Collaboration between DeepMind and Taunton and Somerset NHS Foundation Trust

This content relates to the following topics:

Article information

  • Posted:Tuesday 06 February 2018

Speaking at our event Partnership Models in the NHS on 6 February 2018, Andrew Forrest, (Chief Information Officer, Taunton and Somerset NHS Foundation Trust), Dr Dominic King, (Clinical Lead, DeepMind Health), Luke Gompels, (Joint Chief Clinical Information Officer, Taunton and Somerset NHS Foundation Trust), and Ashley Sumner, (Senior Nurse Adviser, DeepMind Health), speak of the collaboration between Taunton and Somerset NHS Foundation Trust and DeepMind.


AF: Andrew Forrest
DK: Dominic King
LG: Luke Gompels
AS: Ashley Sumner

AF:    Hi, good morning everybody.  I’m Andrew Forrest, I’m Chief Information Officer for Taunton and Somerset Foundation Trust.  I would like to just introduce people.  At the end there we’ve got Dr Luke Gompels who’s our Join Chief Clinical Information officer, and from DeepMind we’ve got Dr Dominic King here and right in the middle there is Ashley Sumner who’s the Senior Nurse Advisor for DeepMind.  And we’re very pleased to come here and talk about our developing partnership with DeepMind.  So what we want to cover today, and we’ve got quite a busy thirty minutes, is just to introduce our two organisations, we want to talk about how our partnership came together and set out our combined approach and pick up some of the challenges maybe.  We want to demonstrate a bit of the first solution, just so you can picture it, and then Luke will take you through a quick vision of how patient care could look like if we can actually get it right.  

    So Taunton and Somerset Foundation Trust, we’re in the South West of England in the middle of cider country.  Some cruel person said, “It’s where you stop on the way to Cornwall.”  It’s actually a lovely place to live and to work.  It was actually built in the Second World War by the American Army and was handed over to the NHS in 1949 and became an NHS organisation.

    Here’s a bit of a thumbnail sort of sketch of the organisation.  It’s got 630 beds, it’s a medium-sized acute hospital, very busy place, as all the hospitals are at the moment.  What is unusual is it’s just formed an alliance with the Community and Mental Health Trust right across the Somerset Partnership and also runs a GP practice.  From the CQC report it was judged as Good and the care is Outstanding and the IT and Improvement teams work very closely together and were also singled out by the CQC as an example of good practice.  And that, together with our implementation of the open MAXIMS electronic patient record, meant that we were successful in gaining funding as a digital exemplar site, which means we’ve got money to implement systems for patients and clinicians which are best in class with digital technology.  And we’re also working to link our information up to create workflows for actionable analytics so we can predict what’s going to happen and look at what’s happened in the past. We’re also looking to link together primary and secondary services and consider for the citizen how we best use technology as preventative care.  

    So we’re working in this complex environment and we’ve concentrated so far mostly on the right hand side, on inpatients and outpatients, with our electronic patient record but we’re working with DeepMind in terms of looking after the deteriorating patient.  So patient care is at the centre of everything we do.  We work very closely with primary care.  All of the GPs in our area use the EMIS system and we have an EMIS viewer in our Emergency Department so we can see the medication that people are on when they come into ED but we’re taking that further and we’re linking our systems together so they’re fully integrated.  And, as I say, we’re looking at the possibilities – and, clearly, companies such as DeepMind are ideal for this – where we can look at wearables and apps to help people stay well.

    It is really complicated though and, although we’ve done much of the bottom part of this diagram within the hospital, there aren’t well established standards that we’re able to interlink different systems to create electronic patient records, analytic engines and give patients a portal so they can access their own information.  But what we do want to do as healthcare providers and social care providers is work together with charities and with innovators such as DeepMind to create those solutions of the next few years.  

    So with that I will hand over to Dom.

DK:    Thank you, Andrew.  So I think many of you will have heard of DeepMind, and just to give you a little bit of background I was probably here about a year ago and a lot’s changed since then.  So for those of you that may have been here a year ago, just to give you a bit of an update where we are, we have a pretty grand mission statement at DeepMind which is to solve intelligence and then use it to make the world a better place.   DeepMind was founded in 2010 and was acquired by Google in 2014.  We say ‘joined forces’ though because we retain a lot of independence at DeepMind; the decision to work in Health, the decision to partner with Taunton is taken in London and not in California.  When I spoke here last year we were entirely based in London, although we have opened new offices in California and two offices in Canada in the last six months, and the team has grown very substantially so we now have a team of over 700 people from 40 different countries, the vast majority of which are kind of academics, PhDs, postdocs, mathematicians, scientists, neuroscientists.  

