Tom: One of the things, again, particularly bad in health systems, is once we've adopted an innovation - which often comes back to that issue of the zero tariff model not being as successful as I think we thought it might be at the time - is that often, if you add a cost, even an implementation cost, onto what you might rub as a cost cake, very often our model is to put an extra layer on the cake.
Ian Abbs: What we absolutely have to do is to pull a couple of layers of cost out. So we have to build models of substitution innovation adoption rather than just additionality. And that is another thought that I've certainly been having over the more recent years - but how do we start to do substitution rather than addition?
Tom: That is really interesting. I mean, certainly - so we now, almost since the start, but certainly in the latter years, we will never do anything for free. The free pilot is a terrible idea for all the reasons you just explained. The thing being free almost stops the scale. It stops it being on top of anyone's agenda. The more successful projects are the ones where the organisation has committed financially. And then the point about the cake, I hadn't really thought about it like that, but certainly we see business cases for our product where I'm shocked when I see the final number, because it's twice or three times the cost of the implementation and license fees, because of all the internal costs that have been added. And when you buy from DrDoctor, and I think other businesses do this too, we have quite a full-service offering. So you don't get just the software, you get the implementation, like we did at GSTT way back when, and put boots on the ground to go and do the change. Very effective, a high cost for us, but we do it because we know it's the way to get results. And it's interesting when you see an organisation that's bought that from us, then layers additional cost on top again and again and again. It basically makes the thing very hard, it makes the ROI case very poor. Instead, I think the NHS should become a little braver on leaning on its suppliers. Make us carry the bag, make us responsible for the outcomes. Don't just buy the products from us.
Ian Abbs: Well, I think that's spot on, Tom. I certainly think that implementation capability is a very powerful tool for companies to use to increase the number and speed of implementations. Very often there are capable teams within NHS organisations, but they're often, I think, at the top of their capacity curve and they don't have a lot of headroom, so just adding a bit into that is often helpful. I certainly think there has to be careful examination, pre-implementation, of what this is going to allow us to stop doing, and which layers of our cost cake we could remove - or could we do something differently by implementing these new tools that will actually change our cost cake? Because, you know, I'm sure you've seen this a lot, Tom.
Tom: There's that age-old cartoon of a horse-drawn cart with square wheels being pulled along very, very slowly, and someone standing next to it with a round wheel. And the people driving this poor horse and cart are saying, well, we don't have time to change the wheel, and by the way, we've already got some wheels, so we don't need to redesign it. So it's really a point about - it has to be redesigned, you have to redesign the underlying process, not just believe that putting a round wheel next to a cart with a square wheel is going to help. And secondly, you've got to have some capacity, you've got to have the time to do the change, and put some resources in to do that. But I think if you do those things - redesign the fundamental process, put some implementation capacity in, and challenge ourselves, both supplier and supplied, to take some cost cake out - then that's probably a more satisfying solution than thinking you can just put something in and then leave.
Ian Abbs: Yeah. And sadly, that still happens all of the time.
Tom: So that's the SME innovation end of the spectrum. Ian, you're someone who's unusually overseen two EPR deployments - IPM in the early 2000s and Epic in the 2020s. Talk to me a little bit about lessons learned from EPR deployments, and the benefits and challenges organisations face when they do something of that scale.
Ian Abbs: Yeah. So I think firstly, Tom, I'd start with: why do that? Why do you ever put a large and often pretty expensive IT system into a healthcare organisation? Because actually, managing an individual patient through an individual episode of care is actually easier on a piece of paper.
Tom: Yes. Paper's amazingly good for lots of things, actually.
Ian Abbs: And it comes back to that need to rethink what you're doing. Just putting a digital solution, an IT system, into an organisation that replaces a paper process - and all the different types of paper we use, from notes to drug charts to processes run on paper - without redesigning somehow your underlying process model, your operating model, is a big mistake. Now, I've been guilty, at least partially guilty, of that mistake, because redesigning organisational operating models is hard. I don't think I'd ever articulate that the reason to put an electronic medical record, an EMR, into an organisation is about managing a single patient. It does give you some additional benefits, things like prescribing alerts on drug interactions. It does give you some ability to use electronic systems for risk warnings, so early warning scores, things like that. I think the real reason to put these large electronic systems in is, firstly, to think about what your organisational operating model is like, and what you're going to do today, but importantly tomorrow, to change that operating model. Secondly, what does it allow you to do in a digital world that you can't do in an analogue world?
