Healthcare Leader’s Editor Victoria Vaughan interviewed three leaders in the AI, data and healthcare landscape. Managing director, Government, Healthcare and Life Sciences at Microsoft UK Jacob West, who discusses how tailored versions of copilot will improve clinical care, chief intelligence and analytics officer at Greater Manchester Health and Social Care Partnership, Matt Hennessey, shares details about the integrated care boards (ICBs) new data platform, and director of healthcare for BT’s Business unit Professor Sultan Mahmud, discusses how the communications company is playing a role in the way AI is used in the NHS.
Victoria Vaughan (VV): What is Microsoft doing in terms of AI and healthcare at the moment?
Jacob West (JW): It’s an exciting time particularly over the last year. There’s two things Microsoft is doing in terms of AI. We are building generative AI capability into every part of our technology offering from teams and outlook through to Bing and Xbox: every part of the Microsoft technology stack and we use the phrase Copilot to describe what that means. And really it’s a new category of software, a virtual assistant to augment human capability in potentially every realm of life and will have a very huge potential for healthcare as for other industries.
The second part of our strategy is building bespoke AI capability with customers using both generative and other forms of AI. You can think of that as building copilots for customers themselves, but that draw on the foundation models that we’ve developed through our partnership with Open AI.
VV: What kind of impact could copilot have on a GP or a system leader planning care?
JW: There’s all sorts of different copilots. And not all of them are yet available to everybody but I suppose the one that’s probably most relevant to people who are familiar with everyday Microsoft applications like, Outlook, PowerPoint, Excel and Word is M365 copilot.
That will allow you to comb all of the data that sits in what we call the Microsoft graph. So all of the data and all of those applications, and then allow you to do some very cool stuff on top of that, like generate a PowerPoint presentation from a Word document or other input from scratch without any further input from yourself.
It could be your analyst by looking at Excel and telling you what worksheet might tell you, summarise discussions in teams and lengthy email chains and propose a response. Both clinical professionals and other knowledge workers in the NHS could potentially benefit from that.
There are one or two others. Microsoft search capabilities are now powered by a large language model. So you could use as both your own internal data, but also for external search. And it is a new form of search. So that that’s a powerful tool potentially.
And then one other I’d highlight is power platform. So that’s our low code, no code development tool, and it’s really popular in the NHS. How do you build an application to shrink a physio referral workflow? Something that might take quite a long time to do in analog or even in typically in digital form, but you can shrink that right down to a very short process. And copilot allows you to essentially build an application without any programming skill and using very natural language. Those are some of our cross-industry copilot capabilities I think could have NHS specific productivity potential.
VV: Are you currently working with the NHS on tailored versions of copilot around specific projects?
JW: Firstly, as part of the national agreement with the NHS, there are some steppingstones towards introducing that capability. The first step is teams premium, which does things like intelligent recap and that’s part of what the NHS has bought through that agreement.
We are working with frontline NHS organisations to develop really exciting use cases, I can’t get too specific about that, because it’s commercial in confidence. We do have a healthcare specific copilot that’s already in market, not yet in the UK, but my hope is that it will be very soon it’s called DAX copilot. It stands for Dragon ambient eXperience copilot. We bought a company called Nuance a couple years ago, which is the world leader in voice recognition software and conversational AI. And they already work with the NHS, with GPs actually, but also hospital trusts and other parts of the NHS to take away the sort of drudgery of writing up patient notes.
But the DAX copilot takes that to the next level, it is an automatic transcription in a very structured way, of the conversation between a healthcare professional and a patient. It turns into an accurate medic medical notes and writes it into the electronic patient record. It allows the healthcare professional to focus on the on the patient. It’s technology enhancing the patient care and experience. I suppose you can think of that as our first healthcare specific copilot.
VV: Is the DAX copilot being used the US?
JW: Yeah, it’s deployed in the US. We’ve integrated it into our partnership with Epic. So it’s built very slickly into their patient record now as part of our partnership. But it should be coming soon to the NHS.
VV: So can you discuss some healthcare specific examples?
JW: AI is not something we started on since ChatGPT launched. We’ve been working on some really exciting use cases, with classic AI capability, which is probably worth touching on and then some more emergent work drawing on the power of generative AI.
In the sort of work that has a longer lead time. A great example, is what we’ve been doing with Northumbria. Where couple of Orthopedic Surgeons have used machine learning techniques on historical data for surgical outcomes for orthopedic surgery, so hip and knee transplants and so on, to produce very individualised patient risk profiles that help anticipate the potential post-surgical complications. This also helps doctors and patients think about the best setting for a patient’s treatment, which is important at a time where, waiting lists are super high.
