DeepMind's AI can predict AKI up to two days before it happens, according to new research

The study was published today in Nature.
By Cara Dartnell-Steinberg
02:36 pm
Share

Credit: DeepMind

Currently, more than 26 clinicians are using Streams at the Royal Free Hospital in London, allowing them to detect an average of 11 patients daily who are at risk of AKI, most in fewer than 15 minutes rather than the multiple hours the process usually requires.

Streams processes patient test results, rapidly prioritising the patients most at risk and allowing for more immediate treatment to be identified and delivered. On the app, medical information is compiled and the most important data is displayed first, permitting clinicians to review tests results easily from any location. If an issue is spotted, Streams sends a smartphone alert and the patient’s medical history to the appropriate clinician so they can make a diagnosis. Clinicians can also access the patients’ vital signs and record observations into the app.

WHAT HAPPENED

New research completed in collaboration with the US department of Veterans Affairs has illustrated that AI technology could be taught to learn how to interpret test results and to select the appropriate treatments for individual patients. The AI algorithm can predict the presence of AKI up to 48 hours before it happens.

WHAT'S THE IMPACT

AKI results in the death of 100,000 people in the UK (a quarter of which the NHS believes to be preventable) and 500,000 people in the US each year. AKI requires treatment within hours, or even minutes, of diagnosis. It is one of the conditions most commonly associated with patient deterioration, which is believed to be the cause of 11% of all in-hospital deaths and costs the NHS over £1 billion annually. Upon implementation of the app, healthcare costs reduced by 17% per AKI patient at RFL and recognition of AKI improved from 87.6% to 96.7% for emergency cases.

THE LARGER TREND

The research illustrates the possibilities of AI in improving earlier identification and treatment and allowing for the shift of medicine from treatment to prevention. DeepMind hopes to add more features to Streams, including alerts for other preventative conditions such as organ failure and sepsis.

More broadly, The Joint Commission in 2014 demonstrated that errors in human factors and inter-clinician communication are some of the most common causes of unsafe conditions in hospitals, and National UK data indicates that a large proportion of safety incidents in the NHS were related to the use of papers rather than phones or other technology. By using mobile apps such as Streams, errors such as these can be reduced and the process of diagnosis and treatment can be streamlined, especially with the integration of AI technology in the future.

ON THE RECORD

Sarah Stanley, Consultant Nurse at the Royal Free London NHS Foundation Trust commented, “The true benefit of Streams has been having the information in a useful way in one place and being able to make expert clinical decisions in a relatively short space of time.

Dr Chris Laing, Consultant Nephrologist, who led the clinical implementation of the Streams app, said: “Thanks to Streams we are able to monitor the kidney function of patients through real-time analysis of blood tests 24/7. If a potential change in kidney function is detected, at any time or anywhere at the Royal Free Hospital, a specialist will be notified and the case will be reviewed, in-application, in a matter of minutes, with follow-up bedside assessment, as required.”

Mary Emerson, lead nurse specialist for the RFL patient at risk and resuscitation team, commented: “The Streams app has made a huge difference to clinicians’ ability to respond rapidly to patients who are developing acute kidney injury. This means we can deliver treatment more quickly, and also identify deteriorating patients much earlier. The mobile technology is easy to use and fits with the way healthcare is delivered today. I’m excited about the possibilities this approach might have for other conditions and clinical teams.”