The NHS: Is AI really the answer?
Two doctor-led companies believe it is
Two radically different views of the role AI can play in the NHS this week - one from a former Prime Minister, the other from a group of cancer doctors. Sir Tony Blair told the BBC Today Programme that if the NHS and other public services did not embrace AI then we faced slipping behind other nations, with the health service facing intolerable cost pressures. But a paper by cancer doctors and researchers in Lancet Oncology warned that sorting out the UK’s poor performance in cancer treatment involved getting back to basics rather than relying on “magic bullets” from AI and other flashy tech solutions.
Reading these two views my mind went back to two of the many interviews I’ve conducted with people trying to implement AI in the NHS. Both Ben Maruthappu and Dom Pimenta are doctors turned entrepreneurs and neither is starry eyed about the role technology can play in healthcare. Instead, each believes they have created products that will boost NHS productivity rather than provide miracle cures.
Dr Ben Maruthappu, Cera
It was in 2016 that Dr Ben Maruthappu started Cera with a mission “to take as much health and social care from hospitals and care homes as possible into people's own homes.” A career as an A&E doctor followed by a spell advising the boss of NHS England took a different turn when he came face to face with the system's failings:
“I had to organise care for my mum after she fractured her back and needed support at home at the beginning of 2016 and it took weeks for that care to be organised.” Even when carers were in place he found it hard to keep track of them.. He realised that he knew the names of his Uber drivers, and could track the Deliveroo courier bringing a meal, but didn’t even know the names of his mum’s carers.
“That led me to think that technology could improve the way care is delivered in the home,” he explains.”It could help us prevent people ending up in hospital. It could help us get people out of hospital more quickly. It could give family members of older people more peace of mind.”
Just eight years later Cera is the UK’s fastest growing healthcare company, employing 10,000 people, and making two million visits to homes each month, after signing deals with hundreds of local councils and health trusts. Every visit sees information logged on an app and since 2021 the company has been using its huge and growing data pool to launch AI products.
First came a product designed to warn which patients were likely to be hospitalised in the next week: “Our app and our technology can predict over 80% of hospitalizations one week before they happen with statistical significance….That gives us a week to do something about it and so then we trigger an intervention..”
Later, came an AI system that could predict falls before they happened - Dr Maruthappu explains that an intervention could then range from a change in the patient’s medication to a visit to a falls clinic. “When we rolled this product out, within its first two weeks falls went down by over a quarter.” He has a lot more plans to transform his company’s processes using artificial intelligence and worries that the health service which has been very slow to digitise its operations could be about to repeat the same mistake:
“Unless we want the NHS to be 20 years behind everyone else again, we need to start thinking about AI.”
Dr Dom Pimenta - Tortus
I have had lots of fascinating conversations about the role AI can play in healthcare but until I met Dr Dom Pimenta I had not had a demonstration of the technology in action. In one of my earliest chats, Tom Whicher of DrDoctor mentioned a company called Tortus which was doing amazing work using generative AI to make sense of a doctor’s appointment with a patient.
When I eventually realised how Tortus was spelt and tracked down its founder, Dr Pimenta he told me that the idea for their first product came from a chat with his co-founder and Chief Technology Officer Chris Tan who said: “Don't doctors do a lot of paperwork, and wouldn't AI be quite good at that?”
He also remembered a period in his career as an NHS cardiologist when he was for some reason locked out of the hospital computer system for three months: “It was great,” he says, to my surprise:”I took a junior doctor with me, and they would do all my notes, my letters, my orders, my texts, but I would see 35 patients a day. I was the most productive I've ever been as a hospital doctor.”
What he and Chris Tan set about building was a product that reproduced what that junior doctor did - an AI assistant to capture every interaction with a patient. 18 months after that first conversation a 20 strong team made up of doctors, machine learning experts and commercial staff has come up with a system they call Osler.