VV: So this is moving further towards personalised care?
JW: That’s exactly the way to think about it. Another example is around breast cancer. There was a huge backlog in breast cancer screening built up during the pandemic. We know, generally we’re not good enough in the UK at early diagnosis of cancer.
So we’ve been working with a company called Kheiron Medical Technologies, who’ve built an AI deployed initially in around Aberdeen with NHS Grampian, but they cover about 25% of the NHS now. That’s more accurate and more efficient than a radiologist on their own in looking for cancers.
There’s actually a new study out in Nature recently, using some data from Hungary, which showed that they identify 5-13% more cancers [early detection of mostly invasive and small cancerous tumors], through this technology, so that if you gross that up that’s thousands of cases being identified.
We always say ‘keep a human in the loop’ when you’re thinking about AI technology and healthcare. it’s not about replacing the radiologists outright, but you could, have Mia (Mammography Intelligent Assessment) which is their product, plus one radiologist, rather than two radiologists. And if you think about the efficiency of the workflow, then again, that’s usually, productivity enhancing.
That’s two examples that predated some of the more recent work on generative AI. And then we’ve got a whole host of emerging use cases, both in the UK and overseas, that sort of bracket into a few different areas: giving time back to clinicians; engaging patients in more meaningful ways; taking out some of the administrative drudgery; and enhancing research and supporting medical training.
In every part of that healthcare value chain, I would say there is opportunity for this technology, particularly generative AI to, to allow us to think differently, and more efficiently, about how we approach healthcare.
VV: What’s the future potential of AI in healthcare? Where do you think this could take the clinician and patient in terms of how they will be able to deliver and receive care?
JW: There is opportunity at every stage of the patient journey for every part of the healthcare system. If I was to boil it down, I’d highlight four things. How can we, with others, deploy technology like AI, but also the surrounding technology needing to underpin it to make it easier for healthcare professionals to serve their patients.
We talk a lot about giving time back to care. Making that specific, DAX copilot technology that can give hours back to busy doctors and nurses. And healthcare systems around the world, not just the NHS, can’t hire enough people, they can’t hold on to people. People have got pretty burnt out and giving them time back is a hugely important currency. The new kind of capability we’ve seen in recent years and through some of the things we’ve been talking about today has the potential to do that in ways that we couldn’t have foreseen previously.
The second area then is how we personalising care. It’s engaging with patients and citizens in different kinds of ways. Maybe one example, the average reading age of each patient in the NHS is nine and most medical jargon is really difficult to understand, even for people with a reading age much higher than that, and this kind of technology can really at very low cost cross that, by turning discharge summaries or, other medical outputs into plain English or presenting it very imaginatively for children. That’s really enhance care.
Then probably the third less exciting area is around administration. How do you take away the administrative cost tasks? How do you automate repetitive tasks? How do you make email management easier? How do you make scheduling easier. It is really important and allows more time for the patient clinician encounter.
Looking even further beyond that, healthcare has been plagued by issues of data interoperability. And that makes those sort of friction costs in healthcare, potentially significantly less.
And the fourth is around research and the life sciences. AI is probably more advanced there than it is in healthcare delivery, for drug development and clinical trial design. Again, at every stage of the value chain, there’s really exciting opportunities there as well.
VV: What is your current are of focus in terms of AI and data in the Greater Manchester ICB?
Matt Hennessey (MH): One of the things we’ve just embarked on is to create a longitudinal patient record, which is de-identified. We have data for every registered patient in Greater Manchester and it tells us not just about the interactions they have in primary care, but in adult social care, mental health services, cancer services, acute hospitals, secondary care more generally.
We can see what prescriptions are done, we can see A&E attendances, 999 and 111 calls. We’re getting a complete picture on a patient’s individual pathway without actually knowing who that patient is. And because of that, we can cohort up patients that have some similar characteristics.
We could put together all the longitudinal records for a white male, over 50 years old who has a stent fitted, who has been to A&E in the last six months? You’ve got a cohort there. And we’re able to see for that cohort, what are the antecedents? How many of them saw their GP in the previous three weeks and we can after their A&E visit – what was what was their general discharge profile, were many of them discharged home, where many of inpatients, what were they prescribed?
