Digital healthcare transformation: the role of AI in self-assessment – Don’t get left behind - Q&A Transcript

Digital healthcare transformation: the role of AI in self-assessment – Don’t get left behind

Webinar Blog and Q&A summary – 18 Nov 12.30 – 13.45

Connect Health “Change” brings you the fifth in a series of webinars to make and embed transformation in healthcare. Aimed at system leaders and clinicians across the NHS, the webinars provide practical solutions to the challenging issues we are all grappling with.

Background to AI self assessment

Digital enabled patient care is outlined as a key area for focus for the next 10 years, as outlined in the NHS Long Term plan. As a result, we can expect the existing model of care to look markedly different. The NHS will offer a ‘digital first’ option for most, allowing for longer and richer face-to-face consultations with clinicians where patients want or need it.

Primary care and outpatient services will have changed to a model of tiered escalation depending on need. Senior clinicians will be supported by digital tools, freeing trainees’ time to learn. People will be helped to stay well, to recognise important symptoms early, and to manage their own health, guided by digital tools.

This session explored why AI triage is useful for commissioners and explained where it sits in the pathway. We heard the lessons that have been learnt from Wales and NHS Trusts , and demonstrated how AI tools such as PhysioNow® ,powered by Phio, adds value, and is something that is tangible for the patient.

Watch the Webinar in Full

Q&A

Abi Phillips, Head of Innovation – Economy Skills & Natural Resources Group, Welsh Government – explained the rationale behind the Welsh Government’s Digital Solutions Fund – innovative ways to use digital technology in response to coronavirus, the lessons learnt, feedback from Health Boards and the next steps.

View a video of Abi’s talk

Question 1: What does the future hold?

Answer: So I think we’re going to start seeing a lot more digital health delivery, accepting that obviously not everybody can access digital health delivery. So we’ve invested a lot in Wales in upskilling people. We’ve put a lot of devices out, so our Digital Communities Wales program has been really active in getting devices out into the hands of people who haven’t had them previously. And that’s included people like school children, older people, care homes. So I think we will continue to see that growth, but mindful of the fact that human interaction is still very important, and making sure that some of our more digitally disadvantaged citizens have got access. But I think there’s a long way to go yet.

Question 2: What is the kind of cost to trail software for AI triage? Have you done your cost benefit analysis? Will it have a cost benefit for you within the services? And how will that look?

Answer: I don’t think it’s going to be a cost benefit analysis. The pressure that’s on health care services means that this might better manage demand, as opposed to do away with clinicians which we’re never going to do. So there could be a benefit to time spent by clinicians, time spent by patients waiting, and obviously they’re really important things.

We’re also keen to have positive experience measures and outcome measures as well. I think given that with the COVID-19 pandemic, for patients it could potentially be no physio or access in this way. And then I think we’ve probably more than had our money’s worth. But it’s a very difficult thing to do, particularly when you start talking about cash releasing benefits, because they’re just very difficult to achieve in these types of projects.

Question 3: AI is no doubt an important development, but how can we make sure health literacy and digital poverty issues (with the divide increasing as a result of COVID) are addressed effectively?

Answer: We are doing a lot of work on digital literacy and poverty issues in Wales through our Digital Communities Wales Programme. This includes training and provision of devices. The programme works alongside the Health Boards in delivering this.

Question 4: What is the best thing about digital transformation?

Answer: From a clinician perspective, it offers a much better work-life balance.

Zoe Brewster – Assistant Head of Physiotherapy – Cwm Taf Morgannwg discussed the practicalities of how the Digital Solutions Fund has enabled AI to offer fast-tracked access to healthcare and how it has been embedded within Health Boards.

View a video of Zoe’s talk

Question 1: What is next? And what’s the vision for digital health transformation in Wales?

Answer: So this is top of the radar for lots of people and I’m sure lots of MSK clinicians everywhere. We know that the demand for our services is just going to grow. The aging population and all of those other things that we’re all very familiar with.