    Broadly, DeepMind is split into two parts.  About two thirds of our organisation is working on this goal of artificial general intelligence.  These are algorithms that are capable of learning from kind of raw data and being generalisable, so the algorithms that could power some Google products and services are also used potentially in medical imaging.  And then about a third of us, including me and Ashley, sit in the Applied Team and it’s our role to think about how we deploy this technology safely and effectively in areas that we think are important, and the first of which is in healthcare.  

    DeepMind’s kind of scientific progress has been remarkable over the course of the last few years. Three years ago we published our first Nature paper which showed that our algorithms could play Atari games, very simple games like Breakout and Pac-Man.      Two years ago almost exactly to this day we beat the world champion at Go, the ancient Chinese game, which was watched by about 250 million people – arguably the most complex game ever invented – and then a year ago we showed how our systems could develop dynamic external memory.  Two months ago we showed how our algorithms were able to, without any training, beat the world champions at Go and in recent weeks we’ve been able to show how the same algorithm is the best chess player of all time.  And an important thing is that that is now done without any training whatsoever.  So previously when we showed how we could beat the world champion at Go our algorithm learnt from watching many hundreds of thousands of games of Go; now, within a few hours, the systems are able to learn from themselves just by playing repeatedly over and over again and then reach this kind of world class standard.  So this has some very interesting applications potentially in Health where we’ve previously thought that a lot of the work we do requires extensive training material and that may be less necessary in the future.

    It’s really important to say these algorithms are already having a big impact in the world.  We have a team at DeepMind called DeepMind for Google which is effectively applying these algorithms in Google products and services and AI is now driving many of the services.  Whether you use Google, Amazon, Android, Apple, deep learning approaches, reinforcement learning approaches are driving many of these services.  But these are very digitally mature environments; we can’t really say the same is true necessarily in healthcare, which will explain, I think, some of the work that we’re doing with our partners at Taunton.

    So we launched DeepMind Health in early 2016.  The vision at the time was that...  We really felt that there was a huge potential for artificial intelligence in healthcare but actually we needed to focus on the practical as well as the smart.  Now, we are doing artificial intelligence research in healthcare, particularly focused at medical imaging.  We have three UK or NHS partnership that we’ve announced, the first working with Moorfields Eye Hospital looking at retinal scans, second working with University College London hospitals looking at radiotherapy planning and the third working with Imperial College London looking at mammography.  There were actually articles in the Financial Times yesterday and today and a number of papers about some of our work in this area.  We think over the next couple of years our organisation and many others will be demonstrating very substantial progress in areas like medical imaging.  

    The wider application of artificial intelligence in the NHS is going to be really hampered by the current lack of digital maturity.  So, you know, when I graduated from medical school 15 years ago I was given a pager - thought it was pretty cool at the time, became less so a few years later – used paper lists to organise the work I did, prescribed on paper.  In the vast majority of hospitals in the NHS that is still the case and the idea that advanced artificial intelligence of the types we’re building at DeepMind and are being built by many other types of organisation can somehow be dropped into this environment and be safely deployed and effectively deployed is not actually a credible idea.  

    So alongside our artificial intelligence research we’ve been working with a number of NHS partners to help support them in the best ways we can, partnering closely, to kind of move forward in this kind of digital journey that I think we all think is critically important, you know, the move from paper to modern digital tools and systems.  So we have a secure mobile app called Streams which is in live deployment at the Royal Free Hospital, we have partnerships with Taunton, their close neighbours in Yeovil and also Imperial College Healthcare, and some of the things that we’re working on is the ability for clinicians to view test results, the ability to order tests, to manage their activity, to message securely and make sure that that information ends up in the medical record.  

    Just to be very clear, because it does cause some confusion, we absolutely see these two streams of work or strands of work coming together at some point in the future but it’s not part of our current partnerships with Streams for that to be the case. But in the next couple of years, if we have an algorithm that allows us to detect pneumonia on a chest x-ray or allows us to automatically generate a discharge letter that would usually take a junior doctor two hours to write or to identify a patient at risk of sepsis or acute kidney injury, we feel very much that this needs to be surfaced in modern digital technologies.  

    So we’re incredibly excited about the potential for this work without artificial intelligence.  We have very early promising signs from the Royal Free Hospital about the success in these types of systems in speeding up care and we’ll be publishing an academic peer review paper later this year which will hopefully confirm this, but this work and the work of companies like ours, particularly big technology companies, does generate a lot of controversy.  Some of you will be familiar with Information Commissioner’s Office criticised the Royal Free and our partnership and relationship and the use of data here and I would say I’m very happy to answer any questions either in the session that follows or one to one.  We absolutely see that there’s a lot of learning that we can take from our initial partnerships and have brought into Taunton and future work we’ve done, which you’ll hopefully get a sense of as Luke speaks.  

