18
Examination of witness
Professor Andrew Morris.
Q73 The Chair: Good morning. Welcome to the Science and Technology Select Committee. We are undertaking an inquiry on innovation in the NHS— personalised medicines and AI—and we are very pleased to have as our first witness this morning Professor Andrew Morris, who is director of Health Data Research UK and president of the Academy of Medical Sciences. Professor Morris, I would like to start by asking you to set out for our committee what Health Data Research UK does and how it fits into the wider health data landscape, and what, in your view, it has achieved since you became its inaugural director in 2017. Perhaps also, can you say what is the current state of the UK’s health data infrastructure? What works and what does not work, and where are the most significant gaps, in your view? Over to you, Professor Morris.
Professor Andrew Morris: Thank you, Lord Mair. Can I just say that it is a real privilege to be here today? I am Andrew Morris, a doctor, and a professor of medicine at the University of Edinburgh and vice-principal there. I am seconded to London to do two jobs. One is to be director of Health Data Research UK, which is the national institute for health data science. I also have the privilege of being president of the Academy of Medical Sciences, which is one of the four national academies in the UK.
I should say, just to frame this, that I fell into data about 35 years ago, actually when I was working with Lord Patel in Dundee. I was a diabetes doctor, and I realised that working with colleagues, especially general practitioners, and by following journeys of care of individual patients across primary, secondary, tertiary and social care and joining up that journey with data, was not only good to improve care, but we reduced amputations. Lord Patel will remember we reduced them by 50% in four years and reduced blindness by 40%. One could use the same data to drive public health, clinical trials, research and innovation, and genetic studies, so data as infrastructure across a population. That has endured today in Scotland. There are 360,000 people living with diabetes who are supported on a daily basis using such systems.
With HDRUK, the ambition is to do that for 67 million people. It is the national institute for health data science, and its aim is to unite the UK’s health data to enable discoveries that improve people’s lives in a trustworthy way, for 68 million people[1]. To do that, we have a triple aim. First, we want to accelerate trustworthy data use by sorting the data, which is a huge challenge. Secondly, the aim is to improve people’s lives by unlocking the power of data, by using use cases, programmes which show the art of the possible, deliver tangible research benefit, but also clean the pipes of the infrastructure.
Lastly, this endeavour requires collaboration arguably not seen before. We are trying to shape the future of health data research globally, through infrastructure, data, services, governance, and most importantly, public engagement. The show has been on the road since 2018. I must credit the funders: the Medical Research Council, NIHR, British Heart Foundation and Cancer Research UK, who came together around a national endeavour. Otherwise, we had complete fragmentation of funding across the ecosystem.
As you know, the UK has potential huge opportunities in health data research because of its data assets. If I may, Lord Mair, I must just outline this. We talk about health data and are very focused on the NHS, whereas I like to consider health-relevant data, because a lot of the determinants of poor health lie outwith the NHS. If we talk about buckets, we have got nationally collected data such as NHS England cure rates and hospital admissions, for example. There is then regional data such as laboratory data and imaging data. Then, you have got other buckets—for example, the administrative data the ONS holds, or other administrative educational data, which have huge relevance. Then there is research curated data, such as from UK Biobank and Our Future Health, which you are going to hear from. These data assets are not held in the NHS. Finally, we have got environmental and sensor data. By 2030, there will be 30 billion devices globally attached to the internet of things, and many of those will be collecting data relevant to health. So that is the vision and the ambition.
Since 2018, we have achieved research impact, improved data access, capacity building and public engagement, and we have collaborated at scale. For example, we have performed the first ever study on 67 million people during the pandemic, which showed the safety effectiveness and the uptake of vaccines. Secondly, you will be familiar with the Cumberlege report around valproate, vaginal mesh and hormone pregnancy tests. For valproate, we have run a study to look at how it is being used following the Baroness’s recommendation and localise that across the UK—so drug safety. In total classic academic outputs, the team has published 5,000 articles and 300,000 citations, so the academic side of this is robust.
Improving data access is so important, and I will come to this again and again today. We have made some headway. Anyone globally can search the UK’s data assets of over a thousand datasets of 100 data custodians by searching on a single shop window—the metadata. In terms of capacity building, we have 60,000 learners from 143 countries on a virtual learning platform. In terms of collaboration at scale through our British Heart Foundation data science centre, we have 400 researchers across 50 institutions looking at cardiovascular disease across the UK. Those are just a few examples of how we have tried to harness in a trustworthy way the huge potential of health data research in the UK, but we are still in the foothills of where we want to be.
The Chair: Thank you, that is a very helpful start. You will know that during our inquiry so far we have heard about the NHS data as being potentially one of the greatest assets for personalised medicine and life sciences research. You have mentioned many datasets. Do you have a view as to which are the other most important ones to identify and collate? Is there a wish list of the most influential and important datasets, in your view?
