• Contact Us

AI - the MHRA's approach to the impending regulatory influx

on Thursday, 16 May 2024.

For UK Life Sciences, the MHRA shoulders the responsibility for balancing between making the UK a hub for AI innovation, and keeping products within acceptable limits. A likely rise in AI-related medical products also adds to the challenge.

Following the call in the UK Government's AI pro-innovation regulation policy paper for regulators to publish their strategies, the MHRA published 'Impact of AI on the regulation of medical products' in April this year. The strategy covers the development of AI product regulation, its own use of AI for efficiency gains and its understanding of the use of AI in the sector.

The five principles of regulation set out in the pro-innovation policy paper underpin the strategy. This strategy, which responds for the Secretary of State's letter on the subject, is about the MHRA's efforts to implement those principles.

AI product regulation

The MHRA reiterated that the current regulations for medical devices can also apply to AI used for medical purposes, making AI qualify as a medical device. They advocate for a balanced, proportionate approach to regulation, which includes updating the existing regime and offering specific guidance. This strategy points to existing guidance, in addition to releasing further guidance on key areas such as:

  • Machine learning in medical device development
  • AI as a Medical Devices (AIaMD) best practices
  • Applying human factors in AIaMD
  • Cybersecurity

The MHRA stated that it aims to support innovation safely by using a principles-based approach, supplemented by clear guidance to ensure businesses have the certainty they need while maintaining the flexibility to oversee emerging technologies effectively.

From Spring 2024, a pilot AI Airlock will launch. The intention is that experience from the AI Airlock will help the MHRA to identify and address the novel regulatory challenges for AIaMD.

Expected changes to AIaMD regulation

AIaMD is a subset of Software as a Medical Device (SaMD) and is regulated as such. That is not going to change, and the MHRA points towards its software guidance. However, the MHRA has indicated that many AIaMD products that are currently class I will be "up-classified", meaning the route to market will involve more detailed scrutiny.

Since certain AI systems naturally change over time, the MHRA intends to introduce the concept of optional Predetermined Change Control Plans (PCCP). This would allow manufacturers to update their device status more efficiently.

Transparency

As in other key sectors, there is significant concern about a lack of transparency about how AI technology works causing significant safety and equity issues. The MHRA has highlighted the current requirement for manufacturers to describe the intended purpose of the device, which is supported by existing guidance on how to craft the intended purpose.

International

If you are wondering about the UK's position on the world stage, the MHRA has described its work within the International Medical Devices Regulators Forum (IMDRF), including its position as co-chair of an AI and Machine Learning Working Group. The MHRA intends to work towards international alignment, referring to the standard ISO/IEC TR 24027:2021 and to IMDRF guidance document N65. Therefore, the sector can expect to need to get to know international standards in detail.

The MHRA points out that its guidance on PCCP was the result of a collaboration with the FDA and Health Canada. It is also working with the Council for International Organizations of Medical Sciences on international alignment.

In short, it would be wise, as it has been for a long time, to consider requirements in the wider world, even if you are only considering UK compliance at this stage.

Customer use of AI

The MHRA is aware that businesses in the sector are using AI, but it has concluded that it will not change the safety questions that the MHRA will ask. In this sense, the MHRA's focus is on adapting to customer changes (rather than making its own). It has identified three areas of customer change:

  • Development projects quicker to fail, which should increase the speed of new medicines being generated
  • New clinical trial design approaches
  • Enabling of personalised medicines

MHRA efficiency

The MHRA is also considering using AI to improve its services. Through efficiency gains, it will allow earlier access to the market and prioritise tasks requiring advanced skills.

The AI solutions being considered include:

  • Supervised machine learning to assist MHRA assessors with initial assessment of applications
  • Machine learning and AI to analyse real world data (RWD) relating to post-market safety
  • LLMs and generative AI, although with significant scepticism
  • Medicines Website Checking tool for consumer analysis of suspect online sellers

An important coming step in the near future is the outcome of the Ludlow review into the UK health data landscape. Those interested in analysing UK health data, on which AI technology relies, should watch out for its recommendations.

Comment

There is a lot going on! A lot of innovation, both by the regulator and the regulated. Although the UK lacks specific AI laws, change is coming that will pose several new challenges for life sciences businesses to appreciate, adapt to, and even take advantage of.


Are you developing an AI product or considering using AI technology? If you would like to discuss the UK environment for AI in the life sciences sector, contact Harry Jennings in our Pharmaceuticals and Life Sciences team on 07789 533 122, or complete the form below.

Get in Touch

First name(*)
Please enter your first name.

Last name(*)
Invalid Input

Email address(*)
Please enter a valid email address

Telephone
Please insert your telephone number.

How would you like us to contact you?

Invalid Input

How can we help you?(*)
Please limit text to alphanumeric and the following special characters: £.%,'"?!£$%^&*()_-=+:;@#`

See our privacy page to find out how we use and protect your data.

Invalid Input