We can start to see patient flows for cohorts of patients, that can be defined either by socio demographics, so we can understand the inequalities between different communities, or age groups, ethnic groups, or by a collection of medical conditions. We can see what the standard patient pathways are for people who have both diabetes and cancer for example.
VV: Is this in operation now?
We’ve built the capability to do it. We have had a linked data set for all the secondary care data. So mental health, prescribing data, A&E and we’ve had the shared care record, but only in [September] we had to make a specific application to the national Confidentiality Advisory Group (CAG) to be able to link the primary care data with all that secondary care data. This is something we could do during covid but post covid we have to go through a process to set aside the common law duty of confidentiality, which would allow us to do this.
We’re in the process of making that that linkage but what that will enable us to do is not just to identify the pathways and the those cohorts, but if there is a direct care you case, then a clinician would be able to re identify that anonymised patient so that they can provide direct care. And we’ve got a couple of projects on in development that specifically speak to that ability of being able to identify the risk of a population. And then the clinicians being able to re identify individuals in that population so that they can mitigate the risk that population has.
VV: And does this cover all patients in Greater Manchester?
MH: Yes, all patients who are registered with a Greater Manchester GP. There’s about 2.8 million residents in Greater Manchester, and there’s about 3.1 million registered patients.
VV: Did you did you have to go and ask them to opt in to this as well as a shared care record?
MH: When we made the application to CAG they wanted to see exactly what communications we’d have with patients and the public. They weren’t going to sign this off unless we had a really good engagement process. A lot of this is based on opt out rather than opt in. There are different types of opt out. Essentially, you can opt out of your data leaving the GP practice. That’s what we call GDPR lockdown people. The next is that the data can leave the GP practice and go into the shared care record, but it can only be used for direct care. The final opt out is that it can leave the GP practice, it can be used for direct care but it can’t be used for secondary use, or research. That’s referred to as the national opt out, because at the same time the data comes out of the GP practice to us as a system, it also goes into the national reporting. And so you’re effectively opting out of the national teams using it for secondary use and research.
We’ve had to try and explain all of that complexity to the public, but we’ve done it through the use case stories that show the benefits of data being used in the right way, providing assurances about the security of the data.
We do talk about data leaving practices and it makes it sound like data is flowing everywhere. And what we’re trying to do, particularly in Greater Manchester, is to get away from the model of data sharing and talk about data access. The data sits in one place, known as secure data environments, and the analysts go to the data rather than the data going to the analysts.
VV: How are you using this data?
One of the things we’re concerned about is, the longer people are on a waiting list, there’s a potential for individuals and their condition to deteriorate. We wanted to use the data to produce a kind of risk assessment as to who was likely to deteriorate the longer they wait.
We’re able to create a risk stratification that highlights those people who are at highest risk due to the nature of the condition, but also their external circumstances. Maybe they don’t have carer support. So you’ve got a high risk cultural cohort of people who might experience health deterioration, the longer they wait.
But there’s two ways we can address that by surfacing that intelligence. Either the clinicians who are managing the waiting lists can consider whether it’s worth reprioritising, moving down the waiting list to try and reduce the likelihood of that deterioration, or, more importantly use the data to support community teams to say, ‘This person is at high risk of deterioration while they wait. These are the factors that might drive that deterioration. Can you provide wraparound support? Can the GP practice or community teams provide wraparound support to mitigate the likelihood of that that risk coming into fruition?’
VV: How far back does this data go?
MH: Each service has a has existed for a certain length of time. We’ve had a shared care record of some sort for about 11 years. But not all GP practices were signed up to it. And I think since the pandemic, we’ve got 99.9% coverage of all the GP practices, since 2021. In terms of the data, that GP dataset has 4 billion rows of data. It’s a huge amount of historical data. Some services have only come into being since virtual wards but that’s only a relatively new thing. The idea is we start as far back as the patient record starts. And then we just supplement with whatever service provision data we have, since that kind of start point.
VV: Who are you working with, who is interpreting and managing that 4 billion rows of data?
MH: The only people who are working on the data itself are NHS analysts who work in my team. We’re working closely with universities. We are working closely with local government and NHS providers.
And in particular, we’re starting to do a huge amount of work with the voluntary, community and faith and social enterprise sectors. There’s a big piece of work on to understand what data and capability they have and how we could share that. Because we all benefit. We’re doing quite a bit of mapping work to try and bring the systems together so that we’ve got a really rich picture that’s available at a personal level.
VV: What technology are you using to interpret the data?
