UK Parliament / Open data

Medicines and Medical Devices Bill

My Lords, I thank the Minister for the opportunity to speak today about medical devices that make use of data-processing methodologies, algorithms, artificial intelligence and other technical architecture. In moving Amendment 83, I shall speak also to Amendments 112 and 113.

I am a long-standing advocate of the use of patient data to provide better healthcare. Noble Lords can therefore rest assured that I by no means seek to introduce obstacles to the vital research and innovation that I know the Government wish to encourage and facilitate to improve the nation’s health. However, I seek to guarantee patient safety in our increasingly data-driven health service and to ensure that the UK is well placed to become a rule-maker and world leader post Brexit.

Others in this place have spoken eloquently about the need to prioritise consideration of patient safety in the Bill. In particular, I was moved by the expert and passionate speech made by the noble Baroness, Lady Cumberlege, at Second Reading. I hope and anticipate that the noble Baroness, Lady Bennett of Manor Castle, will remind us of the social and environmental consequences we should pay heed to in encouraging innovation in medicines and medical device development, which, I believe, should result in closer consideration of the implications of the growth in digital healthcare for the climate as the Government prepare to host COP 26. However, it was the noble Lord, Lord Patel, who helpfully reminded us of the importance of precision in the definitions and wording we agree in this place, and it is my intention here to do the same.

In speaking to Amendments 112 and 113, I am recommending specific protections in the form of an up-to-date definition of a medical device, because the Bill currently relies on a definition from the Medical Devices Regulations 2002, which were updated to include the term “software” in 2007 and therefore do not safeguard individuals from the potentially harmful effects of the full range of contemporary medical devices. The effect of my amendment would be to update the definition of a “medical device” and bring it in line with regulatory provisions in the EU and the US. The use of technology has advanced significantly since 2007 and I can see no good reason why the UK should be a laggard at this time.

In moving Amendment 83, I recognise that we are always, in some senses, on the cusp of new developments or breakthroughs in science and technology that have the potential to transform healthcare services as well as outcomes for individuals. I can therefore understand the Government’s desire for flexibility in presenting this Bill as they have done. However, I am certain that the Minister will agree that we find ourselves on the cusp where growth in the use of artificial intelligence in healthcare is concerned, not least because the Department of Health and Social Care has invested more than £250 million in establishing an AI lab this year and is preparing to publish the first national health and care data strategy this autumn, which is expected to state that the Government will prioritise the

“safe, effective and ethical use of data-driven technologies, such as artificial intelligence, to deliver fairer health outcomes”.

As such, I very much hope that the Government will, at the very least, support Amendment 83, the effect of which would be to require the Secretary of State to initiate a comprehensive technical review of the definition of a “medical device” under the Medical Devices Regulations 2002 within 12 months of this Bill coming into force, with a view to addressing the inclusion of artificial intelligence, software and algorithms, including methodologies for the interpretation of data and associated technical architecture, in the legal definition of a “medical device”.

3.45 pm

I thank the Minister, the noble Lord, Lord Bethell, for his detailed correspondence with me since the Bill’s Second Reading. I will rehearse some of the questions and answers we explored for the benefit of other noble Lords. They are, admittedly, broad-ranging and of a technical nature, so I apologise in advance for the detail of what comes next, but I hope they will explain what is at stake.

In his letter of 4 November to me, the Minister appeared to confirm that algorithms are already covered by the term “software”, which would negate the need for my Amendments 112 and 113, but only in so far as algorithms underpinning medical devices amount to what might be termed “static” instructions, as distinct from the dynamic algorithms with which I am particularly concerned. My understanding is that, in addition to software involving static algorithms, modern healthcare is also making growing use of dynamic algorithms. These result in exceedingly complex data-processing functions which are arduous to audit or control adequately from the perspective of versioning.

For example, deep neural network—DNN—algorithms, which since 2016 have facilitated the exponential growth of the artificial intelligence market, are what is termed “black box” approaches where there is no way to work backwards from the outcome to determine the calculation or processing steps that arrived at the decision or recommendation. At present, it is also my understanding that they cannot be scrutinised for approval by a human being in the same way that static algorithms, such as QRISK2, or other traditional clinical calculators, can be. Therefore, will the Minister say whether and how they are currently being regulated in practice, since he appears to take the view that the Bill’s provisions suffice as they stand?