Professor Andrew Morris: It is like the layers of an onion. There are core datasets, which I would suggest are absolutely vital for the UK to get the foundations right—for example, hospital admissions datasets. As you know, when anyone is admitted to hospital, coding takes place according to the ICD-10 coding system, so the so-called hospital episode statistics in England or the SMR dataset in Scotland are key.
Primary care data is absolutely vital across the UK. Today in the UK there will be a million consultations with our primary care colleagues. Every one of those contributes important information across that journey of care. There is also prescribing data, not only of therapeutics, but also devices, which is also a key data asset. In the UK, we spend £20 billion a year on medicines. The opportunity is to have real-time surveillance of safety, effectiveness and value of the use of those medicines and devices. Imaging and laboratory data in the UK are the key assets. Those are core.
A key point is linkage, because when one can link other datasets to those foundational datasets, whether they are genetic datasets or cohort datasets, you do something very simple in epidemiological terms, which is linking exposure to outcome. That longitudinal opportunity is where the UK really flourishes. One example of that, which you may have heard about, is that we sponsored the Moorfields data hub—called INSIGHT, of course, because it is an eye hospital. We worked with the fantastic team there to curate 1.4 million optical coherence tomography scans: the anatomy of the back of your eye. Then, the team there put a large language model across that dataset, and through linkage was able to predict incident Parkinson’s disease, seven years before it occurred clinically. So there is one example of a specialist dataset, which, when linked to these other datasets which define outcome, can reveal true scientific insight that is important for patients.
The Chair: I have one more question before my colleagues ask more. What was the experience from Covid in how health data was treated? Were there any lessons learned from that in your view?
Professor Andrew Morris: What we saw with Covid was a glimpse of what could be possible, not just for revealing insight in the pandemic. One of the first things we achieved in Covid was scale. We were privileged to be invited by Lord Vallance, who was the Chief Scientific Adviser at the time, to lead a national core study on data across the four nations and with the Office for National Statistics. That allowed us to assemble 93 data assets across the four nations. Being able to run studies at scale across the UK reduced fragmentation.
Secondly, what we saw in Covid was speed. You will remember in the early days of the pandemic that the initial recommendation was that we took the AstraZeneca and the BioNTech vaccines twice, but within eight weeks we were able to show that single-dose vaccination reduced hospital admission by 85%, and that changed national policy and freed up vaccine supply. We also showed, for example, when we were at the threat of another lockdown—you will recall it was because Delta lost its growth advantage to Omicron—that Omicron was far more infectious but had much less morbidity, so we did not need to lockdown.
What we saw in the pandemic was the ability to use data at scale. Also, the recovery trial, as you know, randomised thousands of people and we showed that dexamethasone was the treatment of choice for people with severe pneumonia heading for the intensive care unit. The key to be able to use data at scale was use of confidentiality of patient information. All these separate data controllers across the UK were encouraged to share data for a common purpose—the COPI purpose—and that allowed us to use data at scale safely and effectively. Sadly, however, I would argue that access to data has gone backwards considerably since the pandemic.
The Chair: That is interesting.
Lord Patel: Andrew, good to see you again. You are looking not much different.
Professor Andrew Morris: I have taken up running, you see. He does not recognise me.
Q74 Lord Patel: That is why I am making the comment, Andrew. Anyway, to go back to business: you know that the Government are investing £600 million in the Health Data Research Service, so I have a three-part question. First, what role will HDRUK play alongside this Health Data Research Service?
Professor Andrew Morris: That is a good question. First, one of the main stimuli of the Health Data Research Service—by the way, it is fantastic news for the UK that the Government wish to invest in a Health Data Research Service—was the review led by Professor Cathie Sudlow, who I think is appearing later. As you know, at the time she was chief scientist at Health Data Research UK, and our team supported the production of the review. In terms of the Health Data Research Service, her recommendations are first that we need to see data as infrastructure. Secondly, we need a national service, especially in England, where data are made findable, accessible, interoperable and reusable. Thirdly, we need clear and accountable leadership embedded in the Department of Health. Fourthly, it should be UK-wide. Lastly, the service should accredit trusted research environments across the UK.
This is a timely investment. It has the potential to maintain our international competitiveness and is distinct but entirely complementary to Health Data Research UK. Health Data Research UK focuses on the innovation, science and research which sit alongside the required service of data assets across the UK.
Lord Patel: Where does the data they would allow people to use for research come from?
Professor Andrew Morris: The Health Data Research Service, as you know, was announced by the Prime Minister a year ago. It has appointed a terrific chief executive; I am not sure if you are seeing Dr Mel Ivarsson. She has been in post two months. The service is currently setting out its stall. I would anticipate that the main focus will be data from the NHS, especially in England, but that it will also work with the devolved Administrations. I anticipate that national datasets would probably be an initial focus.
Lord Patel: We have heard, for instance, from Professor Birney, that already there are suggestions that expectations for the services that will be set up are unrealistic and that it will take much longer. What is your view about that? Do you agree?
