Artificial intelligence (AI) and machine learning (ML) are transforming the medical device industry, from enhancing diagnostics to automating workflows. But with the power of ML-enabled software comes a regulatory challenge. As algorithms continuously “learn” and improve, these changes should trigger new FDA submissions, creating hurdles for manufacturers trying to keep pace with innovation.
The standard ‘change control’ demands of the harmonised QMSR and ISO 13485 are not flexible enough to deal with this new challenge.
Instead, to address this, the FDA has just finalised its guidance for Predetermined Change Control Plans (PCCP) - a proactive approach to managing modifications to Machine Learning-Enabled Device Software Functions (ML-DSF).
So, what is a PCCP, and how does it simplify regulatory compliance for medical device developers?
Let’s break it down.
A PCCP is a regulatory tool that allows manufacturers to outline anticipated changes to their ML-enabled device software, including how those changes will be developed, validated, and monitored.
The PCCP is submitted as part of a device’s marketing application (510(k), De Novo, or PMA) and, when approved, enables certain changes to be implemented without requiring resubmission to the FDA - as long as the changes remain within the pre-defined parameters.
The goal? To streamline innovation while maintaining safety, effectiveness, and compliance.
As the FDA puts it:
“The FDA reviews the PCCP as part of a marketing submission for an AI-enabled device to ensure the continued safety and effectiveness of the device without necessitating additional marketing submissions for implementing each modification described in the PCCP.”
Machine learning algorithms are dynamic by nature. For example:
Each of these iterative improvements could constitute a “change” to the device software, which—under traditional regulations—would require a new submission (e.g., Special 510(k) or PMA supplement).
It’s easy to see, how, without a framework like the PCCP, medical device manufacturers could risk serious delays, excessive administrative burden, and regulatory bottlenecks.
The FDA’s guidance defines the three major parts of a PCCP:
This section outlines the planned changes to your ML-DSF, including:
The whole point of ML is to allow systems to improve and change as they learn from data. For example, cancer screening tool may specify improvements to sensitivity and specificity as it processes new training data. When self-modification is the ‘way it works’, regulators must find a way to define and validate this functionality.
The Modification Protocol is a step-by-step plan for implementing and validating the modifications described above. It includes:
This protocol ensures that every change to the software is systematically managed within your existing Quality Management System (QMS) a critical FDA requirement.
The Impact Assessment evaluates the benefits, risks, and cumulative effects of the planned modifications. It includes:
It should be noted, the FDA highlights that your QMS is the foundation for conducting this assessment, reinforcing the importance of robust quality processes in managing ML-enabled devices.
The PCCP offers a practical and ‘least burdensome’ approach to managing machine learning-enabled devices, allowing:
The FDA’s Predetermined Change Control Plan represents a significant step forward in aligning regulatory processes with the realities of AI and machine learning.
By embracing the PCCP, medical device manufacturers can innovate with confidence, knowing they have a clear, compliant pathway for managing ML-driven improvements.
This approach ensures a balance between innovation and safety, with clear documentation at its core — proving that in the AI era, strong processes are just as critical as smart algorithms.
If your team is exploring ML-enabled devices, now is the time to build a strong QMS foundation and start your PCCP implementation to help smooth the way for future FDA approvals.