While the UK’s Presymptom Health is by no means alone among healthcare startups globally in using artificial intelligence (AI) to offer products or services that would not otherwise be possible, it is undoubtedly unique among those doing so in how it came about.

Founded in 2019, the company was spun out of Ploughshare, a Ministry of Defence (MoD)-owned entity created in 2005 to commercialise intellectual property (IP) from UK defence and government research. Ploughshare identified work by the MoD’s Defence Science and Technology Laboratory (Dstl) research and development agency as being potentially transformative for infection diagnosis.

Initially focussed on sepsis, the technology could determine whether a patient would develop the condition up to three days before symptoms appeared. Trained on data from a 12-year study, it used AI and its subset machine learning (ML) to diagnose infection at an earlier stage and with higher accuracy than was previously possible.

Infection research

“They were gathering many samples and lots of data from 4,500 patients, some of which would go on to develop illness,” explains Iain Miller, CEO of Presymptom Health, to Medical Device Network. “The insight was, if you recruit them before they're sick, you create a resource to train an algorithm that sees the sickness coming rather than reacting to it and picking it up when it's rather late and you might have less opportunities.”

The output of that study, which was started shortly after 2000, was a biobank containing in the region of 70,000 samples – a resource that would become all the more valuable as the emergence of AI during the 2010s allowed for a step change in data interrogation.

“We had this huge resource, and this is a classic case of where you can use machine learning,” says Miller. “You've got a lot of data, and, to use the parlance of AI and machine learning, they talk about having a lot of ‘features’. What you've got is a lot of features in your data, and you've got outcomes of interest, so you’re trying to train up a set of features that are diagnostic or predictive of the outcome of interest.

“In our case, the features were all of the genomic information and the proteomic information. The outcomes were the clinical statuses of these patients and all the other metadata associated with that. We were able to train from one to predict the other, and so we kind of became something of a software company in that we were doing all that training.”

Test development

The use of ML in this way allowed Presymptom Health to develop a first-generation polymerase chain reaction (PCR)-type blood test similar to those widely used during the Covid-19 pandemic, with samples taken and sent for analysis based on the featured identified in the biobank data. A third-generation test being developed will provide results in situ without needing to be sent away – and Miller says one aim is to take that back into the military.

“Historically, military casualties and deaths, through the history of warfare, really mainly come from infection,” he says. “You get a minor wound, and you die of infection. That's not the case any longer, but, even now, you still want to know if the patient has an infection, is at risk of sepsis or if they're about to get other troops infected. Should they be pulled out of the deployed force because they would bring other people down with them?”

Ahead of the “smaller, cheaper, field deployable” third generation of tests, there are the first and second generations to roll out, and work continues on expanding and refining the company’s knowledge base.

“Part of our job has been running additional trials to get new data, doing additional machine learning and then testing the things that we derive from the machine learning on the new data, independent data sets. We've just finished a nine-hospital trial called Precision, and we're planning Precision 2, a successor trial, right now.”

AI in healthcare

Of the role that AI is playing for Presymptom Health and others in the healthcare space, Miller believes that simply being able to interrogate data in such a way as to identify associations that would not otherwise be apparent is a major development.

He explains: “In my field, with infection diagnostics, the way it would have been done traditionally, they'd say: ‘Well, certain parts of the genome are associated with infection or infection response,’ and you'd have focused on them. But with AI, you look at the non-obvious stuff. You throw a ton of data at it and you find stuff. You find unknown biology, which also tends to be more patentable because it's not obvious.”

By way of example, he points to the third-generation test that Presymtpom Health is developing, which he likens to the lateral flow, or 'dipstick', tests used for Covid-19.

“Going back to the Covid analogy, if you want to make it really cheap and dipstick-compatible, you've got to look at proteins that are easier to look at than genes,” he says. “So, you want to do machine learning with the protein complement of blood versus the genomics.”

More broadly, Miller offers an optimistic – if tempered – view of the role AI will play in healthcare.

“There's no doubt it's really exciting,” he says. “It will be slower to come to healthcare because of the regulatory need to slow it down, in a sense. So, we are definitely AI-derived, but we and all the other AI healthcare companies will not initially be offering AI in the clinic. We'll be offering something derived from AI – but locked down and then offered in a more traditional way thereafter.”

Support and funding

Considering what has set Presymptom Health apart in its successes and helped it to secure funding where others may have been unable, Miller explains: “Two of our big advantages were this big set of data and also the fact we had support from government – because I knew that private investors would be slow to dip into their pockets. Both the support we've had from the Ministry of Defence and, more recently, from the innovation departments within the government and also the Department of Business and Trade have been really incredible.”

Few startups – in healthcare or elsewhere – benefit from having been spun out of a government-owned organisation, and Miller acknowledges that, noting: “We were lucky. We came out of the defence establishment, so we had a toehold in government.”

But that, of course, is part of how Ploughshare is able to help commercialise the valuable IP that comes out of the UK’s defence and government research. It is partly through building on that support that Presymptom Health is one of eight companies – whittled down from 80 – to have been accepted onto the UK’s Innovative Devices Access Pathway, an initiative to bring new medical technologies to the National Health Service to help with unmet medical needs.

“We're working closely with the relevant branches of government on that,” says Miller. He adds though, that, despite having been spun out of the defence establishment and retaining close links with government, Presymptom Health operates fully independently.

Spinning out

“We eventually were deemed to have grown up a bit after about two and a half years and kind of left alone,” explains Miller. “Now, I don't have to go to all the committees, I don't have to do different things, but there was a lot of that for good compliance. There's public money involved.

“There was husbanding a resource, and, while it might have been irksome to me, it was all well-intentioned and useful. It probably pressure-tested me and helped me refine my ideas. So very involved, initially, still involved as an advisor. Always helpful in terms of setting up, introducing people.”

“Ploughshare is good at spinning out companies using government knowledge assets. That's their USP, their DNA if you like. That's what they do.”