We’re really keen to continue to work with Connect Health with PhysioNow, moving forward, to create a self-management element to this system, where we know things like ‘I’ve twisted my ankle’ and it can reply, ‘Well, actually, you’ve answered these questions so we’re going to give you these exercises to get you started.’ You may not need to come into a department, you may not need to come to a face-to-face appointment. And I think the future’s bright in terms of the digital movement.

Question 2: How do you manage the governance around your students accessing the systems?

Answer: We have a long-standing relationship with our Cardiff University students. We take a large volume of students every single year, so having students is pretty embedded in our services. And we’ve got really robust clinical governance structures in place, because we have mentoring and training for those clinicians.

The PhysioNow triage system that is currently set up through the pilot, essentially churns out a document that then just comes into our referral system. So it gets managed in the same way that our other referrals are managed. So in the same way that we have clinical governance structures in place for our students managing new patient referrals. This is no different, they would look at the document beforehand, which is often much better than what they would have had before.

Question 3: AI is no doubt an important development but how can we make sure Health literacy and digital poverty issues (with the divide increasing as a result of COVID) are addressed effectively?

Answer: I don’t think it will be any surprise to anybody that’s joined today that that’s probably the first question that came to us, and certainly some of our GP colleagues were very anxious about who we were excluding.

And I will be honest, we learnt from our experiences and the nature of the pilot, aside from the actual system, was very time limited, and we had to get a certain number of patients through that system. So initially we looked at this being the only access route. So in the same way a lot of other services during the initial phase of COVID shut up shop, we kept urgent only appointments available, but our routine care was very much ground to a halt.

So we looked at this being our initial way to reopen our services. And I have to say we learnt quite quickly that having PhysioNow as the only option was not suitable, in terms of ensuring that we were being accessible and equitable to all of our populations. And we do cover a number of areas with some significant deprivation. Having said that, Gary, one of your colleagues from Connect Health, has been massively influential in gathering the data. I’ve obviously got another role which means that doing all of this is hugely time consuming. But collecting really detailed data about who is referring, when are they referring, what is their age, what is their background.

And we’ve done a piece of work almost overlaying what our normal referral patterns look like, and what our referral patterns through PhysioNow look like. And the geographical basis and the age distribution looks, on the surface, to be very similar. So our old ways of paper referrals or GP referrals, doesn’t look that dissimilar to what we’ve seen in a very short period with PhysioNow.

We would obviously need to have a much longer episode of pilot to be able to really robustly tell you that that is 100% the case. I think our lesson learnt is, this can’t be the only access route and we shouldn’t disadvantage somebody because they don’t have digital access, they wouldn’t go further down the waiting list or anything like that.

Question 4: How much development is required to make this clinically safe? What happens if a patient does trigger red flags? How do we make sure they go to ED promptly? How do we make sure we don’t overburden emergency services?

Answer: I think probably the best way to explain is to tell you a story.

The very first week before we went to go live, we had a test link that we shared with our clinicians for them to see what a patient was going to see. And one of them thought they would test it out. So they answered the questions in the way you never want a patient to answer these questions and within about 45 seconds I had a phone call from somebody to say ‘is this a real patient or is this somebody testing the link?’

So it is being constantly monitored, anything that flags up is dealt with appropriately. The only thing I would say is, that in the same way as a physio in a department or a GP in a practice, we don’t have control over whether that person decides to turn up to A&E or not. If we direct them to 111, ask and tell them we’re concerned about their condition, for example, if that’s what gets generated, that’s the individual’s responsibility. Whether that’s for us, whether that’s different to how your services currently work, that’s probably individual service level implementation that needs to be explored.

But to me, we would advise people to do that now and whether they end up at the end of the corridor, in their car on the way home we don’t know.

Question 5: Does the PhysioNow referral triage to other pathways, or just Physio?

Answer: There are 4 ‘arms’ – 111, routine physio, urgent physio and the self-help arm, which they are currently building. If it isn’t appropriate for physio it would direct to 111 but the numbers are small. I should also say that some patients stop half way through the ‘PhysioNow’ chat if they realise they’re not in the right place – i.e. it’s not physio they need.

I understand that Connect Health can also triage to their CMATS/MCAS/iCATS service in some places, but that wasn’t part of our pilot in Wales due to time constraints.

