Healthy intelligence: How AI is changing the NHS

Some have characterised artificial intelligence (AI) as a pressing danger, raising fears the technology could signal the end of humanity as it becomes more sophisticated, but in fact, the reality is far from that. Forget Hal 9000, Skynet, and Blade Runner replicant Roy Batty, the emerging technology is being touted as a breakthrough early detector for diseases such as cancer and heart failure.

And it is clear that the Scottish Government recognises this following the publication of its strategy on AI in 2021, in which it placed a great emphasis on how the emerging technology could “save lives” in the health service.

The hope is that AI systems will use data to make better informed, more accurate decisions throughout the health service in Scotland – becoming a third eye for clinicians in many cases. The vision for the next ten years is to reduce the number of people coming into hospitals unnecessarily, ensuring accurate and fast treatment, getting patients out of hospital and into an appropriate care environment as quickly as possible, and to intervene earlier in the process of detecting and preventing disease.  

Experts suggest it could solve many of the problems currently faced by the NHS in Scotland, which have become no secret in the last 24 months. The most recent figures from Public Health Scotland (PHS) show that only 68 per cent of patients were seen at A&E within the four-hour standard waiting time. And for the most part, that crucial waiting target has been on the decline since mid-2021. The figures, which date back to 2007, show it had never previously fallen below 85 per cent, and has stayed above 90 per cent for the most part. Cancer waiting times have also increased recently, so much so they are at the worst level since records began. In the last three months of 2022, it was revealed that almost three in ten who were referred with urgent suspicion of cancer waited longer than the 62-day target for their first treatment – with just 71.7 per cent of cancer sufferers treated within the crucial target.  

James Blackwood, an AI expert working with NHS Greater Glasgow and Clyde, says that at the national level, AI will be able to be deployed to speed up and improve decision-making in the NHS, while reducing the workloads of healthcare professionals.  

“This will start with something like diagnostics, where AI is currently most used. And then it will be spread out across other disciplines and areas – operational, not just clinical. So, areas like planned and unplanned care, the primary and secondary care interface, and then the secondary into tertiary and community care interfaces. 

“The vision is doing that as nationally as we can, the preference is not to go health board by health board.” 

To help achieve this, in June 2022, NHS Scotland announced the Accelerated National Innovation Adoption (ANIA) pathway, which was set up to accelerate the process of bringing innovative ideas which potentially have national significance and impact into practice. The pathway involves a value assessment to prioritise innovations that will improve patient outcomes and experience, improve staff experience, and are both financially and environmentally sustainable. 

The health boards, through the three national testbeds – the NHS West of Scotland Innovation Hub, the NHS Health Innovation South-East Scotland, and the NHS Grampian Innovation Hub – provide evidence that an innovation is going to deliver value. The ANIA pathway independently and objectively looks at that evidence, assesses whether there is a value case, and if it is deemed to be viable, it recommends it for national procurement. One AI technology which is currently under consideration is an application that tries to determine signs of lung cancer on chest X-rays.  

Blackwood believes the ultimate role of AI in the health service is to improve workforce resilience and says as the technology starts to reduce the amount of unnecessary work a clinician must perform through the provision of clinical decision support, it will improve better patient outcomes, and reduce the number of unnecessary treatments and care. 

There are no AI systems that have been rolled out nationally that are reducing the workload of healthcare professionals, but Blackwood says we are “at the tipping point” between research and adoption. 

“With the announcement in England of £21m of funding to accelerate the adoption of AI, although that was not available to Scotland, it demonstrates for the first time that we are not just funding research, we are now funding adoption. 

“If we look at how AI is used in Scotland it is only used in clinical practice in a few areas – for example, paediatric bone growth measurement, it is used almost universally across Scotland for that, and it is used in radiotherapy which automatically identifies the dosage of radiation required. Besides that, it has been used in pilots by individual health boards, which are pushed through the regional innovation testbeds.” 

He continued: “Most of the work going on just now is piloting and evaluation. The one area where we have started the national procurement approach is in stroke thrombectomy (a medical procedure used to treat some cases of stroke caused by blood clotting). There is a national procurement running just now to purchase an AI solution to assist with that type of stroke. That process has been running for about a year now.”  

One trial completed by NHS Grampian and the University of Aberdeen has proven to detect abnormalities in breast cancer screenings that would have been missed using current screening procedures. Using an AI tool named ‘Mia’, which was developed by Kheiron Medical Technologies, 220,000 mammograms were analysed from more than 55,000 people to determine how well it could detect breast cancers. 