Professor Andrew Morris: I would not underestimate the size of the challenge. There are some principles which will enable success and I am happy to go through them. First, build on what already exists: look at other data ecosystems in other industry verticals. Secondly, do not assume that if you build, they will come. Use cases need to be a focus for delivering societal value and the potential of the service. Thirdly, there needs to be user-centred design. Who is going to use the services? Is it going to be the research community? Is it going to be the Government? Is it going to be industry? Is it going to be the NHS? User-centred design is going to be key. Finally, a key principle is design for connectivity, not monopoly control. If one looks at other data ecosystems, it is about addressing the interoperability of existing data assets rather than creating a huge national data lake. The metaphor I use is power stations on a grid. How do you join it together? I could give Lord Patel £100 in the next 10 minutes because his bank and my bank, although highly competitive, both subscribe to the same messaging system called SWIFT, which ensures that rapid, secure communication of data. The Health Data Research Service will need to look at those standards of interoperability so that data assets hosted across the UK can be joined together.
Lord Patel: You have mentioned that, and I now understand the challenges it will face. What would the metrics and timeline of success look like?
Professor Andrew Morris: I would suggest there will be qualitative and quantitative metrics. In terms of qualitative metrics, demonstrating trustworthiness and maintaining the social contract with the public will be key. There is also the active participation by data controllers who want to subscribe to a national Health Data Research Service, bearing in mind that there are literally thousands of data controllers across the UK of health data. That participation and maintaining trustworthiness are two qualitative metrics.
In terms of quantitative metrics, going back to data access, time from application to data access is going to be key. The impact of the number of multi-dataset research and innovation projects enabled will be an important metric, as will, most importantly, measurable patient benefit and efficiencies within the NHS. So there is a scorecard of metrics which will need to be developed.
The Chair: Just to follow on from Lord Patel’s question on timeline, you have talked about metrics, but what in your view is a realistic timeline for the delivery of real benefits?
Professor Andrew Morris: If we build on what we have got, and if the Health Data Research Service under its excellent leadership accelerates existing activity, I would hope we would see success in 12 to 18 months, but we all need to get behind it.
Q75 Baroness Willis of Summertown: Just a quick follow-up: I am not a medical scientist—I am a biologist—but HDRUK and HDRS are the two acronyms that have been used. Do you think it would have been better to have a different title for these research services than what has been provided? Your explanation of what the differences are is very helpful, but I am getting confused, and people looking from outside will really get muddled. My other follow-up is that I understood where Health Data Research UK comes from and understand the different organisations and sorts of datasets that are in there. But where does something like UK Biobank then sit? Does it sit with you, or with the Health Data Research Service?
Professor Andrew Morris: In terms of taxonomy?
Baroness Willis of Summertown: Exactly.
Professor Andrew Morris: I must say, I meet the chief executive of the Health Data Research Service every two weeks, and I know this is something they are actively considering. A principle I always subscribe to is that we need to reduce complexity and have transparency with the publics. We are completely open to having a very sensible decision around terminology and being clear about purpose and functions. The way we are provisionally expressing it is that we are the research engine of the Health Data Research Service.
Baroness Willis of Summertown: But what if you were going to rename it? Its title, unless I am really confused now, is the Health Data Research Service. How does that then align with what you just said?
Professor Andrew Morris: It is something that we should address.
Q76 Lord Winston: It is very good to see you here. I was going to write to you about the question I am about to ask, so it is quite convenient and I feel rather annoyed that I did not write to you in advance. You said in your opening statement that we have improved public engagement. That was the bland statement you made, and I was very interested to explore that aspect of the academy’s work. It seems to me that we have certainly increased our attempts at public engagement, but I do not think we have changed public trust very much, and although we have tried metrics on that, there have been things that have been missing.
Some years ago, the research council set up training for talented young individuals who were clearly at a high level in science and might actually do fellowships in this area. At that time, there was also a committee looking at public engagement directly with the Prime Minister. I think it was chaired by Lord Vallance at the time, but I am not certain about that. Walport was one of the people involved with that, and Paul Nurse was also certainly on that committee.
However, we do not seem to have a good idea of why people are so resistant to accepting so much of this. The Government still make slightly foolish suggestions about how we are going to change our use of data, particularly in the area of DNA, but it is not at all clear that this is actually necessarily properly policy-driven. I wonder whether you would comment on some of those issues, which are causing increasing concern.
Professor Andrew Morris: It is a very good question. The golden thread of our own work and that of the Health Data Research Service has to be to demonstrate trustworthiness in society. We actually draw a lot from Baroness Onora O’Neill, who you will all be familiar with. Often, one reads that we are going to build trust. Actually, you cannot build trust, because trust is earned or it is given. We talk about demonstrating trustworthiness, which is about being honest, reliable, and competent in the way that we use health data.
