The MHRA, FDA, as well as Health Canada have jointly gone on to publish five guiding principles for machine learning-enabled medical device- MLMD manufacturers so as to develop predetermined change control plans- PCCPs.
The idea behind these guiding principles is to assist manufacturers of MLMDs by decreasing the regulatory requirements for reconsideration in case of specific modifications or improvements to their respective devices.
Building upon the 10 guiding principles for good machine learning practice, the 5 guiding principles for manufacturers of MLMD state that a properly constructed and carefully planned approach must be:
- Bounded and focused: Describing the particular modifications that manufacturers intend to make.
- Risk-based: The intent, design, and execution of a PCCP are guided by an approach based on risk that follows the tenets of risk management.
- Evidence-based: Demonstrating that the benefits of using it outweigh any potential risks throughout its entire lifecycle.
- Transparency: Ensure that all stakeholders, including patients and healthcare professionals, are provided with precise and pertinent information, as well as comprehensive plans for maintaining transparency.
- Total Product Lifecycle Perspective: Enhancing the Quality and Integrity of a PCCP by continuously taking into account the viewpoints of all stakeholders to improve the overall quality and integrity of the product.
It is well to be noted that at present, in the UK, manufacturers must go on to inform their conformity assessment body whenever they make effective and significant alterations or updates to their medical devices. This communication is essential to ensure that the modifications that are made do not have any adverse impact on the operations or security of the device. As a consequence, the device could go on to face reassessment to verify its ongoing compliance.
Medical devices that make use of AI as well as machine learning may necessitate frequent updates, and this in turn can lead to certain long revision processes whenever a change is made. The implementation of this could enforce a substantial burden of regulation when it comes to both developers as well as assessors.
PCCPs helps manufacturers of medical devices with machine learning capabilities to showcase the changes that are proposed and also provide updates they are going to carry out. By doing so, they can show their bent to maintaining the safety as well as efficacy of their products without ever requiring any sort of regulatory intervention.
These guidelines outline the domains under which the MHRA, FDA, and Health Canada share similar standards for an acceptable PCCP, with an objective to minimise or eliminate the requirement for reassessment.
While these core values will go on to support the development of PCCP across the UK, US, as well as Canada, it is important to note that each regulator goes on to have its own specific national guidance that manufacturers have to adhere to. It is well worth noting that the guidance from the MHRA is expected to be published in 2024.
Dr. Paul Campbell, who happens to be the head of software and AI at MHRA, stated that AI and MLMDs have gone on to become more common, thereby requiring regulators to alter their procedures so as to facilitate improvements for patients while still prioritising safety. By working together with the FDA and Health Canada to create these core values, one can establish a clear understanding of shared expectations for a change control plan that’s effective. This collaboration will also contribute to easing the burden of regulation, which is often faced by manufacturers.
The collaboration among regulators about these guiding principles showcases the advantages of working with international partners. This partnership aids in the establishment of flexible regulatory processes that effectively promote innovative manufacturers and also patients across the world.
