DeepMind, a UK-based artificial intelligence company owned by Google, unveils an incredible breakthrough in predicting a deadly kidney condition two days before it occurs.
One of the major causes of death in the hospital, acute kidney injury or AKI is the rapid decline in kidney function. It kills hundreds of thousands of people every year, but it's challenging to prevent since it happens very quickly and with little warning.
To help doctors and patients fare better against AKI, scientists from DeepMind turned to artificial intelligence for help.
DeepMind Figures Out How To Detect AKI Early
In a paper published in the journal Nature, DeepMind revealed that they have created a new artificial intelligence algorithm to find signs suggesting a patient is at risk of AKI and predict the condition 48 hours before it occurs.
Findings showed that the AI algorithm was 55.8 percent accurate in predicting overall AKI cases. In cases that are severe enough to require dialysis, it was 90.2 percent.
For the study, the researchers used data of 703,782 adults from the records of the U.S. Department of Veterans Affairs. The records include over 600,000 health indicators, such as vital signs, blood tests, procedure history, and more. From these indicators, the team picked out 4,000 that are potentially danger signs of AKI. From these indicators, DeepMind's AI algorithm can detect the risk of a kidney failure before it occurs.
"Clinicians being able to move from reactive to the ability to predict [AKI] two days before — that's due to the richness of the data," Dominic King, DeepMind Health product lead, explained in Scientific American.
Predicting AKI in advance allows medical professionals more time to prevent or mitigate its effects on the patients.
What Other Researchers Say
Joseph Vassalotti of the National Kidney Foundation, who was not involved with the development of the algorithm, told Scientific American that the program is promising, but he's not certain how it would work in an actual hospital. Since the VA records used by the DeepMind scientists are extremely comprehensive, it's not certain whether the algorithm would perform as well in hospital settings when patients don't have extensive prehospitalization records.
Vassalotti also added that the study also showed plenty of false positives, but according to King, the false positives weren't a big concern since half of them correctly simply didn't predict in 48 hours or were trailing false positives where there was an AKI episode with the algorithm predicting higher risk after. The other half occurred mostly in people with existing renal issues.
