Artificial intelligence can help accurately predict acute kidney damage in burns

Many burn victims suffer from acute kidney damage, but early recognition of these infections remains difficult. Now, an AI / ML model developed by UC Davis Health and described in a new study can predict acute kidney damage more quickly and accurately than ever before.

The ability to predict AKI in burn patients using AI is a potential breakthrough for burn treatment centers. If we can say that a patient could have a kidney injury, we can institute measures to prevent it. "

Tina Palmieri, Professor and Director, Burnfighters Burn Institute Regional Center for Burns at UC Davis Medical Center

What's an acute kidney injury?

Acute kidney damage is sudden renal failure or injury that causes a build-up of waste in the blood and fluid imbalance in the body. The AKI usually occurs during the first week of severe burns due to inadequate resuscitation, especially during the first 24 hours critical. Developed in approximately 30% of cases, AKI is a common complication after a severe burn, with a mortality rate as high as 80%.

Diagnose acute kidney injury

Physicians generally rely on traditional biomarkers such as serum / plasma creatinine and urinary flow for diagnosis. However, urine production and creatinine measurements are considered as poor biomarkers of AKI.

"UC Davis was the first to identify the role that a new biomarker, called lipocalin badociated with neutrophil gelatinase (NGAL), plays in the early prediction of AKI in critically burn patients," he said. said Nam Tran, badistant clinical professor in the Department of Pathology and Laboratory Medicine. at UC Davis.

Despite its strong predictive power, NGAL was not available in the United States and its interpretation required a greater number of clinicians and laboratory experts. This challenge has led to the development of an artificial intelligence machine learning model to facilitate the interpretation of NGAL test results.

Machine learning improves recognition of acute kidney injury

Sometimes, in the AI ​​/ ML world, it is badumed that more complex algorithms such as deep neural network are better than more traditional algorithms for building ML models. This hypothesis is not always true.

"We have developed a powerful ML model through our k-Nearest Neighbor approach, which accurately predicts AKI in our patient population in a much shorter time frame," said Hooman Rashidi, Professor, Department of Pathology. and from UC Davis Health Laboratory Medicine. "Based on the admission data, the model can reduce the two-day diagnostic delay."

The models were trained and tested with clinical laboratory data for 50 burnt adult patients with NGAL, urine output, creatinine and NT-proBNP measured within the first 24 hours after admission. The AKI was developed by half of the patients in the database during the first week following admission. Models containing NGAL, creatinine, urinary flow and NT-proBNP have achieved 90-100% accuracy for AKI identification. Models containing only NT-proBNP and creatinine achieved an accuracy of 80-90%.

The average time between admission and diagnosis with traditional biomarkers was 42.7 hours. The average time of use of the ML algorithm was only 18.8 hours. The ML model has gone beyond the traditional method of almost a full day – a critical time to prevent and treat the AKI.

"For our study, the IA / ML showed the potential clinical utility of predicting burn-related AKI by using only a few routine laboratory results," he said. added Tran.

Applications and implications of the new model

This model has applications for use in the field, including for military casualties. Given that troops can be sent to hospitals without the facilities needed to treat kidney damage, the IA / ML method could identify patients with IRA faster, so that they can be transferred more quickly to advanced medical facilities. This optimizes the limited resources on the battlefield and speeds the transport of patients to where they need to go. The same process applies in the civilian world.

"We envision the integration of such machine learning platforms into a variety of contexts outside of AKI, which could eventually improve various aspects of patient care in the field of clinical medicine," he said. added Rashidi.