Cutting the red tape: MHRA publish new guiding principles for AI-based medical devices

In a positive move for manufacturers of AI-based medical devices, the MHRA has published new guiding principles on Predetermined Change Control Plans ("PCCPs").

Regulators of medical devices in the UK, US and Canada have been collaborating to publish guiding principles for manufacturers of medical devices which use AI. These efforts form part of the global movement to improve regulation around the use of AI in healthcare.

The use of AI in medical devices brings about new risks. For example, AI relies on the data fed into it. This means medical devices may be trained on data sets which underrepresent certain population groups. As a result, these devices can have reduced efficacy or safety when used on patients from those underrepresented populations.

To mitigate these new risks, the UK is working towards developing new regulatory frameworks. The MHRA is due to roll out changes to existing regulations over medical devices in 2024.  As an interim measure, in 2021, the MHRA, together with the FDA and Health Canada published 'Ten Principles of Good Machine Learning Practice'. The principles give guidance to developers of AI technologies on how to mitigate these new risks and best prepare themselves for compliance with future regulation.

Building on the existing 'Ten Principles' the MHRA, FDA and Health Canada have now also published 'Five Guiding Principles for Machine Learning-Enabled Medical Devices'. These new five principles offer a way for manufacturers of medical devices to avoid heavy regulatory burdens when updating and upgrading the AI systems in their medical devices.

Currently, when a medical device undergoes a sufficient degree of change, the manufacturer must inform its conformity assessment body. These bodies may require that the device is reassessed for approval. Since AI in healthcare needs to continuously adapt and respond to new input data, this creates a catch-22 for developers. They will want to ensure their devices are performing to the highest standards and adapting to new data generated by increased usage of the device. However, in doing so, they may be changing their device which in turn necessitates re-approval.

The five guiding principles aim to address this issue by offering manufacturers an alternative regulatory route where they have a 'Predetermined change control plan' (PCCP). This is a plan where the manufacturer specifies:

• Planned modifications to a device;
• A protocol for implementing and controlling those modifications;
• An assessment of the impact of those modifications.

This is good news for manufacturers. A robust PCCP can reduce regulatory burden by avoiding the need for reassessment. A PCCP does this by demonstrating that changes are safe, effective and will not negatively impact the performance of the device. Manufacturers will benefit from these principles when trying to navigate an uncertain and transitory regulatory environment around the use of AI in healthcare.