A new £21 million investment in AI aims to modernise NHS diagnostics and improve wait times. But does the funding overlook other low-hanging fruit? Owen Hughes reports

The recent £21 million commitment to rolling out AI tools across the NHS shows that the UK government is increasingly willing to entrust new technology in its efforts to tackle the NHS’s big challenges.

The AI Diagnostic Fund aims to spur the deployment of machine learning (ML) tools that help clinicians diagnose and treat patients with serious health conditions, including AI imaging technologies for analysing X-Rays and a commitment to roll out “AI decision support tools” across all NHS stroke networks by the end of the year.

Dr Katharine Halliday, president of the Royal College of Radiologists (RCR), points to a number of areas within NHS diagnostics services that stand to gain from a digital update.

“Bookings in the radiology department, for instance, are made over the phone, and that’s a monumental waste of loads of people’s time,” says Halliday.

“There’s huge potential for us to automate that and use AI, particularly in things like screening programmes. Again, for breast screening, a lot of that system is very old and clunky, and if we could use AI to speed that up, that would be a win.”

Low-hanging fruit

Arguably, at its core the £21 million AI investment is an effort to tackle a challenge that touches all parts of the NHS – waiting times.

After all, any technology that helps clinicians diagnose and treat patients more quickly also helps hospitals clear appointment backlogs.

If that’s the case, could it also be argued that the funding overlooks more fundamental NHS processes in need of digitisation?

Ben Court, head of analytics software at Civica, doesn’t dispute the motive or the reasoning behind the AI Diagnostic Fund, but agrees that “there’s other low-hanging fruit” when it comes to addressing existing pressures.

“I think if you engage with the users and look at what their biggest pain points are, there are things like paper-based administration, directing patients, standard admin processes or planning processes…There are probably dozens of administrative processes supporting those challenges,” Court says.

Civica has built a variety of universal, “reusable” AI tools that can be deployed across different public sector markets, ranging from “the very simple to the very complex,” according to Court.

Its optical character recognition (OCR) technology, for instance, can be used to scan various clinical documents and store them as digital files, and will be included in Civica’s cloud-based Cito electronic health record (EHR) software later in 2023.

Such tools might be a far cry from the more fanciful predictions and presumptions about the role of AI in healthcare, but the reality is these are exactly the sort of applications that NHS staff are calling out for.

“We’ve definitely seen better uptake of solutions that remove the bits of people’s job that they don’t want to do, or they don’t feel is valuable or is useful,” Court says.

Doing the grunt work

Indeed, while Halliday can envision a future where artificial intelligence is capable of interpreting an image autonomously, she adds that “we’re a long way from that, and I don’t think that’s going to happen very soon.”

Says Halliday: “The majority of the AI that we’re looking at is something [that can be used] in conjunction with either a radiologist or another health professional. AI can help us do some of the grunt work, if you like.”

AI/ML diagnostic tools can also give staff the confidence to determine whether a patient in their care requires urgent treatment by a specialist, says Halliday.

For stroke patients, and in particular, those who require a mechanical thrombectomy, time is especially crucial. “If you can do that [procedure] within a couple of hours, people just return to normal,” Halliday says.

“The AI gives the stroke doctors who aren’t experts in imaging the confidence to say, ‘I’m going to refer this patient, and I’m going to get those wheels in motion.’”

Jay Verma, GP partner and president-elect of the GP and Primary Care Section at the Royal Society of Medicine, very much views AI as a tool for augmenting clinicians’ roles, rather than replacing them.

“I don’t want the AI to make clever diagnoses,” says Verma.

Verma, who is also CEO of informatics company Data Care Solutions, believes the role of AI should be to help clinicians “separate the wheat from the chaff” and focus on the key parts of their job, particularly amid ongoing shortages of staff.

“Can we use it to help with administrative workload, which will free up clinical time, which makes us more efficient, and hopefully, more effective?” he says.

“If you think about all the different processes – appointment bookings, seeing patients, filing their lab results, looking at the clinical documents, reviewing investigation results, doing prescriptions…those are the genuine things that will cover probably about 80% of my workflow.”

The rise of the (chat)bots

Chatbots are also becoming increasingly valuable in healthcare, predominantly for directing patients to appropriate services and helping them with common queries.

Court notes that natural language processing (NLP) models have become a lot easier to build in recent years due to their explosion in popularity across nearly every industry.

Civica and Belfast-based digital studio Big Motive spun up a chatbot platform for the Northern Ireland Department of Health in 2020 to give citizens access to important information about Covid-19.

The company is now working on more advanced NLP using both foundation (i.e., pre-trained) and custom-built models to teach its chatbot platform “the language of the NHS” and help support the codification of clinical documents.

Further down the line, these chatbots might be capable of generating “cheat sheets” for clinicians that provide a concise summary of a patient’s medical history, helping healthcare providers keep track of what occurred in previous appointments making timeline analysis possible, says Court.

“There’s quite a lot you can do once you’ve given that information some more context.”

Knowing where the issues lie

NHS trusts that plan to bid for the AI Diagnostic Fund have to prove that the tool they want to use represents value for money – not just in theory, but in practice.

This means knowing exactly which bottlenecks need to be addressed, as well as having a clear understanding of both the end-goal and the way in which NHS staff will interact with – and apply insights from – these tools.

“AI and ML are like any other software and technology: it’s only as good as the engagement you have with the users,” says Court.

“If you aren’t clear on what you’re trying to achieve, and what the user is going to do with that insight once they get it, you may just be throwing good technology at the wrong problem.”

A key concern for Halliday is the lack of established processes for evaluating the impact of AI tools on existing systems and users.

This has prompted the RCR to begin compiling a registry of AI imaging tools used in NHS trusts that decision-makers can refer to when researching which technologies are available to them, and whether they can be deployed successfully.

“We don’t yet have the kind of mechanism and infrastructure in the NHS to assess these products, and give the trust the confidence that it’s OK to introduce them,” explains Halliday.

“We need to do long-term audits to really understand how the patient’s outcomes are affected…There needs to be a whole infrastructure, really, around ongoing QA (quality assurance) of these products.”

Verma agrees, arguing that a more exhaustive overhaul of NHS IT will be required particularly if the NHS hopes to leverage AI/ML for population health initiatives.

“It’s a very archaic system and [IT systems] don’t talk to each other. I feel like it’s been designed for technical people and not for the patient, which is my biggest frustration,” he says.

“System redesign has to be putting the patient upfront. I’m not just talking about technology – clinical services have to be for population health, thinking about the needs of the locality and how we manage that as best we can.”

A sticking point for Verma is the fact that the AI Diagnostic Fund overlooks primary care completely.

While he acknowledges the substantial caseload in secondary care, he argues that strengthening GP services has a knock-on effect throughout the care system.

“Even that one patient we don’t refer onwards and manage in primary care is one patient less on a potential waiting list,” says Verma.

“I think this is a great opportunity. I think we’re in a wonderful time where change can happen and impactful change can happen. But it requires us to stop, think of all the processes that occur, and redesign it around the patient.”