Intelligent Critical Care Center (IC3) researchers have published a new study in Surgery about an IC3-developed, dynamic deep learning model that predicts post-operative acute kidney injury (AKI) stages for the next 24 hours. After surgery, AKI can affect 10%-30% of patients and is associated with the development of chronic kidney disease, and severe AKI suggests more aggressively deteriorating clinical outcomes and a greater chance of death.
The article, “A deep learning–based dynamic model for predicting acute kidney injury risk severity in postoperative patients,” details how the IC3 deep learning model was more accurate than traditional logistic regression models in predicting every next-day AKI stage, except Stage 3 with Renal Replacement Therapy (RRT) where the predictions were of similar accuracy.
To test the deep learning model, IC3 researchers created a dataset that included electronic health records (EHR) data of 51,806 surgeries from 42,906 surgical patients admitted to University of Florida Health between 2014 and 2021. The deep learning model made the next-day AKI predictions based on EHR data from the hospitalization period and the one-year history prior to the hospitalization.
According to the IC3 study, “The proposed model demonstrates that temporal processing of patient information can lead to more granular and dynamic modeling of acute kidney injury status and result in more continuous and accurate acute kidney injury prediction.”
Ultimately, the goal is to implement the model as part of a clinical decision-support system that can alert physicians and thus help them to prevent AKI from developing and to prevent AKI from progressing in severity.
Authors included IC3 co-directors Azra Bihorac and Parisa Rashidi, associate director of research Tezcan Ozrazgat-Baslanti, assistant director for clinical AI Benjamin Shickel, research scientist Yuanfang Ren, and postdoctoral associate Esra Adiyeke.
Read the full Surgery article here.