    I’ll just say a little bit about how the partnership started before handing over to Luke.  Sometimes kind of serendipity in random meetings is a very good thing.  I think something that’s incredibly positive about some recent announcements in the NHS is the support being given to creating this new generation of doctors like myself and Luke who have an interest in informatics and technology and actually putting us alongside people like Andrew to think about the clinical impacts of the technologies that we’re all keen to see deployed.  And myself and other clinicians on the DeepMind side met Luke and some of his colleagues and we just kind of got on very well and shared a kind of common vision about how this could work. On that note I will hand over to Luke who will tell you a bit about how the partnership has progressed.  

LG:       My name is Luke Gompels, I’m a rheumatology consultant by background and have been involved in some of the digital work at Musgrove in particular over the last four or five years.  And I share the role of Chief Clinical Information Officer with a colleague, Tom Edwards, who’s a colorectal surgeon, so we split that about fifty/fifty between the two of us, and which is just a great opportunity to be clinically involved in the progression of digital developments within the hospital in keeping with our GDE site.   And it’s just really great to then have the ability to be able to work with DeepMind looking really intelligently at the problems that we urgently need to solve on a very practical level in the NHS, particularly for patient care.  So it’s really that meld between the sort of local and also the opportunities that we can get from this kind of partnership.  

    So I just want to run through some of the things that we thought were touch points as being really important.  A really great opportunity for us to learn from each other but particularly for an NHS organisation to be learning about the way in which we can take design to a different kind of level with user-centred and more agile design processes than may have been in evidence with some of our previous IT developments over the years, to really work hard building on this relationship model with our patients to get meaningful and practical patient engagement, person engagement, coproduction.  We’ve got people involved it the programme who’ve never touched the hospital in any shape or form and patients who’ve got chronic illnesses, certainly from my own specialty, you know, experience of people who’ve got lifelong conditions who’ve got great insight and expertise in how care or how they feel care could be better managed.

    We’re going to talk a little bit about some of the transparency and governance and clinical safety that needs to go on around a much more creative development and how we try and evaluate.  So we want to try and move through these things quite quickly because we want to show you a little bit more about the deployment of this as well and we’ll have a chance to pick up on some of these things in discussion.  

    So looking at the user-centred aspects, we’ve really had a great education about how to look at user-centred design but also apply that particularly to the patient at the bedside, that’s the core thing; a key aspect of hospital work is the unwell patient, how to identify a deteriorating patient, how to escalate care.  These have always been issues and remain so, and it’s surprises some people who’ve not been in that environment for some length of time that this is still the case but recognising when people are unwell and giving the opportunity for staff and individuals involved in that care to do the right thing because that’s what everybody wants to do.  

    So this is a simulation involvement, and we’ve done lots of these and you learn a lot about yourself and a lot about your organisation by doing these kind of things, and patients learn a lot about the care that needs to be delivered.  So it’s been a really great creative process for our Trust.  And we’ve then looked at particular individuals in real detail.  This is an example of an F2 who’s under considerable pressure in their day to day practice.  All sorts of things come out of this but you can look at things like “So much of what I write is already written and captured somewhere else,” you know, “Why are we using our most intelligent staff to act as scribes on a day to day basis when they need to be by the bedside?”  All these little things act as touch points for how we look at developing an intelligent solution.

    More importantly, looking at the bedside, a day in the life of a nurse at Musgrove.  Absolutely stunning insights into the level of interaction that individuals have with patients.  You can see actually just by way of a pointer, you can’t quite look at the bottom there but the red dots are when doctors are on the ward – not very often.  So there’s all sorts of opportunities for different healthcare professionals to be involved in the care of the patient, things that we can make great advantage of, outside of just this work itself.  So it’s been a lovely opportunity to look at the kind of roles that individuals have within a Trust, what are their pain points, what motivates them at work, how can we gear those individuals to do a better job.  Part of that is providing the right digital technology and then really apply that to an agile design process. 

