change Protocol?
The SaMD Pre-Specifications (SPS) describe what will change: the types of modifications, their scope, and the device aspects they may affect. The Algorithm Change Protocol (ACP) describes how those changes will be made: data management, validation methodology, acceptance criteria, and documentation requirements for each modification event.

How does a PCCP integrate with QMS design change control under QMSR?
Under ISO 13485:2016 Section 7.3.9, every design change must go through formal design change control. A PCCP modification event is a design change and must produce a corresponding change control record in the QMS. The PCCP defines which modifications can proceed through an expedited pathway — it does not eliminate the QMS change control obligation.

What happens when a planned modification exceeds the PCCP's SPS boundaries?
The ACP must include a documented escalation protocol. That escalation routes to a formal design change control review, which determines whether the modification requires a new marketing submission. FDA may issue a warning letter for unauthorized changes if a manufacturer implements out-of-scope modifications without a new submission.

This post created by and appeared first on Cloudtheapp