In terms of putting that into practice, in recent years, true public engagement and involvement have become more disciplined and have been woven into initiatives from day one, rather than being seen as an afterthought further down the track.
Thirdly, bringing the behavioural science, social science, legal and ethics skills and expertise into health data research is something that we have done. One example is deliberative public engagement, where one works in partnership with the publics. For example, over a weekend, there were two very prominent events, one in London, one in Manchester, where one works with the publics to talk them through these nuanced issues around data access, data security, safeguards, benefit-sharing and industrial data access. The suggestion of those deliberative consultations is that polling may overestimate the opposition. With deliberative research, there is a more stable social licence for research and commercial partnerships—as long as public benefit is clearly articulated, stewardship of the NHS is clear and reciprocal, and NHS value or benefit share back to the system is explained.
Lord Winston: Do you remember the attempt to undertake what was called dialogue at the time, when we looked at groups of people who had no particular background in the science and highly qualified people at the same time? Then, we looked at how opinions could change in a working environment where it was relevant to social issues.
Professor Andrew Morris: I am aware of it but not in detail.
Lord Winston: It was supported by Government. I do not think it got very far, but it certainly produced results. Access to synthetic biology was one thing; nuclear waste was certainly another thing that was discussed. There were a number of other medical issues. I wonder, therefore, whether you felt that sometimes the academy could perhaps do more about that itself.
Professor Andrew Morris: I am glad you are writing to me, because one of our five new priorities for this year is to increase public trust in medical sciences in an era of global misinformation. One example is that we are working in partnership with the Royal Society to look at vaccine hesitancy in society, because, as you know, we are seeing outbreaks of diseases such as measles, which we should not be seeing in the UK.
Addressing public engagement is an academy priority, but being a president of a national academy I am keen that the national academies do things in partnership, because the social science expertise resides in the British academy. I think this is where the national academies could and should stand up and try to be the trusted, independent voices in society, especially around misinformation in science. That programme is currently being scoped and we are seeking to fund an anti-misinformation programme within the academy.
Q77 Lord Drayson: Professor Morris, good to see you. I would just like to go a little further on the issue of public trust in the data. During your tenure leading this, as you say, really important drive within the UK there has been a huge change in geopolitics. You touched on this, I think, in your comment just now, mentioning that one of your priorities this year will be addressing global misinformation to maintain public trust. Now, of course, for data to be to be stored and recorded, that requires software and the UK has over the years predominantly chosen software from the United States for its electronic patient record systems, and has now chosen Palantir for its federated data platform. We have seen in the context of recent geopolitical challenges very clear feedback from polling of the difficulty the public has with the direction of the US Administration under President Trump and the alignment of many US technology companies with that agenda. I would really like to hear your opinion as to the wisdom or otherwise of the NHS choosing software from companies aligned with that worldview.
Professor Andew Morris: The NHS is not a software house. It is not a chipmaker. It is not a hyperscaler. To my mind, the future of medicine and healthcare delivery will be increasingly data-driven and computational. It will therefore be essential to work with technology providers who provide these skills. In terms of who to choose, I would have a principles-based approach that would articulate how you make the choice—
Lord Drayson: What would those principles be? For example, if you are using a large language model—which, as you have suggested, has very successfully analysed eye data at Moorfields—the way in which a language model is trained and the biases embedded in it are in the hands of the people who train the model. Therefore, if the United Kingdom chooses only to use US or Chinese models for the analysis of its health data, it is importing the biases inherent in those models. Is that not the case?
Professor Andew Morris: This is a live discussion. Shall I go through my principles first?
Lord Drayson: I think that would be helpful, yes.
Professor Andew Morris: This is effectively about choosing a partner for national data infrastructure, so my sense is that we need to set a very high bar. The principles are, first, that the NHS must remain the unquestioned data controller. Secondly, radical transparency of any contract and its extension is really important. Thirdly, and it touches on your point, there should be explicit anti-vendor lock-in architecture. Vendor lock-in means a company making it very hard to exit a contract. What does that mean? It means open APIs, documented schemas, exportable ontologies, mandatory interoperability standards such as FHIR—sorry, I am getting a bit technical. Fourthly, the privacy-enhancing technology part of the stack has to be independent of the prime vendor so that the privacy characteristics of the stack are not within the same workflow, so you have an additional safeguard. Touching on the geopolitics, what I would call ethical supplier suitability must be continuously reviewed. Finally, and I think we are witnessing this, the adoption of technology—artificial intelligence—into the health service is more about the people than the technology. Workforce trust and partnership are a key part of the adoption.
Lord Drayson: Given the undoubted controversy at the moment with regard to the Palantir contract, with the BMA suggesting that its members should not support the use of the Palantir system, and the fact that from the polling we are seeing that patient trust is being eroded, was the decision to award the Palantir contract consistent with your principles?
Professor Andew Morris: I think that is a decision for others to make.