     And I’m talking through these slides like I know all of this but, you know, this is where our partnership has given us a great opportunity to learn some of these techniques.  And actually some of these things we’re running in parallel with other types of more traditional work so it’s not that it’s been a rigid process that we’ve gone through, but these are just some of the examples of things that have been brought to bear, bringing everybody together, looking at particularly priorities, recapping on where things could be improved.  A particular thing is lots of different individuals, and I don’t want to use that term too much but stakeholders have different viewpoints about where things should be improved, and so there’s an opportunity to collaboratively vote on the most important aspects, bringing those together to form some kind of meaningful solution.  And, importantly, testing prototypes as early as possible so that we can start saying whether that looks clinically right or whether we have to actually rip up our own models of care and look at something different.  And testing and learning, so this is a sort of day five kind of, you know, one to five being a sort of week, is really important.  But we’ve put in a slide here about how we’ve brought that to bear in terms of local practice, looking and working with individuals - so the top panel there is some of our HCAs, the middle panel is some of our patient partners and in the back there is one of our nurses, and the bottom panel is one of the Hospital at Night team - and then actually dovetailing that in with real data so that we can actually look at some of the information collected by the bedside so that we can quantify some of the aspects of what’s happening.

    Meaningful patient engagement is absolutely essential and we’ve all been involved with that at different levels.   The question we had to ask ourselves as an organisation is “How do we bring that to bear so that our information is shared appropriately with patients, patents have security, people have security that their information is being used for the right purposes?”  And really emphasising the point that we are very concerned here about the direct care of patients, the unwell patient, and how to make them better in the simplest possible form and that in that we have legal responsibilities and duties to look after their data in the most secure way.  And it’s been great to work in a partnership that’s been so open and transparent about that and being able to take that right to individuals themselves and have that direct discussion.  So that’s been a real learning curve and started us to move to different areas that are a bit outside of our comfort zone.  

    So we’ve done some town centre events.  It’s actually quite interesting when you just go up to somebody and they say, “Oh would you like to talk about your care at your local hospital?” and they say, “Well who are you?” and I say, “I’m one of the consultants at the hospital,” and they say, “Well why aren’t you in clinic?”  So you know.  And you say, “Well actually we’re here because we’re trying to improve things.”  And people have been really enthused and engaged by this approach which has been absolutely fantastic and given a real sort of sense of energy to the team.  We’ve got one of our critical care ICU nurses in the top panel there and we’ve got one of our patient partners, Leonard, who’s given us some fantastic insights to some of his own care within the hospital which has given us a real driver to change things for the better.

    We have a legal responsibility to be able to look at things in terms of the ICOs Dom’s touched upon and it’s been really helpful to do that at all sorts of levels and make sure that we’ve got the right security and the right approach and the right governance around that, and transparency has been a real key point of that.  

    And that leads into sort of clinical safety and making sure that we’ve got the right safety processes because at the end of the day whatever you’re doing you want to be doing it as safely as possible, and that needs to occur both on a sort of more national level and also very much that we’ve got local ownership and testing and hazard assessment.  And through our GDE status we’ve worked incredibly hard over the last number of years and it’s something that we’re very proud about in terms of how we take clinical safety to a level where you’re proactively testing the deployment and the changing process that occurs on the wards and in all sorts of other clinical environments.  And this is an example of our hazard safety testing which is actually trying to flush out problems before they become an issue, and some of that actually examines some of the existing risk which is in the system.  It’s so easy to forget that a dropped piece of paper on the floor is an information governance risk, it’s a clinical safety risk, and nobody is quantifying that, so it’s absolutely a key part of this that we have a chance to reassess the risk, but in doing that sometimes people inflate the risk that you’re then assessing rather than actually analysing the risk that was in the system in the first place.  So it’s a real sort of step change in how we look at things which has been absolutely fantastic.

    And then we’re involved in evaluating how we look at this product in terms of a digital intervention, how can we actually quantify that we’re making a stepwise change in care.  So we’ve talked about the hazard workshops at the top there, we’ve talked about pilots and simulation, how can we make a quantitative evaluation of whether this digital project is actually improving care.  And just one fascinating little piece of early data that’s come out from the work that’s been going on, we’ve been able to feed a little bit from all the things that have been going on to look at this.  And this is one of our colleagues who’s working at Musgrove actually and also in collaboration with Exeter Department of Life Sciences looking at some of the test results that come out from the lab, and if you look at the purple line there you’ve got a peak of test results that come out around about midday and then the blue line here is actually when the test results are viewed.  And if you go to the second panel, which is the red one, you can see this enormous scatter of the time taken to view a test and the variability increases towards five o’clock.  So what you’ve got is a very tight area in the middle where there’s no so many dots and that’s because the tests are taken and seen in a much shorter time period and then everything starts to expand out of that.   So if we can get something by the bedside where you’re taking digital observations and a pathology results viewer, putting them all together so that care is delivered much earlier in the day, as an example, you can start to really reduce the variation in when people are looking at results, which is going to improve their care, make sure that tests are acted on as quickly as possible and make for a better outcome.  So, again, taking this to the sort of numbers level has been a real insight.  So I’ll hand over to Ashley now who’s just going to take us through some more of the detail.