MH: One of the things I’ve tried to do for the Greater Manchester system is we’ve got an analytics and data science platform (ADSP), an often when people describe platforms, there are a proprietary single supplier platform. You could mention Cerner or Graphnet.
Our platform is supplier agnostic, I describe it more as a tool belt. We have something called Snowflake, which is our data cloud. So that’s our data orchestration engine, where we’ve got another component, which is Matillion, which is what moves the data around and helps transform the data.
We use DataRobot as our enterprise, machine learning and AI tool. We use Tableau to visualise our tools. And there’s a whole suite of components in the tool belt. And the idea from my perspective was that we didn’t get a kind of vendor lock-in or there wasn’t anything that was too sticky. If our visualization software seemed not to catch up with the with the latest, we could just swap it in and swap it out without destroying the entire infrastructure.
VV: So you can see, patient’s interaction with the health and care services along their journey. How do you then translate that into direct care?
MH: As part of the platform, we’ve created a sort of single front door, because you’ve got all these technologies, and there’s always different stakeholders. And the idea is everyone should go through the single front door, have a login, that login identifies the legal basis that you have to operate.
If you’re logging in as a GP, then you would have the legal basis to identify patients in your practice. If you’re logging in, as a strategic manager, planner, or commissioner, you wouldn’t have the legal basis because you’re not providing direct care. It’s not designed for patients to access it, it’s more about the public sector access, but all the people going in there will get to see the data that they need to do their job.
We’ve got visualisations and presentations of the population health level for where in Greater Manchester is the community with the highest risk of this disease or this deterioration, you can visualise that if you go in as a clinician, and you say, there’s this 30 People in this cluster of high risk health conditions. I want to see who they are, so I can phone them up and bring them in and do some blood pressure checks or whatever it may be.
VV: What can you see being the possibility of this work? What are you aiming and hoping for?
MH: I’m hoping that we, we get to a point where we automate what can be automated, and we use AI in in that in that space. But that actually, we recognise that there is more data and more information than anyone could ever consume. So where we need to move to our workforce is in the interpretation and the application. It’s the intelligence bits of the data to decisions. You can you start with data. If you transform it, you can make it into something presentable, which becomes information. Then, if you link it with evidence, and you link it with wisdom, and you connect it to clinical and patient experience, you turn it into something that’s insightful and intelligent and that’s the thing that makes a decision work.
Victoria Vaughan (VV): What potential do you think AI has for healthcare?
Professor Sultan Mahmud (SM): AI has huge potential in healthcare because of the ability to analyse large quantities of really complex data. But trust is the most important thing.
The approach we take at BT is to make sure it’s clinically robust, technically safe, and it fits in the workflow as well as having patient confidence.
If you think about AI, there are different sorts of trust. There’s technical trust – how have these algorithms been built, and can we trust the maths behind it? Then there’s settings-based confidence – is the technology accurate and how does it interact with the clinician? And finally, do the patients have trust in it?
VV: BT’s Vanguard programme is about setting things up from the clinical perspective, rather than designing tech in isolation and then finding it’s not used. So how does that approach work?
SM: Normally, with product development, people come up with an idea, develop the product and then it’s presented to the NHS.
But my own experience of being in the NHS – I came into BT during the second half of 2021 – is that it works a lot better if you co-create. So, we’re working with a range of partners and NHS organisations, really getting into pain points to look at what’s causing problems. We have an ecosystem of partners – people who will work with us to bring solutions for real life problems.
VV: How are you using an AI in your solutions with healthcare now?
SM: One thing we’re working on is a Care Close to Home programme – remote monitoring technology in primary and community care form the basis of that.
And from that, there’s anticipatory care where we have AI-enabled devices and applications looking at datasets and contextual factors. The clinician can decide whether there is a need for an anticipatory approach – for example, do we need to give some prompts to Mary’s carer around a particular situation or explain that if this happens then do that? Or do we send out a healthcare professional in advance of the patient arriving back at A&E or GP practice?
We are also looking at diagnostics in secondary care. That has always been a bottleneck because there’s demand and supply issues of access to tests and access to clinicians’ processing information – these things take a lot of time and effort. There is huge potential to save clinician time and workload and provide better access. It’s a priority area for BT health because it’s amenable by technology.
We’re working with a company called Deep C on a radiology AI platform that can help clinical professionals access multiple valuable AI tools to help in the diagnosis process. It’s positive in terms of workload reduction for breast screening. And it helps in chest X ray analysis with faster reporting times and, with fracture detection, the sensitivity is increased.