Tom: And that includes the geographic basis of care.
Ian Abbs: Yes, that's interesting. What does it allow you to do about the risk equations we talked about earlier? What does it tell you about the risk of the status quo. And then ultimately, as I think I've learned about other industries increasingly - it's not really the organisation, maybe that's not the right term - business intelligence comes from an understanding of large datasets.
Tom: Not singular pieces of information. Can we draw trends across our sort of -
Ian Abbs: Yeah, exactly. Large combinations of data, analysed over time, which is very difficult to do on singular paper records. So I'll give you one example of that, and I'll come back to what it was actually like. One of the things I believe very strongly in will ultimately be the ability to manage risk in populations over time. So, for example, if you think about the higher clinical risk groups within populations which are outside hospital - hospitals are actually, and I'll say this with all due regard to my colleagues managing hospitals, relatively simple, linear places to manage.
Tom: Yes.
Ian Abbs: As are patient pathways in hospitals - they are relatively predictable, they have a beginning and an end, usually. A much more difficult place to manage is the out-of-hospital space. But the evidence is pretty compelling that, in a geographical population, there is a segment - it could be 10-ish%, something like that - that is at particularly high risk of requiring health and social care resource utilisation. Understanding that group of people, finding better ways to help them have better management of their risk today, and therefore preferably a reduction in their risk tomorrow, and therefore better experiences of health and living than they have today, really requires you to have the ability to interrogate large datasets with predictive capabilities. Now, of course, it's important for those outcomes to be delivered for people. But there's a plus side for the health system here.
Tom: Yeah, that's interesting.
Ian Abbs: That 10% of a high-risk segment - probably we can argue about the numbers - but if you look at the global evidence, probably consumes 35 to 40% of total health and social care resource. If we could improve how that group of people experiences health and care, we have a better chance of containing, and possibly even reducing, the expenditure, so we have more money to spend on patients elsewhere. You cannot do that without predictive analytics based on large datasets. So actually, the big reason I wanted to particularly do our most recent iteration of a large IT system was in fact not about managing the hospital or individual patients within the hospital. It was actually about managing a population, for the future. The implementation itself was hard work.
Tom: Yeah, I bet.
Ian Abbs: I mean, for lots of different reasons, the healthcare EMR software market has contracted. There are relatively small numbers of significant players, despite what were early forays into the industry from many of the technology companies - Apple, Google, and others all looked at the possibility of large organisational software solutions, but none of them survived. The choice within the market is relatively limited. We thought about new solutions pretty hard.
Tom: Yeah.
Ian Abbs: And then, at Guy's and St Thomas', we opted for Epic, an American system well used in academic centres in the US. As you mentioned earlier, its original code was from the late 1970s, and it has the advantages of something that's evolved over time in a unitary product. It also has the disadvantages of something that's evolved over time in a unitary product. We went through a very large implementation programme. We did it with King's College Hospital, our partners in Southeast London.
Tom: To make it bigger and more complicated.
Ian Abbs: And it was the right thing to do, particularly because our populations cover the same geography, coming back to that population management future - but it did make it a bit more complicated. We opted to make it more complicated and go for a big bang go-live. But I think that was the right decision, one that Clive Kay, my colleague and CEO at KCH, and I made together. We went for a big bang implementation in the end, one of the largest, I think at the time the largest, global single go-live in the history of Epic since the late '70s. That initial go-live was successful, but it takes a huge amount of work for colleagues to get used to. It will probably take five to ten years for it to fully bed in. The other thing one has to overcome is understanding the product roadmap for these large EMRs, but also the product roadmap for the agile organisations providing other types of digital services. I've always thought of them as complementary, although not everybody does.
