DeepMind's work with the NHS to help alert doctors to patients whose health is at risk is laying the groundwork for delivering information that one day will be powered by artificial intelligence.
The Google-owned British pioneer is working with the Royal Free London NHS Trust on a smartphone app called Streams. It currently uses an NHS created algorithm to provide information to clinicians relating to acute kidney injury (AKI), with DeepMind creating the method of delivery that includes "breaking news" style notifications. It does not use AI despite the companies expertise in the technology.
But, the work on Streams is in preparation for a future where artificial intelligence will be behind the information being delivered to doctors and nurses.
"The reason why we've gone to this trouble.... is it allows us to think about how we enable AI in the future," said Dominic King, senior clinician scientists at DeepMind speaking a the AI Summit, indicating that it could be behind such algorithms, but equally it may be university researchers, other startups or another healthcare organisation.
"So that when someone, or we, come up with a better algorithm for detecting acute kidney injury, or when there is the ability to identify problems, we now have a way [to deliver an alert or notification]," he said.
While researchers and other companies are working on AI powered algorithms for healthcare, King said delivering it would be the problem.
"We think it [AI] can really help the NHS and the current challenges it faces, but I would say strongly that for clinical impact, we need solutions that are both practical as well as smart," said King.
Hospital blood tests are usually done in the morning, but results are not necessarily looked at until the afternoon or next morning. The Streams app can alert experts to signs that a patient is at risk before symptoms escalate. The cost of AKI to the NHS is estimated at £1bn per year and its hoped earlier detection can help save costs in the healthcare system.
King also lifted the lid on how the trial of Streams is going three months in and how it is trying to get the delivery and presentation of data right for clinicians. Information from from more than 2,000 blood tests on average per day are being analysed, identifying 66 changes in patients' kidney function.
King said this number of alerts was "unmanageable" for specialists however, and has worked with them to filter them down to the ones they really needed to deal with - around 10 or 11 a day. 60 per cent were reviewed in less than one minute.
While the figures are anecdotal, it is currently making a more detailed evaluation, said King, and the company plans to roll out the Streams app to new teams within the hospital trust which will cover other conditions such as sepsis and organ failure.
DeepMind's five-year deal with the NHS trust has proved controversial in regard to privacy and transparency. It has since upped its efforts to assure data is handled in a transparent way.