AS:    I have to say I’m delighted, I’ve been given the best bit; I get to show you a little bit of an insight into what the collaboration has been able to achieve in working together.  But before I do that I think it’s really important again just to reflect back on something Dom said earlier.   All of us in all of our organisations start out at a different point in our digital transformation journey and no matter where we are in that journey there is still an awful lot to do.  So one of the first things that the partnership and the collaboration had to consider was “What do we focus on?  What problem in care delivery or what issue should we be best focused and thinking about?”  So we could have been looking at dashboards, results viewing, collaboration tools, roster management, a whole range of things, but essentially the collaboration determined that the focus of this early work should be around the deteriorating patient or the acutely unwell patient.

    And really just to illustrate this, this is really where a lot of us are starting from, lots and lots of records at the end of the bed where the clinician has to try and pick out the salient pieces of information from hundreds, possibly thousands, of datapoints on each individual patient to determine what’s happening and when best to intervene.  And I know probably, like other colleagues in the room, I’m quite sure that we’ve all missed opportunities to step in and do something early enough to really support the patient and make sure that the patient has the best possible outcome.  This is where we have arrived to today and this is some of the work that we’ve been working on that we’re able to present for you.

    So I’m going to hand back to Luke who’s going to take you a little bit more on the journey about where this might actually take us, particularly at Taunton.

LG:    With a co-CCIO as a colorectal surgeon there’s been an exceptional amount of focus on the bowel movements for some reason, I’m not quite sure why, but we’ve tried to keep it to a minimum.  So this is just really more of a thought piece, to be honest, and we could take this in all sorts of ways but you can quite easily see that there’s opportunities across all of the points where a patient becomes unwell to make a real difference to care.  So whether it’s a patient who's collapsed potentially with a head injury or in a diabetic coma that they’ve got some form of wearable sensor that causes an alert, whether that’s a patient in their final years of life or who requires additional care, all sorts of opportunities there.  Looking at the medical record we’ve seen that, pages and pages of notes, so taking the viewing of the medical record so that it actually starts to form a genuine piece of work about what’s happened both in the community and in the hospital and that that care can be delivered much closer to the patient is a real need.  This is like a kind of candy box wish list, if you like.  Looking at the care record so that you’ve got the right information at the right time and at the right place so that you’ve got a presentation that links both your assessment of results and generation of task activity.  And in a sense paper does both of those but it’s discontinuous from both where you are and where the patient is and where your collaborator is.  So you’ve got two individuals that might need to be collaborating on that patient and they’re in different places and the notes are in a different place, and then you’re trying to drive a task from a piece of paper.  So bringing that all together so that the care record can be acted upon would be a real need.

    There’s a huge focus on trying to make sure that care can be organised better, particularly in the current environment that we have, so can we, in keeping with some of Rob’s work, start to predict where the real needs are and change our resources so that we can cope with the demand.  This is going to help a great deal with that.  Task management and notification for those individuals that need to be focused on mobilising themselves to the point of need is going to be an essential requirement and how we prioritise the work that’s available.  Do we have any real prioritisation for our junior doctors?  We have some degree of that that occurs on a manual basis but one individual on one side of the hospital running to the other side of the hospital may not be the most efficient use of resources.  There could be somebody down who’s on a ward who could just be deployed locally, do a quick fix, make an assessment, call somebody and you wouldn’t have to have all of this traffic going up and down.  So our step count might go down but care would improve.  

    How do we record the outcome of our consultations for the best care for the patient and for their own satisfaction, happiness and wellbeing?  How do we triage patients so that we can more intelligently use the facilities that we’ve got?  How do we use algorithms to make sure that we can actually improve care by giving and facilitating the right answers for complex interpretation?  How do we look at imaging so that we get more consistent results, reduction in variation and a more rapid answer by the bedside?  How do we code our outcomes so that we can improve care and not least make sure that we are being able to record what we do?  And, most importantly, how do we empower patients and people so that they’ve got the right information that they need so that they can look after themselves and how can they get advice and help and expertise when they need it?  How do they look at their medications?  How can they review things?  You can start to see that this all starts to tie up with that overarching health need.  

    So on that note thank you very much and we’d very much like to take any questions and open the discussion.