Of course, the data that a deep neural network approach relies on is as crucial as the algorithm or model used in any software, and updating a training dataset can significantly alter the function of an AI model through changing the weighting of components in an algorithm. In this instance, the algorithm is unchanged, but the weighting of different parts thereof does change. Would such an algorithm be deemed a new medical device in the eyes of the regulatory framework that the Government propose to introduce? It could, after all, result in a change with the potential to cause harm, such that it could be said to pose a risk to patient safety if left unchecked.

Beyond this, there are leading-edge approaches, such as reinforcement learning, where over time and through iterative processing an algorithm self-alters as the outcomes of different clinical decisions become apparent. These techniques are such that software per se is constantly changing and, as each case is resolved in a clinical setting, the ongoing function of the algorithm is altered. In practice, this means that on any given run of the software underpinning devices in relation to healthcare, the output could be different. Here, there are services, tools and devices available in different markets which use this technology in healthcare that would not be classed as a medical device as the Bill stands.

A specific example is GOQii, which is an app that tracks lifestyle conditions including, for example, diabetes and hypertension:

“Its ecosystem of wearables can currently track blood pressure and heart rate. The app mines this data and uses AI to make predictions about a possible fatal health situation that should serve as a warning for people.”

GOQii does not currently appear in the MHRA’s register of medical devices, and the developer’s website states that the platform does not amount to a medical device—but it would constitute a medical device that would be difficult to audit due to the lack of versioning control were all algorithms covered by the term “software” in the Bill, as the Minister appears to be suggesting is the case.

Moreover, reinforced learning has been applied in experiments that date back to 2017 and were designed to optimise antiretroviral therapy in HIV, tailor anti-epilepsy drugs for seizure control and determine the optimum approach to managing sepsis. Is it really in the interests of patient safety not to take further safeguarding action at this time? Without a formal commitment to update the definition of a medical device as I have proposed, I have to say that I remain concerned about the nexus of issues that Amendments 112 and 113 seek to address.

In writing to me, the Minister also helpfully sought to address the relative merits of including reference to the term “artificial intelligence” in the Bill, which is relevant to the review that Amendment 83 would require to take place within 12 months of the clause coming into force. Here, however, I must be candid and say that I am less persuaded still of the stance that the Minister appears to have taken, because he himself acknowledges that Her Majesty’s Government intend to

“examine the options to strengthen the regulations for software and AI”,

echoing his commitment in response to my contribution at Second Reading.

Does this stated intention to examine the options perhaps reflect the fact that the pertinent MHRA guidance is non-statutory and currently “under review”, with further non-statutory guidance anticipated by the end of the year? Or is it perhaps the case that the Minister is concerned about how decisions regarding “edge cases” are taken or monitored in practice in the absence of formal “accountability and regulatory measures”—a situation that is potentially exacerbated by the lack of a clear definition of “medical purpose” in the Bill? Certainly, these are all issues with which I am concerned.

Finally, why would the FDA and pertinent bodies in the EU be looking at how to treat artificial intelligence in relation to innovation in medical device development were it the case that AI was already adequately defined? I wonder: might the Minister have in mind an alternative form of words to Amendment 83, calling for a review over the course of the coming year, that would provide me with the comfort that I seek if I am not to reserve the right to revisit my amendments on Report?

I know that the Minister is absolutely committed to patient safety as well as to facilitating innovation in the development of medicines and medical devices, as was made plain only recently when NHSX launched its digital technology assessment criteria for health and social care in beta. However, there has perhaps never been a more thoroughgoing appreciation of the inextricable link between health and wealth than is the case at present, during the ongoing public health emergency. Algorithms that impact the exam results that people receive impact their life chances. We must ensure that algorithms with the potential to impact people’s health, and which could even represent a threat to life, are appropriately regulated. As such, I hope that the Government will make good the Minister’s commitment to champion patient safety, in parallel with taking steps to ensure that the UK is among the best places in the world to develop, test and deploy an array of next-generation medical devices. I thank noble Lords for their patience in listening to this long speech and I beg to move.

About this proceeding contribution

Reference

807 cc464-7GC 

Session

2019-21

Chamber / Committee

House of Lords Grand Committee
Back to top