The risk of acute kidney injury (AKI) in intensive care unit (ICU) patients is as high as 50%, which not only may lead to deterioration of other organ functions but also increase hospitalization days, the risk of renal replacement therapy, and mortality rates. The team at Taichung Veterans General Hospital has developed a predictive system for acute kidney injury that can infer the risk of AKI in ICU patients 24 hours in advance, enabling early intervention to improve outcomes. 

Hourly prediction of AKI risk for early intervention and improved prognosis 

Tsai-Jung Wang, MD stated that patients in the ICU are critically ill, and their condition can change every minute. However, there has been a lack of continuous AKI prediction systems in the past. Once AKI occurs, clinical intervention becomes difficult, and only supportive treatment can be provided. Therefore, the development of a continuous warning system capable of predicting the probability of AKI occurrence early is necessary. 

The acute kidney injury prediction system developed by Taichung Veterans General Hospital is an interpretable machine learning model that can infer the risk of AKI occurrence in ICU patients 24 hours in advance. Dr. Wang Cai-Rong mentioned that the model has undergone cross-institutional verification and joint training with four other medical centers, maintaining good predictive discriminative power, with an AUROC value of above 0.911. Once installed in the hospital's server, it can collect and transmit patient physiological, medication, laboratory, and related data in real-time for online inference. In addition to providing the probability of AKI occurrence in patients after 24 hours, it also presents the ranking of feature weights to enhance clinical interpretability. 

Four highlights of the prediction system! Precision, safety, convenience, intelligence 

the breakthrough technologies of this system include: 

• Using a reasonable number of features to analyze large amounts of data, utilizing real-time clinical physiological, medication, and laboratory data to predict the risk of AKI in patients. 
• Deploying a federated learning platform, allowing collaborating hospitals to train models without exchanging proprietary medical record data, thereby maintaining data confidentiality. 
• Designing interactive dashboards for healthcare personnel to view prediction results in real-time and provide feedback interactively. 

• Integrating hospital information systems for bidirectional transmission of real-time data into inference models and inference results feedback. 

The acute kidney injury prediction system developed by Taichung Veterans General Hospital is an innovative AI technology capable of continuously predicting AKI risk in ICU patients, aiming to enhance the quality of critical care and improve the prognosis of critically ill patients. 

Resource (Mandarin): 

AI預測急性腎損傷　提早24小時預警ICU病患風險