Tom: No, no. And I think that's a live and active debate, isn't it - the system of record innovating itself, versus allowing a constellation of innovations to happen around it. So Epic was a hugely successful go-live. I don't think asking 25,000 people, plus however many people work at KCH, to change happens overnight. Definitely something to be very proud of. The thing we're always amazed by when we look at Epic, and we talk about this internally at Dr. Doctor, is it's one of the few tech businesses that can dictate to its customers how they're going to work. That's something I always find really fascinating - you buy Epic, you get a team of people to implement it for you, but they very much tell you: this is the Epic way. Does that have disadvantages? Does it allow you to standardise?
Ian Abbs: It has some advantages and some disadvantages. There are certain aspects, particularly of the go-live, that Epic colleagues will have learned through experience over hundreds of go-lives internationally - they know that doing it a certain way is likely to be successful, and doing it other ways is likely to be unsuccessful. So one can take some learning from that. One of the issues, and it comes back to the point you raised, is that there's an advantage to something developed within the Epic world, tested within the Epic world, and then executed in the broader Epic user world - but in that type of iterative process, there is a time sequence. I think it's an interesting question within healthcare organisations, around the types of time sequence one can best use. So in a world where - and I think the big changer is artificial intelligence, particularly generative AI and agentic AI -
Tom: And given - how long's ChatGPT been around?
Ian Abbs: I mean, I think like four years, 2022, late '22.
Tom: So, you know, three and a half years, something like that, maybe a bit less.
Ian Abbs: And I think it's had an impact on all of our lives, we've seen a big change. Agentic AI - and here I don't think this is getting into the world of science fantasy - one of the healthcare sustainability questions is: if you look at outcomes, using mortality as a crude marker of health outcomes globally, and you look at a graph of outcomes versus financial resource inputs, you go up the curve a bit as you start to increase resources, but actually, in many high-GDP countries now, however far you push the financial envelope out, those outcomes don't increase much.
Tom: Right, yeah, there's a plateau.
Ian Abbs: And actually, they're starting to flatten, and in fact decline. What's very interesting is that if you look at some countries - and this isn't just true of the US, which has a particularly bipolar distribution in the population - you see a flattening, and in fact a decline, in outcomes. So for lots of different reasons, pushing more money into the healthcare system may not produce linearly increasing outcomes. And the same is true if you look at another factor in the supply-side equation, which is people. We've tended to respond to demand-side increases, since the end of World War II, by pushing more money and more people in as demand for healthcare has gone up.
Tom: As with our sort of early 2000s conversation.
Ian Abbs: Yeah. So we've pushed money and we've pushed people into the supply side. But if you look at the people factor, which I think is in some ways more important than the money factor, we're simply running out of people. In this country, we employ 1.5, 1.75 million people in health and social care. In the US, healthcare and social care overtook retail and manufacturing in about 2018, 2019, and it's predicted to require, by 2031, about another 8% of the total American workforce. Now, that is not possible. So if we come back to the population health management example I gave earlier - if you took, let's say, a million people, an average-sized town or small city, and took 10%, that's 100,000 people. To manage 100,000 people out of hospital, in a very complex environment with iterative risk analytics predicting - as it will do, as it gets better - certain types of events, you really need a way of interacting with those people on a pretty regular basis, let's say every day. The people can't do that. The compute power we might get from the investments we've made in IT systems could probably give us information about who is at risk. But to help those people at risk, we don't have enough people in the system to manage them. That's something we'll be able to do with agentic AI, I think. I am quite optimistic about the power of AI agents to help us augment the human workforce, and to do some of the things we simply can't do at scale, for the cost it would entail, even if we had enough people - which I don't think we do. Now, whether those developments can easily be built by the sorts of companies providing these very stable, large, safe IT systems, or whether they'll come from the more agile organisations, including potentially Dr Doctor -
Tom: Yeah. Well, we'd like to think we're one of those.
Ian Abbs: Yeah, we try.
Ian Abbs: And it's an interesting conversation, Tom - I think about whether DrDoctor is effectively a supplier of useful software to manage hospital processes, or is in fact a data and analytics company. And actually, what do you know now from the data you have on the millions of episodes that Doctor Doctor has managed in the English NHS, and what predictive capability does that give you?