Question 6: How will this work alongside First Contact Practitioners in the PCNs and how will Connect Health mitigate the impact of digital inequality?

Answer: PhysioNow is used as an ‘option’, we can still take referrals via telephone, and we are currently embedding FCPs in Wales.

Question 7: Is your plan to move away from seeing patients face to face?

Answer: Definitely not! Our clinicians have trained as physiotherapists because they want to work with people, not at the end of the phone!

Question 8: Are initial appointments following triage referred to all face to face or remote?

Answer: The next step is based on clinical need. We have had plenty of discussions, along with the impact of COVID, over recent months about what is best, and it is an ever developing discussion.

Question 9: The UK is a multicultural and diverse country with many languages other than English as a first language. How will AI be able to support those who do not speak English or have poor quality of understanding of the English language?

Answer: Part of the life sciences/digital innovation fund was a requirement to test the Welsh language element. I anticipate the same would be possible for other languages.

Question 10: How has the use of AI triage impacted on the amount of triage required of clinicians in both MSK physio and MCATS services where piloted?

Answer: Since this was just a pilot run for PhysioNow, we haven’t reduced triage for CMATS or physio at present.

Question 11: What is the best thing about digital transformation?

Answer: Putting the power back into the hands of the patients, they can choose how and when they can access healthcare.

Peter Grinbergs – Co-Founder and Chief Medical Officer, EQL, discussed the appropriate use of AI in MSK medicine and the potential to revolutionise healthcare practices, customer experience and health outcomes.

View a video of Peter’s talk

Question 1: How do you think AI will complement clinician behaviour?

Answer: The power behind artificial intelligence really lies in its ability to collect, analyse large data sets and then present back for interpretation. And I think it’s about presenting back that’s the key part, because clinicians are the experts, and effectively what we’re able to do is aggregate data in a way which is potentially meaningful, give insight.

But ultimately the whole point is that it’s passed back to clinicians, to effectively evaluate, sense check, see whether or not it’s something that is potentially useful and evaluate, and then incorporate that potentially as part of their practice. So I see it as another facet, another kind of tool that potentially sits with that clinician and effectively is there to be able to provide meaningful data and allow them to clinically reason in a way which hopefully has been enhanced through the insight that’s been collected.

Question 2: What is the long-term impact of technology like this on musculoskeletal clinical careers?

Answer: The healthcare system at the moment deals with people that effectively understand that they’ve got an injury, they can do something about it, and so they’re seeking healthcare in order to overcome the challenge. But I think there’s a huge amount of people, potentially, that don’t know about the fact that they’ve got things that can be resolved, but importantly can be self-managed. And as we know, MSK pain is a gateway for co-morbidity generally.

And so I wonder what the knock-on effect is for those people that don’t proactively engage and seek healthcare when it’s completely necessary, or when it’s absolutely necessary? So I think, potentially, that’s where there can be some real impact. It’s just increasing awareness, increasing engagement early on, and in doing so then create a kind of culture where people are much more open to understand the need to engage in healthcare, and are willing to do so themselves. Personally that’s where I think there can be some real value added into the system.

Question 3: How much development is required to make this clinically safe? How do we make sure they go to ED promptly? How do we make sure we don’t overburden emergency services?

Answer: Effectively, making sure that we get proper clinical review on the outcomes as they’re coming out of the back of the solution, and having robust systems in place to also sense check onward referrals where they occur.

But I think you’ve had some interesting data. I know one of Connect Health case studies was where the data is showing that what we’ve been able to do in certain areas is actually significantly increase the rate of referrals into the emergency department where it’s been 100% needed.

At high level, it’s really just about making sure that initially when we design the system, that we use expert opinion. That we sense check what the local frameworks are. But before anything goes live it’s subjected to significant rigorous testing, both with real life examples and synthetic examples, i.e running through numbers of scenarios to make sure that the outcomes are where they need to be. But then when it’s also being deployed, it’s actually the way that it’s intended to be used, is that it is checked, the outcomes are checked and validated, and as we know there’s been some good data around what the acceptance rate is there. And there’s an ongoing process of again checking that, not getting to a position where we become reliant that it’s working, but continually sense checking and continually updating.