Q78 Baroness Nicholson of Winterbourne: Thank you for the most interesting points you are giving us. I have a couple of questions. First, inevitably, these large organisations able to offer platforms for such research are likely to be fairly linked with other Governments, the USA particularly. Our system is fundamentally 100% different from the USA’s. How does this inhibit you in using US-based systems such as Palantir? It is 100% a different system. Here, our data effectively belongs to the Government because the NHS is Government, even though it is not assembled in that way at this moment. None the less, it is seen that way by the British population and is controlled and owned by government funding. Is it not rather difficult to use another system based on a commercial understanding of how data is used? The only country which has our particular system is New Zealand. Do you not think we should build up our own platform in order to research our data and not rely on outside ones, even though this would take longer and would be expensive?
Professor Andew Morris: It is a very good question and I think it aligns to Lord Drayson’s point. The question is here is: should we be assembling sovereign AI and capability? My own view is that this is a prerequisite for UK ambitions, and we are seeing it in other countries as well, such as Switzerland, Vietnam and Singapore. It would also potentially provide a platform for home-grown AI companies, adoption and skills. As you know, the world of AI is moving so quickly. Its growth is exponential. We are also seeing a tension between the American model, which is made up of hyperscalers, large language models and heavy compute, and the Chinese approach, which looks more to open models and so-called small language models, which are much more agile. So we are looking to a future that will potentially be dominated less by the hyperscaler frontier systems.
To my mind, that will give the UK an opportunity, because these small language models are much less compute-demanding and can be deployed locally. Bearing in mind that the feedstock of AI is data, the opportunity for the UK will be in curating data and possibly contributing that to a sovereign AI initiative. This is under active discussion at the Council for Science and Technology, for example, where Nigel Shadbolt from Oxford—you may know him; he is a national authority on this matter—is suggesting that, first, the two domains where this may have the greatest potential are defence/cyber security and health.
Secondly, he says that we need to anticipate agentic AI, which is where AI governs workflows and decision-making within systems. He has a lovely phrase that I quite like: computer programmers are moving from being bricklayers to being architects. However, if we are to succeed in this place, we will hit a brick wall if we do not curate our data and do it well. Having data in an electronic health record does not mean that you have AI-ready health data. That curation piece is key; it is why the National Data Library and the Health Data Research Service are going to be so important if we can curate data to that standard so that we have a ground truth upon which we can develop sovereign capability.
Baroness Nicholson of Winterbourne: My follow-up question is the reverse of the same coin. In essence, our democratic system is based on a mistrust of Government—in other words, the desire always to change someone for somebody else whom you believe may be better, and then you want to change that one. It is based on an essential mistrust of anyone who has authority over you, which is extremely healthy, in my opinion.
This means that, in terms of persuading the population to trust the Government on owning data, it does not really matter whether or not the Government release it because nobody will trust it anyway. How are you attacking that problem? Is it by using the data, or is it by pointing out the benefits of some particular merging of data? How do you tackle that while respecting the fact that this is the way the population think automatically?
Professor Andrew Morris: Sorry—I did not catch some of that.
Baroness Nicholson of Winterbourne: How can you build trust among the population when, in essence, the public are correct not to trust the Government, whoever they are? How do you segment your work, in that sense, when it is already government-owned? A&E—the emergency room—is the reverse of that, as it were. How can you build trust? What are you standing on to create that trust?
Professor Andrew Morris: First, we are demonstrating trustworthiness by being honest, reliable and competent in what we do. Secondly, we involve the public in the design and delivery of projects. Thirdly, the primary benefit has to be a public benefit, whether it is a reduction in heart attacks or new treatments for dementia. The narrative has to be about the patient and public benefit.
Baroness Nicholson of Winterbourne: Might it be worth considering trying to diminish the public’s belief that health data is private, particularly when you look on the box and you see OnlyFans or something like that— which I hope you do not? Is it really necessary to think that your personal health data is absolutely private for ever when it is almost impossible to keep it private? Are you part of that discussion?
Professor Andrew Morris: As a doctor, I would say that most things in medicine are about risk versus benefit. It is about maximising the benefits and minimising the risks. In recent years, we have articulated the safeguards in the system to protect the privacy of data.
One example of that, working in partnership with the Office for National Statistics, is the so-called Five Safes framework. We have done public consultation. The public understand and support the Five Safes. The first is safe data, which are anonymised. The second is safe places because, traditionally, we have had lots of data travel. As soon as you let data out the door, you do not know what happens to it; this goes to the concept of a trusted research environment. The third is safe people; this means that we know which people are accessing data and for what purpose, as well as whether they are accredited. The fourth is safe projects, which are those that are in the public interest. The fifth and last one is safe outputs to minimise disclosure control.
We now have systems that maximise the benefit, in terms of the value to the public and patients of the use of data, while minimising the risk. Having that constant dialogue with the public is absolutely vital.