We’ve got live trials in two places at the moment at big NHS organisations.
VV: When could the diagnostics product be ready as an actual offering to the NHS?
SM: I wouldn’t want to give a timescale because when it’s something as new as this, there are so many things to consider, such as whether the input data is of good quality and so on. We need to understand the bias in these systems.
I wouldn’t say it’s going to be years, but equally we’re not going to rush through it. We want to get this right and test it. It needs to be worked through with patients and clinicians before taking it out further.
VV: Do you have a plan for where you might go with AI?
SM: There’s huge opportunity around medical imaging in terms of image analysis, detection of diseases and improving the accuracy and speed of diagnosis.
Patient interaction is another area. We’re developing a solution to help patients communicate with various parts of the NHS, from their GPs to their hospital consultants.
We’re also thinking through how we can educate by using AI powered chatbots and virtual assistance. Then there’s remote monitoring technology and anticipatory care, which I’ve already mentioned.
VV: How do you think the way BT works with NHS organisations will change?
SM: Increasingly, the NHS wants us to be a trusted partner – an interface between tech businesses and the NHS.
SMEs and startups that are developing AI are looking for an organisation to help take them to market. And we say: ‘Absolutely, we’ll do that, but you’ve got to comply with NHS clinical and operational standards, and you’ve got to abide by our ethics and delivery policy.’
We’re also looking at how that fits with infrastructure and connectivity. As demands for data and connectivity increase, we need to make sure that the resources and infrastructure needed are ready.
VV: What are the common things that the tech industry – the people creating products – misunderstand about NHS requirements?
SM: I wouldn’t say that they misunderstand. It’s just that, often, product development is presented to the NHS as a kind of decision. It needs to be more of a two-way thing. We need more understanding of the problems and working with the NHS.
We also need to think about how we expedite things. The speed to market is a problem. Things move quickly and people are forging ahead in countries like Estonia, Israel, and China.
VV: Is the regulatory process in this country stopping things from moving faster?
SM: The regulation needs to be really smart. We need to understand where regulation ends and delivery happens.
When you develop technology, it needs to combine workforce strategy, operational pressures, and future demand. We need a digital transformation strategy that understands there are gaps and vacancies in the workforce. And one that takes on board what’s going to happen in the next 10 or 20 years in terms of multiple comorbidities and patient cohorts because demand is going to increase.
If you’re going to develop a national strategy around making this tick, you need the best brains in the world in tech and healthcare working together.
VV: How are you working with integrated care boards and NHS system leaders? Do you want them to approach BT? How do you see that relationship developing?
SM: We want to hear from ICB leadership about their problems. We know that adoption of technology can feel hard at the start – the funding and how fast it moves – because there’re so many headwinds facing people.
We’re open for business, but we’re also massively open to serve and working in partnership. It’s not just about profit for us. We want the NHS to succeed and, whether it’s AI or any other technology, we’ve got some really sharp minds here.
Our approach is to understand before being understood. Across the country, we have account teams that, for the first time in a generation, work purely with the NHS. They don’t work with commercial organisations like the financial sector, for example – it’s only the NHS. So that kind of in-depth knowledge and focus is the way that I want BT Health to progress.
VV: You’ve talked about everything coming from the bottom up so that specific problems are solved. But at the same time, the tech needs to be able to speak to other bits of tech and eventually it all needs to work together. What is BT’s role in that interoperability space?
SM: We live in a market economy, there are multiple players. We can’t say that everybody has got to be on the same system.
At the same time, if there are hundreds of systems, and they don’t talk to each other, the patient’s going to say: ‘I thought you were one NHS?’
As we develop things with our partners, we make sure of interoperability standards as well as the data and ethical considerations. If you’re being paid for by the NHS, you’ve got to open your data channels. There are things we can do in procurement and contracts to make that happen.
Interoperability is also one of the key things we deal with in Vanguard organisations.
VV: The more we become digitally enabled, the more connectivity becomes a health issue. Does BT Health do any lobbying to make sure that rural areas have the high-speed connectivity they need?
SM: If we’re serious about remote care, anticipatory care and care close to home then we need to make sure rural areas are connected.
So yes, we are lobbying but it needs more. It’s about needs assessment and the problem statement. So, if you were to say, that the top 5% of patients who were using 50 to 60% of resources across primary, secondary and community services could be addressed with better connectivity then that’s a case easily made.
We’ve got to work together to build that case. If anybody wants to work with us on that, we’d be open to it; having two data points to create a narrative would be very helpful.
