VA, Alphabet's DeepMind develop AI system to predict acute kidney disease 48 hours in advance

One out of five patients admitted to the hospital for serious health conditions develops acute kidney injury, with a more than fourfold increased likelihood of death, according to one study.

DeepMind, a London-based artificial intelligence lab owned by Google parent company Alphabet, says it has developed an AI system that can predict the onset of acute kidney injury earlier than most doctors can spot it, giving physicians a 48-hour head start in treating the condition.

In a collaboration with the U.S. Department of Veterans Affairs, researchers used deidentified electronic health record data from more than 700,000 patients collected from 1,200 VA inpatient and outpatient healthcare facilities to develop the model. The AI system analyzes the patient’s health records, including blood tests, vital signs, and past medical history, and can accurately detect acute kidney injury up to two days earlier than it is currently diagnosed, according to VA and DeepMind officials.

The model correctly predicted 9 out of 10 patients whose condition deteriorated so severely that they then required dialysis, Mustafa Suleyman, DeepMind co-founder and head of applied AI, and Dominic King, DeepMind's health lead, wrote in a blog post.

"This could provide a window in the future for earlier preventative treatment and avoid the need for more invasive procedures like kidney dialysis," Suleyman and King wrote.

The team's research was published in the July 31 edition of Nature.

RELATED: Lawsuit accuses University of Chicago of sharing identifiable patient data with Google

Acute kidney injury is notoriously difficult for doctors and nurses to detect; when it occurs, patients often deteriorate very quickly, according to VA officials. The AI model permitted the identification of over 90% of the most severe acute kidney injury cases 48 hours sooner than with usual care. That early detection permits improved medical care that can reduce progression to serious consequences such as the need for dialysis, the VA said.

“These are exciting times for research and innovation at VA,” VA Secretary Robert Wilkie said in a statement. “Studies like this can have a significant effect on not only the Veteran community but people throughout the nation.”

Suleyman and King said the model researchers developed also addresses the "black box" problem, one of the key barriers for the implementation of AI in clinical practice. The model provides the clinical information that was most important in making its predictions of deteriorating kidney function, they said.

RELATED: VA taps Google’s DeepMind to predict patient deterioration

"It also provides predicted future results for several relevant blood tests. This information may help clinicians understand the reasoning behind the AI-enabled alert and anticipate future patient deterioration," Suleyman and King wrote.

Research projects that involve corporate labs getting access to patient data from hospitals and health systems have raised questions about patient privacy. DeepMind has faced scrutiny over a controversial patient data-sharing arrangement with Britain's National Health Service. In 2016, DeepMind was accused of violating patient privacy after it struck a deal with the NHS to process medical data for research, according to the New York Times.

Last month, a lawsuit filed against the University of Chicago, its medical center and Google accused the hospital of sharing hundreds of thousands of patient records with the tech giant that retained identifiable date stamps and doctors' notes. 

Moving forward, the VA Palo Alto Health Care System in California will be looking at ways to bring the AI system into clinical use, VA officials said. The work leading up to this clinical trial involves complex interdisciplinary coordination to build and integrate a user-friendly platform to assist clinicians with treatment decisions, VA said.