Waiting is usually the worst part of receiving medical care.
“When you’re staring down the barrel of a gun, and you’ve just been told you’ve got a cancer of the tonsils, or something like that, any day that you have to wait to start treatment is devastating,” says Dr Rajesh Jena, an oncologist at Addenbrooke’s Hospital, Cambridge.
For some conditions, waiting can dramatically shift the odds against recovery. With the fastest-growing cancers, each day without treatment reduces the likelihood of controlling the tumour by two per cent, says Jena.
The oncologist is one of a growing number of clinicians using AI to work faster, and better. In clinical oncology, bottlenecks occur when doctors manually mark up CT and MRI scans to show what cancerous tissue needs to be targeted with treatments such as radiotherapy. Jena has long been interested in AI and coding – even studying it at university – but it was a chance encounter that led him to develop an AI to solve this workflow problem. When Jena saw a Microsoft engineer giving a talk in Cambridge about using AI to better recognise the human pose in 3D in order to improve the experience for gamers, he saw the potential the technology had for medicine.
“I said to [the engineer], ‘No, you don’t want to be doing that. You need to use this inside the body to recognise structures within the body,’” he recalls.
That led to more than a decade of collaboration, which has now come to fruition. With funding and support from Microsoft, Jena and his team developed an AI tool that could look at the data from a scan and mark it up around 13 times faster than the human eye. AI could save billions for healthcare systems, while improving quality of care. But in the year that ChatGPT and other AI tools have taken over the world, and as the UK government gears up for its AI safety summit in November, take-up on medical AI remains slow.
Research in the US suggests that only 5 per cent of healthcare organisations there make use of it. Earlier this year, the UK government announced it would be investing £21m in AI for the NHS, adding to the £123m already invested since 2020. These efforts are led by the NHS AI Lab, which has 86 live projects, including initiatives to reduce antibiotic use, scan for skin cancers, and predict the risk of certain eye cancers. In August, nine AI technologies were approved by the National Institute for Clinical Excellence (NICE) for use in radiotherapy. The NHS AI Lab did not respond to requests for data on AI uptake across the health service.
A plethora of healthcare data is available to clinicians, and AI can analyse that at speed.
“My sweet spot is in medicine, where you build systems that can stratify patients at scale,” explains Dr Peter Fish, chief executive of Mendelian, a UK med-tech company. Ultimately, he says, it would provide more personalised care. Mendelian has developed an AI tool to interrogate large volumes of electronic patient records to find people with symptoms that could be indicative of a rare disease. Fish is focused on shortening the “diagnostic odyssey”, as he calls it, that many patients go through when they have a rare condition such as Ehlers-Danlos syndrome – a group of conditions that affect connective tissues – and Behçet’s disease, which causes chronic inflammation of blood vessels. Such conditions affect less than 2,000 people in the UK, and can take up to seven years to diagnose due to their rarity.
Mendelian’s AI has been piloted in the NHS for three years, winning an NHS AI award to trial it at a larger scale. The second phase will look for 40 rare diseases in 750,000 patients over the next year. The focus of the project is on diseases in which early diagnosis and intervention can lead to significantly better outcomes. This approach could also help address problems with access to care. Patients who live near a specialist centre are more likely to get a diagnosis. Fish believes AI can help promote health equality in deprived and remote areas.
The danger of algorithmic bias is well known, and, in healthcare, experts are beginning to discuss this risk. In May the World Health Organisation issued a statement urging caution in the development and use of AI in healthcare due to the potential for unfairness. In 2019, a study in the US found an algorithm developed to predict healthcare needs was biased against African Americans, for instance, because data used to train the AI was inaccurate. As a historically marginalised group, African Americans were under-represented in the data set.
But there is a lack of research outside the US on how common this is, says Niki O’Brien, a visiting researcher in the Faculty of Medicine, Department of Surgery and Cancer at Imperial College London. O’Brien’s research looks at digital technologies and how they can address global health inequalities. Her work highlights the importance of ensuring equity while the use of AI grows.
“The benefits of these technologies, both on systems and for patients, have the potential to create a wider divide between high-income and low-income settings,” she says. For example, solutions developed for use in Western healthcare systems may not work with available healthcare services and treatments, or even be affordable to healthcare providers, in low- and middle-income countries. And the issue of bias in the data used to train AIs can hurt public trust in healthcare services. O’Brien wants to see more funding to develop AI outside of wealthier nations. “Critically, though, with an emphasis on country-led skills and homegrown innovations,” she says.
Jena is working in partnership with another company to develop their AI into a tool that can help surgeons plan operations. The aim is to enable clinicians to use virtual reality environments to collaborate with colleagues on the most effective treatment.
“That would be truly disruptive; that wouldn’t then just be a workflow acceleration, which is what our AI does at the moment – it would actually transform the way that we work together,” he says.
Jena is no “evangelist for AI”, but he tries to encourage people to lose their fear of it. Some of the most useful AI, he says, “is doing really, really simple things”.
This piece first appeared in a Spotlight Healthcare print report on 13 October 2023. Read it here.