AI in the NHS: Pearse Keane, Still Dreaming

The long road from code to clinic

For my second interview in this series about AI and the NHS I have chosen a man who can see both the enormous potential of the technology and the challenges in getting it deployed in the health service. Pearse Keane is a consultant ophthalmologist at London’s Moorfields Eye Hospital and a Professor of Artificial Medical Intelligence. His life changed in 2015 when he read an article in Wired Magazine about Google Deep Mind, the London-based business doing extraordinary things with AI.

Pearse Keane

He ended up working with the company on an algorithm to triage the millions of sophisticated eye scans now coming out of the UK’s High Street opticians, which risk overwhelming the NHS with false positives for dangerous eye conditions. By 2018, the work was done, and the team including Pearse Keane published a paper in Nature Medicine claiming that the algorithm could diagnose a host of eye diseases as well as or better than world-leading eye doctors.

I have interviewed him a number of times and he has often given me a physical expression of the difference between the time taken to come up with an innovation, and the years it takes to put it into clinical practice. “From idea to code,” he will say, opening his hands and spreading them just a little, “maybe a couple of years. But code to clinic…..”, at which he spreads his arms as wide as possible. Still, when I spoke to him in January 2022 he was hopeful that within two years an AI system that could benefit millions of patients would be in use around the world.

When I got in touch this week, however, it was clear that this had not yet happened. But he reminded me just how big a difference there is between a piece of software described in a paper for Nature and a finished clinical commercial product, approved by regulators:

“What you've got when you publish a paper is a piece of experimental code - and what you need to make is a product. So you need to rewrite the algorithm under a quality management system, essentially documenting what every line of the code does in an industry standard way. That takes considerable software engineering.”

The good news is that Moorfields Eye Hospital is leading the next phase of development to get the product market-ready, aiming both to benefit patients and to provide a commercial return to the NHS. The bad news is that getting through that process is a lot harder than people assume.

He goes on to explain that the NHS is not set up to run functions such as a customer service department, and the kind of revolution that implementing an AI system will involve will not be easily done without wider collaboration. But Pearse reminds me again why the need for help from such a system is growing ever more urgent.

He tells me that ophthalmology is the busiest outpatient service in the entire NHS, accounting for nearly 10% of all appointments, up by a third in the last five years. Last year Specsavers alone did nearly 8 million scans using advanced machines which can spot conditions like AMD (age-related macular degeneration.)

“There are people going blind in the UK because of delays in being seen and treated,” he says. “Imagine if you're blind in one eye, and you're losing sight in your good eye and you're told you need to wait a few weeks to be seen and to be treated. That was the motivation for me to do this.”

Pearse Keane insists he is not giving up: and that “the dream is still alive” but he recognises the need to keep up with the fast-evolving AI field

For the NHS and other healthcare systems, however, there is a difficult dilemma when it comes to implementing a fast-changing technology. The Silicon Valley mantra “move fast and break things” really doesn’t work in health - witness the disaster of the Theranos blood-testing business. You can understand why, given the wider safety concerns about AI, an inherently cautious NHS is proceeding with extreme care, however frustrating that may be for those of us eager to see the UK make good use of innovations born here.