The analysis found that the AI tool was successful at identifying potentially missed cancers, known as interval cancers which are detected between screening visits.

Initially the team found that the tool was “too sensitive” and would have suggested recalling women when it may not have been necessary. But when it was adjusted for the local conditions of people living in the local environment, the team found that

Mia would have suggested recalling 34.1 per cent of the women who went on to develop cancer in between screenings. Using current screening measures these cancers remained undetected until the women developed symptoms. 

Dr Clarisse de Vries, a radiology imaging researcher at the University of Aberdeen who led the data analysis explained: “Currently, two experts examine each mammogram and decide whether the person should be invited back for additional investigations. If the two experts disagree, a third expert makes the final decision.

“Similar to a human expert, Mia can examine a screening mammogram and give an opinion as to whether that person should be invited back for additional investigations.  

“Mia has previously been developed and tested on some groups, but until now had not been used on data from the NHS in Scotland.  

“Our finding is a massive step forward in using AI technology in diagnostic medicine – we showed that once tuned to the local environment, AI can be of enormous benefit to clinicians and importantly, people who may be at risk of developing cancer.” 

She added: “Our results show that AI, and in this case Mia, offers huge potential for detecting cancers that may otherwise be missed.  

“Fundamentally however, our study shows that AI tools must be tested first and tuned for the local population and conditions, and we have been fortunate to have been able to do just that here in Grampian.   

“Previously, it was unclear whether AI tools developed elsewhere could be used in different settings and screening centres. Now we know there are risks in just taking an AI tool developed elsewhere and implementing it locally. You must first test the tool on the local data to ensure it will work as expected.”

The issue of equity in health is often brought up within the conversation around health in Scotland, and choosing a national approach enables a more equitable health service, argues Blackwood. He thinks of the technology as automation which fits into “that bucket of doing things faster and quicker” without needing human beings “to assemble the data”.  

“In the first instance, what AI can do is interpret the data, which allows you to target care at patients who most need it. One of the things that does, particularly in smaller health boards that are perhaps more resource-constrained, is it allows them to use the resources they do have significantly more effectively.  

“The other thing that it does, is that without people having to come into hospitals, the AI looks at a combination of data, like someone’s clinical history; data the patient inputs into a device; data monitored on their person, and it starts to predict, not just the risk, but also what we should expect the intervention and the care may be. What that means is that it is preventing people from having to go into hospital to get lots of tests and it is intervening as we detect it becomes necessary. 

“It’s not just equity of access, but also equity of delivery.” 

Often, the introduction of AI is characterised as revolutionary. Georgios Leontidis, director of the University of Aberdeen’s Interdisciplinary Centre for Data and Artificial Intelligence, described it as such last year, saying that the emerging technology is “uniquely placed to provide solutions to major challenges” faced in the NHS.  

But revolution is not a word that Blackwood necessarily agrees with when describing the transformation of the health service. He says it is easy to “get excited” about AI, and he points out that the average time it takes to get an AI device onto the market is five years and could take a further five years for it to become mainstream. 

“When you talk about a revolution in healthcare, I think there are ‘mini revolutions’ that can happen within these pathways. Take lung cancer for example, AI that exists right now will look at a chest X-ray and tell you if it looks like lung cancer, it will then look at a CT scan and tell you if there is lung cancer, help quantify and track the cancer. We now digitise tissue slides, and AI will quickly look at a biopsy and quantify and stage that.  

“At that point, if the patient is assessed to need treatment it will calculate where, when, and how the radiation needs to be targeted.”

He describes how this is already a “mini revolution” in the lung cancer pathway, as the current time of referral from a GP to get a CT scan to confirm lung cancer “can be six weeks”, but using AI “that will be two weeks”. This is crucial as in late-stage lung cancer, each week that is “brought forward” is a one per cent increase in the chances of survival past five years.

“Stage four lung cancer has a less than five per cent chance of survival, stage two has a 35 per cent chance of survival, and stage one which has a 99 per cent chance of survival. The sooner we can automatically pick up whenever a test is done that someone may have cancer, we can increase the chances of survival from less than five per cent to greater than 99. That to me feels like a revolution, but it is a revolution in a specific lung cancer pathway.  

“There are hundreds of those pathways, but if we tick them off one at a time then you get all of these mini revolutions. That will take some time, given we have only adopted one piece of AI across the entire lung cancer pathway. For the next five years, I would hope that we can adopt AI into each stage of that pathway. It is not going to completely change the way the NHS works in the next 10 years, but it is for sure going to have a big impact.”