Question 4: Is there a level of clinical knowledge and experience that enables effective integration of AI – is it different between novice and expert? Does it support the development from novice to expert?

Answer: I think it is a very broad area, there’s lots of different ways to apply AI, there’s a lot of confusion around AI and I think there’s a couple of things that need to be unpicked there.

First of all, we owe it as the innovators, or the people working in the space, to be able to provide a clear explanation as to what the system does, what it absolutely doesn’t do, and what the parameters of use are. And I think that’s more of a broad thing. But obviously when we’re developing and deploying a solution, we need to make sure that we fall within the regulatory framework that we must abide to as part of that solution. And things like post-market surveillance are really key. And I think interestingly, within that effectively is how do we check that it’s continually safe when it’s being deployed?

It’s not an excuse to say ‘well your AI can be biased.’ But what you do have to do is come up with strategies that can effectively mitigate against that risk, so it’s how are you getting additional information back. For example, you send someone to A&E, it turns out they had a diagnostic. What does the diagnostic say? Or what was the follow-on opinion?

And I think that’s really where the power is, particularly the power of some of the technologies that we’re looking at, it is in that ability to look at that data, particularly to be able to look at that closed loop data which is where you get full end to end transparency over a patient journey, and then start to kind of distil some of the insights out and see where some of the inefficiencies are. And I think by creating systems that effectively have these sensible sense checks built into them as part of their design, I think that’s where you start to get confidence in its ability to be able to be deployed. And to providers, they get comfortable that the decisions that it’s making are correct and that the safety mechanisms are in place.

Question 5: Do you use behavioural insights/science when implementing and evaluating AI models/products?

Answer: I can only talk about this from my perspective and what we’re building.

We absolutely include as much information as we can include, so it’s not just biological information, but there’s the whole psychological component to what we do. And MSK is a complicated area, I mean, it’s difficult to define at the best of times what exactly the area of MSK is because, it does cover so many different broad areas of healthcare, and it has such a vast impact on so many different areas. But I think what we have to do is make sure that we use good research, that we’ve sense checked and validated, that meets the criteria that we need to use.

Question 6: What is the best thing about digital transformation?

Answer: Better clinical engagement.

Dominic Cushnan – Head of AI Imaging, NHSX – talked about the importance of getting AI right and the NHS AI Lab.

View a video of Dominic’s talk

Question 1: How do we get involved with this? What support do you offer?

Answer: One of the things that we want to really strive to do is be collegiate. This is not about us sat in a black hole in the centre trying to figure things out on our own.

So we have set up through the NHS Futures website a mechanism for you to come collaborate with us. Obviously if you’re looking for funding, you can go through the NHS AI award, where we can work with you to get your technologies potentially funded. But I think everybody brings a various mix of skills and capabilities to the conversation, and we would like to have an ongoing conversation with you as we build our programs, as we think about safety, as we think about validation etc. So through the NHSX website on the AI Lab, there’s a link directly to the community that we’re building on it, to make sure that we get your voice heard through all our programs.

Question 2: AI is no doubt an important development but how can we make sure Health literacy and digital poverty issues (with the divide increasing as a result of COVID) are addressed effectively?

Answer: We’re doing a number of different activities to make sure that we do address any potential inequalities, and my colleague who heads up Ethics and Equality is doing quite a lot of work with us, and we’re looking at hopefully the next couple of months to make some announcements about some programs.

The other bit I think you’re alluding to, is what are the digital capabilities? So we’re working with Health Education England through our activity to make sure that we feed directly into that. And there has been ongoing conversation about literacy for quite a long time, even across the analyst community, about the literacy around being able to develop algorithms, or being able to understand statistical analysis inside our organisations. So I think it’s a massive multifaceted problem and Health Education England as the body for that have been working closely with us.

Question 3: Is there any document that discusses the data warehouse for it’s readiness for implementing AI in the NHS?

Answer: There are a number of activities across NHSD, NHSE & NHSX around data quality, and there are a number of data sets that are already being developed.

