On October 24th, 2023, the FDA, Health Canada, and the MHRA published a joint document providing harmonization for machine learning-enabled medical devices (MLMD) that use predetermined change control plans (PCCPs). PCCPs are plans proposed by the manufacturer that state the specific modifications to a MLMD, the process for implementing these modifications, and the assessment of impacts from them.
The document details five guiding principles for MLMDs in an effort to set a foundation for PCCPs and encourage collaboration on them. According to the UK government’s website, these principles are:
- Focused and Bounded: Describing specific changes that a manufacturer intends to implement.
- Risk-based: The intent, design, and implementation of a PCCP are driven by a risk-based approach that adheres to the principles of risk management.
- Evidence-based: Demonstrating that benefits outweigh the risks throughout the product lifecycle.
- Transparent: Provide clear and appropriate information and detailed plans for ongoing transparency to all stakeholders, from patients to healthcare professionals.
- Total Product Lifecycle Perspective: Improve the quality and integrity of a PCCP by continually considering the perspectives of all stakeholders.
Here are some examples of how these principles could be applied:
- Focused and bounded: A manufacturer of an MLMD that diagnoses cancer might develop a PCCP to implement a change to the algorithm that improves its accuracy in detecting a specific type of cancer.
- Risk-based: A manufacturer of an MLMD that monitors a patient's vital signs might develop a PCCP to implement a change to the algorithm that reduces the likelihood of false alarms.
- Evidence-based: A manufacturer of an MLMD that delivers medication to patients might develop a PCCP to implement a change to the algorithm that improves the accuracy of the dosage.
- Transparent: A manufacturer of an MLMD might publish a white paper that describes the device's algorithm and how it was developed and tested. The manufacturer might also make available a user manual that provides clear instructions on how to use the device safely and effectively.
- Total product lifecycle perspective: A manufacturer of an MLMD might collect feedback from patients and healthcare professionals on how the device is performing after it is marketed. The manufacturer might also use this feedback to identify and address any potential problems with the device.
The five guiding principles for MLMDs using PCCPs are based on the 10 guiding principles for Good Machine Learning Practices (GMLP) published in 2021, which were designed to help medical device manufacturers develop and deploy machine learning models that are safe, effective, and high quality. Similarly, the goal of these five guiding principles is to help MLMD manufacturers develop and maintain safe and effective products that meet the needs of patients and healthcare professionals. They are also intended to streamline the regulatory process for MLMDs, making it easier for manufacturers to bring new products to market and make updates to existing products in a timely manner.
If you’re looking for additional information about MLMD requirements in the US, join Rimsys and MDQR Solutions for Taking SaMDs to market in the US on Thursday, November 30th, at 1 PM ET. We’ll discuss the various types of SaMDs, considerations to make when obtaining market clearance, and how the FDA is regulating AI/ML in devices. Those interested in attending can register here: Taking SaMDs to market in the US.