AI in the NHS - thinking small
"It's just maths" says Polygeist, the nerdy minnow winning AI contracts
“Nobody ever got fired for hiring IBM,” was the old cliche about IT procurement and a safety first mindset that saw giant companies given huge contracts, with little regard for value for money. In the case of the NHS the likes of Fujitsu and EDS have also been among the giants handed colossal sums for ill-fated projects, with smaller suppliers sitting on the sidelines, struggling even to get a meeting, let alone a contract.
So I was interested that the name of a small company kept cropping up in relation to efforts to use artificial intelligence in healthcare. Polygeist had been asked to work on two NHS AI Skunkworks projects. One, described to me by the Skunkworks boss Giuseppe Sollazzo, aimed to use machine learning to find ways of identifying patients who might stay longer in hospital than expected, another to speed up the time it took to identify Parkinson’s in the brains of deceased donors.
So I wanted to know more about Polygeist - where did it come from, how did it get noticed by the NHS, just how small a business was it? The answer to that last question when I got on a call with the CEO Dr Peter Coetzee was…pretty small.
Polygeist was started nearly five years ago by Peter, a computer scientist, Dr Brad Pearce, a neuroscientist who’s moved into computer security and Dr Vince Calloway, an applied mathematician. But recently the workforce has doubled: “We've reached the lofty heights of half a dozen people now,” says Peter Coetzee with a grin. He says they have recruited people with similar academic backgrounds - “I think everyone we've hired has been a proper nerd!”
So - how on earth did this nerdy minnow get into the NHS? It turns out that the unlikely route was via the Home Office. They began with security projects for defence and law enforcement and were soon talent spotted by a body called ACE, the Accelerated Capability Environment. This Home Office unit acts as an advisor on technology procurement to various government departments, keeping a list of verified suppliers big and small, helping run the tendering process and getting projects up and running within weeks.
Polygeist spotted the Skunkworks project to cut overstays in a Gloucestershire hospital on ACE’s list and won the contract, The idea was to develop a system which would grade patients arriving in the hospital from 1 to 5, with 1 meaning they were likely to be out in a couple of days, 5 that they were very likely to overstay with the elevated danger that they would never go home. “We said, give us eight weeks and we can tell you whether it's possible,”says Peter Coetzee. “And then give us four more weeks if it is possible, and we'll get it into your hospital patient record system, and it turns out, they really liked that.”
Polygeist soon showed it could predict overstayers pretty accurately, enabling doctors to adapt their care to lessen the risk. After the 12 week initial trial they have gone on to refine the system, which is now 95% accurate in differentiating between overstayers and those who can leave hospital faster than expected - knowledge that can improve their care too: “Let's say I have a frail elderly patient coming in, if I know they're only going to stay for one or two days, don't cancel the care package at the start of their stay that they have at home.”
The Parkinson’s project, which also came to the company via ACE, is at the other end of the spectrum. Speeding up the process by which scientists spot signs of the disease in the brain. after death may help advance the long term challenge of understanding the condition but does nothing for patients in the short term. But Brad Pearce, who’s running this project, says there are hopes that the technique they have developed could be used to detect evidence of other neurological conditions in live patients.
At a time when there is so much hype around the potential of artificial intelligence in healthcare Polygeist’s co-founders were refreshingly cautious about it. “We aren't an AI company,” Peter Coetzee goes as far as saying. “I don't really care what technology we need to use to be able to build a solution to an operational problem. I care about whether we are able to provide a solution.”
“Our job really is to go on the opposite side of the hype train,” says Brad Pearce. "We're going the other way and trying to explain - actually, no, no, it's just maths.”
Of course there are plenty of giants - from Microsoft to Google to China’s Tencent -making a lot of noise about their investment in AI for healthcare. But Polygeist seems to show that a cautious nerdy minnow can also make an impact.
Hi Rory, I work in healthcare and have found that there are quite a few small companies that are really good in the AI space. Like you've said, it's "just maths", and perhaps you don't need that many people to that maths - but they need to be top quality uber nerds to do a good job. I can imagine it might well appeal to these people to set up their own firms, and/or work for smaller companies.
I worked with a Dutch company called Xomnia who are like a slightly bigger version of Polygeist.