Q79 Lord Ranger of Northwood: Professor Morris, you are obviously thinking deeply about this area and are embedded in it. For me, the most impactful thing you have said—among many great things—concerns the results of the eyeball scan and the potential identification of Parkinson’s seven years in advance of it occurring clinically. That is amazing when we think about it as one tiny example of the many things that could be found out. Is it not the case that those kinds of outcomes and use cases can really get the public on board, in terms of trust, when they see that kind of value from a health perspective?
The other side of that is the perhaps more controversial point about health data. How can the value associated with the data for commercial companies be leveraged? The Health Innovation Minister said recently that the Government will leverage the UK’s data “for the benefit of the Treasury coffers”, as well as that of patients. So we see the patient benefit, but we are now looking for a potential commercial benefit to the Government as well. However, we are also seeing many UK research companies doing excellent work but not being able to exploit that because they cannot access the data.
So we have a challenging conflict between public benefit and potential government income, with UK businesses struggling to commercialise the innovation that they are delivering here. What do you think is the best way to think about how health data research can be managed in order to get both the outcome that we are looking at and the commercial outcome, particularly from a UK perspective, for UK businesses? Do you think that the Government’s approach to that at the moment is the right one?
I was taken by a couple of the examples you gave. You mentioned how you can move data from one place to another, just as a bank moves money via SWIFT. As someone who looks extensively at the future of money, including digital money, I see that we are moving away from the traditional because the bureaucracy of the big banks sits behind that process. We may not want to replicate that, so is there something that the Government should be doing to find a new approach in how we use this data and how we can get the right commercial and public outcomes?
Professor Andrew Morris: The Government have made some positive strides in this area, but there is still work to do. That is my sense of things. I will maybe talk about two things: first, industry partnerships and how you frame them and, secondly, value. On how we work with industry—it is not just the pharmaceutical industry but diagnostics, devices, analytics and other things—public benefit must come first in any relationship. Secondly, transparency that a lay person can understand is absolutely vital. The third thing is these safeguards that I talked about: coming to the data is key.
A key thing that we learned from our public and patient involvement is that reciprocity and fair value return is key to public support for this activity. Value can come in various guises. It can be monetary value—for example, back to the NHS—and it could be equity or royalty stream value. It could also be getting a technology first to the NHS or getting it at cost price. So value comes in different dimensions.
We in the UK need to build up a series of use cases with the publics to demonstrate how the use of data in a trustworthy way not only creates public and patient benefit but also has an economic value. On a benefits hierarchy, I would say: put patients first, NHS and society second, and the Treasury third.
Lord Ranger of Northwood: Thank you for that. That is excellent. If you look at it from a patient perspective, many of the things you said about benefits for the NHS do not necessarily get felt by that individual patient who is probably being asked for a transaction around their data. We may need to look even closer at how patients actually feel some tangible benefit. If there is a monetary benefit to the NHS—forgive me, but all we tend to hear about is the NHS needing more money, and more money being given to the NHS—it might just fall into a big money pot that does not really impact the debate about what the comeback is.
Professor Andrew Morris: What we are talking about here—it is not just within the health data domain—is the value of public sector data as an intangible asset in the UK. There is a sense of urgency about this because, with the development of AI Foundation models, the right to access datasets will be absolutely vital to this value creation. So this is an area that needs more nuanced thinking on how we have fair trade rather than free trade.
Q80 Baroness Jones of Whitchurch: That leads nicely into my question. You have already explained the huge opportunities of shared data and given some really exciting examples of all that. But on top of that we have AI, which is with us now—it is not something for the future. So, in your role in your academy, how are you beginning to face up to those extra opportunities and challenges of AI, which lie on top of all the good work you have already done? It is almost a bit of a Wild West now—not quite controlled under the rules that you might like. Do we have the right infrastructure for controlling AI in the use of our NHS data? Can you perhaps give us some good, positive examples of where you think that will really enhance the opportunities for you in analysing data?
Professor Andrew Morris: That is a very good question. The first point is that AI is here in health. AI has capabilities across what I call the clinical thresholds of specialist performance, such as reading X-rays. An AI algorithm can perform as well as human specialists can.
The second thing we are seeing is an investment surge in AI in healthcare. It is estimated that, of an approximate $300 billion global spend, $40 billion to $50 billion is on healthcare. In some digitally mature health systems, we are seeing shifts from AI pilots to scaled implementation in screening, clinical decision support and operation workflow management—so it is here. The challenge we have is how we regulate that in a trustworthy way and, again, look at the risks and benefits.
The Government have a lot of policy in this space: they have the AI Opportunities Action Plan and the AI for Science Strategy, and AI featured heavily in the NHS 10-year plan and the Life Sciences Sector Plan. So there is a policy wrapper around this. As ever, the challenge is how we deliver in a scalable and reliable way. There is an excellent report from Cersi-AI—the centre for excellence in AI and digital health—suggesting that there is no consensus currently on what constitutes acceptable risk in this space. The sophistication of both attacks and defences is continuing to evolve.
