Recognizing kidney injury due to burns is improved by artificial intelligence
Machine learning cuts critical time to managing acute kidney injury
Many burn victims suffer from acute kidney damage (AKI), but early detection of AKI remains difficult. Now, an artificial intelligence/machine learning (AI/ML) model developed at UC Davis Health and presented in a new study can predict acute kidney damage more quickly and accurately than ever before.
"The ability to predict acute renal failure in burn patients with AI is a potential breakthrough for burn centres," said Tina Palmieri, Professor and Director of the Firefighters Burn Institute Regional Burn Center at UC Davis Medical Center. "If we can say that a patient may have kidney damage, we can take steps to prevent it."
What is an acute kidney injury?
Acute renal injury (ARF) is a sudden renal failure or renal injury causing an accumulation of waste products in the blood and fluid imbalance in the body. Acute renal failure usually occurs during the first week of severe burning due to inadequate resuscitation, especially during the first 24 critical hours. Developing in about 30% of cases, IRAQ is a frequent complication after a severe burn, with a mortality rate reaching 80%.
Diagnosis of acute kidney damage
Physicians generally rely on traditional biomarkers such as serum or plasma creatinine and urine flow for diagnosis. However, urinary output and creatinine measurement are considered to be poor biomarkers of acute renal failure.
"The University of California at Davis was the first to identify the role that a novel biomarker, known as neutrophil gelatinase associated lipocalin (NGAL), plays in the early prediction of acute renal failure in severely burned patients," said Nam Tran, Clinical Professor in the Department of Pathology and Laboratory Medicine at the University of California at Davis.
Despite its strong predictive power, NGAL was not available in the United States and its interpretation required more experienced clinicians and laboratory experts. This challenge has led to the development of an artificial intelligence automatic learning model to facilitate the interpretation of NGAL test results.
Machine learning improves recognition of acute kidney damage
Sometimes, in the AI/ML world, it is assumed that more complex algorithms, such as deep neural networks, are better than more traditional algorithms at building ML models. This assumption is not always true.
"We have built a powerful ML model with our k-Nearest Neighbor approach that is able to accurately predict ARF in our patient population over a much shorter period of time," said Hooman Rashidi, Professor in the Department of Pathology and Laboratory Medicine at UC Davis Health. "According to admission data, the model can shorten the time to diagnosis by up to two days."
The models were trained and tested using laboratory clinical data for 50 adult burn victims including NGAL, urine flow, creatinine and NT-proBNP were measured within the first 24 hours of admission. Half of the patients in the data set developed acute renal failure within the first week of admission. Models containing NGAL, creatinine, urine and NT-proBNP have achieved an accuracy of 90-100% to identify IRAQ. Models containing only NT-proBNP and creatinine have achieved an accuracy of 80 to 90%.
The average time from admission to diagnosis using traditional biomarkers was 42.7 hours. The average time of use of the ML algorithm was only 18.8 hours. The ML model beat the traditional method by nearly a full day - a critical time to prevent and treat acute kidney failure.
"For our study, AI/ML showed the potential clinical utility of predicting acute renal failure due to burns when only a few common laboratory results are used," added Tran.
Applications and implications of the new model
This model has applications to be used in the field, including for military losses. Since troops can be sent to hospitals that do not have the necessary facilities to treat kidney damage, the AI/ML method would allow patients with acute kidney failure to be identified more quickly so that they can be referred more quickly to advanced medical facilities. This optimizes limited resources on the battlefield and speeds up the transport of patients to where they need to go. The same process applies in the civil world.
"We are considering integrating such automatic learning platforms into various contexts outside the IAR, which could improve various aspects of patient care in the clinical medicine arena," added Mr. Rashidi.
