AI in the NHS - Hype or Hope?

First in a series of conversations about AI and healthcare

If there is one area of human activity where artificial intelligence promises the most exciting and beneficial changes it is healthcare - or so my email inbox would have me believe. Even when I left my job as BBC Technology Correspondent two and a half years ago, press releases about robot surgeons or the automation of diagnostics via machine learning algorithms had been coming thick and fast for some years.

But the arrival of ChatGPT signalling the Generative AI revolution brought the hype to a whole new level, just as it has in almost every other industry. From slashing the time and cost of discovering new drugs to enabling hospitals to predict demand for beds with much greater accuracy, we are promised that AI will bring an even greater transformation in healthcare than the digital revolution.

And in the UK, where the National Health Service is lurching from crisis to crisis, many are latching on to AI as the answer to boosting productivity and starting the long-awaited move to a preventive approach to healthcare, rather than waiting until people are seriously ill before engaging with them.

But while I have heard plenty of stories about radical AI innovations, both from commercial businesses trying to engage with the NHS and from a new wave of healthcare professionals eager to play a part in this revolution, I have heard very little evidence that many have actually been deployed widely in hospitals or GP surgeries. In particular I’m hearing from lots of tech companies that the NHS is very open to pilots of AI-driven innovations - especially if they are free - but. then even when a technology has been proven, very reluctant to spend the money to bring it to patients.

So I thought it was time to take the temperature of AI in the NHS by talking to some of those battling to take it forward. Here is the first in what I hope will be a series of interviews where I try to discover whether, behind all the hype, we can hope to see artificial intelligence give us a better health service.

************************************************************************************************************

I am starting with Tom Whicher, chief executive of DrDoctor, because he gives me a cautiously hopeful vision of how AI can be implemented in the NHS. For more than a decade Tom’s company - which I wrote about in January - has been working to improve the way the NHS interacts with patients, notably in areas like the management of appointments.

Tom Whicher, Founder,DrDoctor

Tom is down-to-earth about the almost hysterical wave of excitement about AI in the NHS - “there's absolutely a hype cycle around it - you know, ‘if you're not doing AI, what are you doing?’ which is a bit ridiculous really.”

But he separates the opportunities for AI deployment into two areas:  “There's clinical AI which is your cancer diagnostic imaging type stuff, which is really well proven.  The bit that we're good at is the other half, which is AI to support back office - and it's a bit less sexy but the opportunity is almost greater.”

I point out - and he agrees - that while the “diagnostic imaging stuff” such as the development of AI systems which can understand scans and detect cancer as well as or even better than human radiologists is well proven, there are few examples of them being put into clinical use.

Whereas boring old unsexy stuff like helping hospitals understand which patients are likely to fail to turn up to appointments is providing good business for DrDoctor. Tom Whicher says this is a case of using AI - or more specifically machine learning - “to improve what we already do.” Prediction algorithms allow their hospital clients to address the “Did Not Attend” problem which costs the NHS £1.6 billion a year

“This person here, they always miss their appointment. They're going to miss it again. They have diabetes.” That means eventually they may become an emergency admission with a worse outcome for them and a much higher cost for the NHS.”

So the algorithm suggests a targeted extra intervention. “Phone them up usually, so that you have a conversation with them. You can talk to them about what's stopping them coming in.”

This sort of AI technique, which works well if there is sufficient data available, has been around for years. So I wonder where Tom sees generative AI playing a role in interactions between doctors and patients. He points to a vast stream of unstructured data flowing round the NHS in the form of letters. often between hospital doctors and GPs.

These have a peculiar format, often starting with “this charming gentleman” and then rapidly adopting some extremely technical language to describe the results of a consultation with the patient. DrDoctor realised that these letters contained all sorts of data which could be used to answer questions from patients such as “what was the result of my blood tests” using a Large Language Model.

“What we realised was, we can get the test result from the computer system and we can use a Large Language Model to read the letter, pull out the interpretation, summarise it and display it next to the test results for the patient digitally. Suddenly, you can give context to this data. And so that's an example of AI which is administrative but, my gosh, it makes a big difference to the person.”

Tom also points to a problem we are all familiar with where generative AI could help - your doctor focusing on their computer, not you. “You're there with a doctor and rather than looking at you, they're over here and they're typing. And it's not a really good experience for anybody.” During our conversation over Zoom, I am focused on Tom, rather than typing away making notes. That is because I am using a program called Otter.ai, which records our conversation and transcribes it, and in the last year or so has also produced a rather clever summary of our discussion and the main themes.

Now, he tells me, a clutch of companies are designing something similar for doctors.

“So the consultant goes, ‘I'm going to use AI to record this conversation. Are you happy with that?’ Yes I'm happy with that……. And the AI makes the notes like Otter does, but it also pulls out the pertinent details. This patient has the following co-morbidities, the diagnosis is this and the next steps are that, and then it can code that all up into the EPR (Electronic Patient Record). And that's huge for doctors because in an average clinic, a doctor spends 25% of their time making notes.”

This kind of use of AI as a time-saving tool may prove less controversial than its application in diagnostics and other clinical areas where there is a fear from some doctors that it may cost jobs. But I put to Tom Whicher that the real frustration I hear from companies like his is with NHS financial managers who are eager to allow free pilots of innovative technology but less keen to pay to deploy it. But he says both sides share some blame for the slow pace of AI adoption in the NHS:

“I think organisations often use pilots as a way of getting things for free, because they don't have any money. That's kind of unhelpful because if you want to scale something, you have to fund it. And I think there's also often a naïveté from businesses - just because they built something which is kind of cool that doesn't mean it's value adding.”

In this series of interviews on AI in the NHS, I’ll be looking out for ideas which are both cool and value adding. Do get in touch if you think you have something that fits the bill.