This is an area where we have a good policy wrapper, but more practical thinking needs to take place about how we have reliable governance systems, especially as agentic AI, which will affect clinical workflows, is governed in practice.
Baroness Jones of Whitchurch: I know that some people feel that we should have regulated this earlier, rather than just having good policies. Do you think that we have sufficient controls now over what is happening with the application of AI and health data? We can all imagine that there are huge advantages, but now that we have given that information away and large language models have been trained on it, we have kind of lost control of some of that information now, have we not? How do we get it back so that it is under the very good principles that you have been outlining?
Professor Andrew Morris: Again, it is work in progress but, through the same organisation, there is an AI readiness assurance framework, which is a step-by-step approach to identifying potential causes of harm, mapping risk controls and then identifying the gaps. But this needs to be sustainable and scalable and to evolve into best practice. At the moment, it is work in progress.
Baroness Jones of Whitchurch: Do you see, in your experience, examples where you feel that we have lost control of the data—that it has been used for one purpose but now you find that it is being applied for another purpose that it was not originally intended for?
Professor Andrew Morris: That goes back to the Five Safes principles: the purpose of data analysis has to be clearly articulated and justified in the public benefit or the patient benefit.
Q81 Lord Patel: I am changing the subject slightly. We have heard in our inquiry so far that genomic data—for instance, genomic data from sequencing—has a role in developing personalised medicine. But we have also heard that genomic data on its own has limited predictive value or power. So, for individuals, it has less predictive value, but, linked with that and used in conjunction with other datasets, it might be more valuable too. How far do you think the UK has got with data linkage with genomic data or genetic data, what needs to happen to unlock it further, and what is the scale needed for personalised medicine?
Professor Andrew Morris: The UK is world-leading, I would say, in its large genomic programmes. I estimate that probably up to 10 million people in the UK, which is remarkable, have consented to participate in genomic studies such as UK Biobank, Our Future Health, and the REACT study run out of Imperial College. The power of these studies is enhanced remarkably through linkage. It goes back to my previous point around exposure to outcome. Being able to link genetic characteristics to future health outcome allows prediction at scale. That linkage is absolutely key, and we are beginning to see some of the benefits of that with UK Biobank and Our Future Health, for example, with the recent direction from the Secretary of State for linkage to primary care data. The opportunity has to be taken in a trustworthy and well-governed way. It may well be that the Health Data Research Service takes on board enabling that linkage to health outcomes, whether it is primary care health outcomes, secondary care health outcomes, or other data assets. Linkage in itself is vital and we do not have the current systems and services in place to make that routine for these large genomic studies, especially around primary care data linkage.
Lord Patel: Does that mean that this vast amount of information we may have about genomic data is, on its own, not much use for individual personalised medicine?
Professor Andrew Morris: I think it is being used, but its future potential is immense and the opportunity is to put the infrastructure data linkage services in place to realise that potential.
Q82 Baroness Willis of Summertown: I just want to move on a little to barriers to use of this data, particularly in research. As the president of the Academy of Medical Sciences you have this unique perspective on the research landscape. Do you think the UK is training enough data scientists, bioinformaticians and health data specialists? It is a very different skill set to the ones that we normally have had going through the medical sciences division in my own university. I would just be interested to know whether you think there is a knowledge gap or a workforce gap developing there.
Professor Andrew Morris: It is a good question. I have a couple of comments. We are living through an amazing age. We are seeing a fusion of biology and medical sciences with computational, physical and social sciences and the traditional way of training the workforce does not meet that need, so I would say we are probably not training enough people. It is about a new generation of people who I call T-shaped people. They may be medical or may be in the nursing profession, but they understand data science, bioinformatics, governance and ethics. There is a need for future hybrid roles. Secondly, we are seeing demand rapidly increasing in this area in terms of, as we have talked about today, NHS digitisation, AI and healthcare genomics and precision medicine, so the demand is very strong. We have great universities and should get behind them to develop this next generation. Future doctors will not look like me; they will be increasingly computational. Of course, there is a challenge of retention because of pay structures and the attraction of industry, for example. I think the UK has a strong base, but we will not realise many of the things we have discussed today unless we have the workforce, the skills and the talent to lead it on for the next generation.
Baroness Willis of Summertown: Can I just ask one follow-up question? It is partly from personal experience from my own research group, who try to use the UK Biobank. Because we are not medics, the time and the layers of bureaucracy to get access to some of these large datasets is prohibitive to research, because it can take up to 18 months when you have a three-year research grant. I was very taken with your comments about the Covid experience and you feeling in some ways we are going backwards, and I wondered how much of that is because of layers of bureaucracy. I totally understand the need to have confidentiality, but do you think we are getting so bogged down with the bureaucracy that it is not really providing the research capability it could do?