It is currently being used in 25 GP practices and Tortus is talking to a number of hospitals as the product goes through various regulatory processes. On a video call Dr Pimenta offers to give me a demo. I play a patient worried about chest pains and we have quite a detailed conversation where I combine imagined symptoms with my real medical record. We discuss my Parkinson’s and the medication I take for it, I bring up the ocular melanoma I have been treated for since 2005, and the doctor ends by arranging for me to have an MRI scan.
Throughout the conversation Dr Pimenta is looking at me, not a computer or smartphone screen, and he isn’t having to pause to take notes. At the end, Osler produces a transcription of our conversation - it is a pretty good one but there are plenty of tools already offering this service. What is really impressive is the way it compiles notes for the doctor of the key elements of the consultation and then composes one of those famous letters to a GP which always seem to begin “I saw this charming gentleman….”
It produces a letter that is both comprehensive and well written - not universal characteristics of letters composed by human doctors - and comes with notes translating some of the medical jargon for the patient. Tortus relies on a Large Language Model to power Osler although not one that it has trained itself. “That's not our business,” says Dom Pimenta. “Our business is clinical evaluation of the models that are out there.” He says that by using and fine-tuning the very latest LLMs, Tortus can stay at the cutting edge of this fast moving technology.
The appeal for customers is that a fifteen minute GP or hospital appointment becomes more productive with the doctor spending much more time engaging with the patient. Then the tedious admin afterwards - dictating letters, ordering tests - is accelerated. While the hope is that Tortus may prove more accurate than an overworked doctor in recording what happens in an appointment, Dr Pimenta is keen to stress that it is the doctor, not the AI, who is in charge: “In terms of liability, it's always going to be the clinician's responsibility, and I can't see that changing realistically for a very long time.”
So, two examples where it seems AI can help solve some of the productivity problems facing the NHS. But I had spoken to my two interviewees long before that sceptical paper from cancer experts convinced that the technology is overhyped. I went back to each of them to get their reaction.
Dr Ben Maruthappu said,”AI on its own is not a cure-all: its potential can only be realised when coupled with an effective workforce and a functioning system - whether that be for cancer patients, elderly patients, visitors to A&E or anyone else in need of healthcare.
But the safe introduction of responsible, proven AI across health and social care can only help us deliver better care, to more people.”
Dom Pimenta wanted to stress that the impact of Large Language Models capable of clinical understanding promised to augment rather than replace skilled medical professionals: “Far from a ‘magic bullet,’ AI in this use case represents the only meaningful and immediate change we can realistically make to the system today—doing more with the staff we already have while not burning them out further.”
Given the level of hype around AI it is easy to understand why some in the medical profession are dismissive of it. Some projects will turn out to be a waste of money and it will take time and patience for others to pay off. But, just as with the arrival of computers and smartphones in every area of our lives, this does not feel like a revolution that is going away, so the NHS had better get on board sooner rather than later.
Two things which look like obstacles to acceptance.
1) don't call it AI. What people understand by "tracking my uber driver" is that they know where their uber driver is. What people understand by AI is that it makes stuff up, draws 6 fingers, gets stuff wrong and it's a black box so they will never know what it got wrong or why.
2) I feel that there's a huge resistance to getting NHS data into the hands of "private companies". If the one set up by doctors or spun off from a university department is OK (but how would people know that it was), the assumption would be that it would get bought out by a predatory US company which would exploit the data for profit against people's best interest. History is not on the side of public trust here.
We got back at 7:00 last night after going to the local hospital for a 3:30 appointment, a consultation that took 10 mins and an hour trying to get a prescription from doctor to pharmacist- eventually delivered on paper by my wife, after the electronic one was abandoned.
Everyone involved was lovely and helpful - but the admin! As an ex-computer programmer, I am appalled at how poor it is.
The NHS is not alone in this, but it means that recent small crisis in my wife’s health involves us doing our own record-keeping, cultivating contacts inside the system, policing the drugs and hunting down lab reports and treatment records.
A decent IT system must be a priority.