If the questions are on the standard itself, yes there are standards sat within NHS Digital about the types of things you should be including in your data sets. When it comes to implementing ML and AI, how do you identify whether the quality of data is representative to be able to both train and independently validate, whether an algorithm is safe for deployment, that’s something that we’re finding out at the moment.

But in terms of readiness, it really depends on where you sit in terms of whether, first off, is it a medical device? What is it you’re trying to do? And the kind of steps that are needed to regulate and approve your technology.

Answer: So I think this is where I struggle generally when we talk about AI. Because AI is a broad science. It covers quite a lot, it’s a bit like saying ‘do you understand the science?’

I think where we need to get really practical, is in the real world use cases. So would a radiologist be comfortable with a decision, with an augmentation to say that there is pneumonia on a chest in a particular area to support them in their diagnostics? So are clinicians ready to be able to use these technologies above and beyond the existing tech that they have?

I think it varies, I think it’s part of the wider digital capabilities that we have in the healthcare system, about adoption of AI technology, or adoption of technology generically. There is an appetite for some people, who are a little bit more ready to look at AI. And you see this in the working groups of the people that are having these conversations. They can see the opportunity and the helpfulness of some of these technologies in their workplace.

But, as others have mentioned, there are some that are not being engaged, and how do we keep them motivated that this is a potential solution to help them?

Question 5: Do you use behavioural insights/science when implementing and evaluating AI models/products?

Answer: Just for reassurance for people I think who are probably new to some of this field, we have a very robust regulatory system in the UK. We have bodies that are exploring this work with us, about how do we embed further safety and mechanisms to do that. NHS Digital has published guidance for those who are new to some of the stuff around clinical risk management, and you need to do this for any type of digital health technologies. We’re embedding this on top of the existing good processes of making sure that technology is robust. It’s not to say there isn’t more work we can be doing, because there is and I just want to kind of reassure people that there are processes in place.

Question 6: What is the best thing about digital transformation?

Answer: I’d actually like to take this time to recommend a book by Eric Topol. It’s a book on deep medicine and how AI can make healthcare human again. It’s about removing the burden, and how can we help them to do the best they can to support the patient with limited resources.

Graeme Fletcher – Chief Information Officer, Connect Health

Question 1: How do we integrate AI into our clinical systems?

Answer: I think it’s an interesting question, and one of the key challenges we have in a lot of the healthcare world, and particularly our estate, is how we get systems to talk to some of the big clinical apps like SystmOne? There’s no natively available integration and APis and readily available platforms that we would normally use.

So, in this instance, what we decided is we already have a fairly sophisticated robotic process automation platforms sitting in the cloud. And we’d already built that, and implemented it to do patient registration, because it allows us to do this 24/7, it frees our people up to do more value-add work. So what we did is use that technology to allow EQL to talk to our SystmOne clinical application. It allows us to do it 24/7 should we need to do that. It allows us to scale so that we can handle as many transactions as we like.

Within that, it has its own AI implementation inside that platform – it intelligently understands where the load is. And we can swarm more virtual workers at any time of the day or night over where the loads are, because obviously it’s not the only thing we use.

We need humans in the process where it comes to some of the more clinical decisions. But when it comes to the administration we just didn’t. So we built that in and it allowed us to move that forward, and it just gives us lots of different ways we can play with that going forward.

Question 2: What is the best thing about digital transformation?

Answer: For me, it’s allowing our humans to move into a more valued workplace, take away some of the manual labour and let them focus on the more important stuff – the patient.

Question 3: What is the difference between PhysioNow and Phio?

Answer: Phio is the original platform developed by EQL. PhysioNow, powered by Phio, has an optimised clinical decision tree developed by Connect Health which has different algorithms and bespoke outcomes. PhysioNow is un use by Connect Health and has been used in the Wales digital pilot. Connect Health and EQL have together developed and tested PhysioNow, with senior clinical input to ensure safety, accuracy of decision marking, clinical quality and effectiveness.

Digital healthcare transformation: the role of AI in self-assessment – Don’t get left behind – Q&A Transcript

27 November 2020