Professor Andrew Morris: 100%. Data access is arguably one of the biggest challenges we face. It takes 18 months, two years, three years to access data. We are seeing UK researchers and UK companies going to the United States to access data. In November last year, because I always think look internationally for best practice, I brought John Halamka here. John is the chief information officer of the Mayo Clinic. He and his colleagues have put 14 million Mayo lives on to a protected Google cloud across Wisconsin, Michigan, Florida and Arizona and have an abstracted de-anonymised dataset. They have reduced data access from 18 months to 18 hours. That should be our ambition from the UK, with good safeguards and constraints.
Q83 Lord Winston: I have a very quick question. Some years ago one of your predecessors advised this committee that we should be supporting clinical academics more effectively. Do you think that is still true? My view is that it would be helpful.
Professor Andrew Morris: I completely concur. The academy is working in partnership with the Medical Research Council, NIHR and other funders. My argument is that the innovators in the healthcare systems are clinical academics. Whether medics, nurses or physiotherapists, they are bilingual and understand the deep-domain clinical problems that patients face but also understand the research method. In the last 10 years in the UK we have seen a decline in clinical academics. There are 6% fewer clinical academics today than in 2012 while we have seen a 50% increase in NHS consultants. In general practice only one in every 200 general practitioners is a clinical academic. I think this is an important area otherwise we will not have the workforce, the ideas and the commitment to drive innovation into the health service. Today’s research, as you know, is tomorrow’s care.
The Chair: Lord Willis wanted to come in online.
Q84 Lord Willis of Knaresborough: What a fascinating morning this has been and thank you for that. One very simple question I would like to ask you is about data and data access, particularly as larger amounts of data are now being used via AI. Most of the data you have talked about this morning is raised in the UK—I presume virtually all of it—but do we obtain any data at all from the US, China or the EU to be able to build data levels in order to get results?
Professor Andrew Morris: That is a good question. What we are seeing internationally is that the UK is not unique in wanting to assemble its data in a trustworthy way for public benefit, societal benefit and economic benefit. We are seeing initiatives such as Health Data Research UK develop internationally. For example, we have a partnership with the Government of Singapore, with whom we share best practice. They have adopted many of the principles that we have in the UK around governance, infrastructure, public engagement and data standards as part of the “national trust” programme that they are leading.
However, we are also seeing federation, which means that, understandably, there is a great reluctance to allow the egress of data beyond sovereign boundaries. The key to international data analysis, research and innovation will be the ability to leave the data in situ but send the code to the data in an integrated way that acts on standardised datasets. What we are seeing is the development of federated networks of analysis, rather than data travel. I think that, because of geopolitics, federation and the ability to run analyses across different countries without moving the data will be a major focus of future research and innovation, rather than the data.
In fact, the European Health Data Space has adopted the “trusted research environment” concept that the UK has promulgated. We are seeing Health Data Research UK-type entities developing across European member states; they will rely on federation.
The Chair: Lord Berkeley, your question has already been covered, I think.
Lord Berkeley: Yes, thank you.
The Chair: Professor Morris, you have been wonderful in answering so many questions. Our final question comes from Lord Verjee.
Q85 Lord Verjee: Thank you very much, Professor Morris, for your very detailed and interesting presentation today on a world of real and deep exponential change, both technologically and geopolitically. As you know, our inquiry will ultimately make recommendations to the Government on how to accelerate innovation in AI and personalise medicine in the NHS and the life sciences sector generally. From a bird’s-eye view, what would be your key priorities? What key recommendations should we make sure we include in this report?
Professor Andrew Morris: Thank you for your time today. I have a few thoughts on that. Remember that the feedstock of AI, which is the main focus of this inquiry, is data. Treat data as critical national infrastructure and get behind it. Sort the data. I am delighted for the HDRS; we need to give it its best chance of success. Its resources are tenfold what HDRUK has, so it represents significant government investment—something like £600 million. At the same time, though, I remind the committee that the NHS spends £600 million in a day. That would be my first recommendation.
My second point concerns data access. How can we, in a trustworthy way and with clear safeguards, put in a stable legal framework for timely and safe data access? Eighteen months is unacceptable. We have programmes that are going to have to halt because we cannot access data.
My third point concerns priority datasets, which were suggested by Lord Mair. Work with our general practitioner community to fix the primary care data gap along that journey of care by using use cases that deliver value to patients and general practitioners.
Fourthly, build on what we already have and what already exists, such as our secure data environments, our policy structure and our public engagement.
Finally, and most importantly, demonstrate trustworthiness in everything we do by working with the public in a very open and transparent way. Also, embed involvement and engagement through the dialogue programmes, which should be continuous rather than periodic.
The Chair: Professor Morris, you have been wonderful in answering so many of our questions. We very much appreciate it; thank you very much indeed. We are going to suspend the session now because we have Professor Sudlow coming in online. Thank you for your time.
