Interpretable Machine Learning for Early Prediction of Acute Kidney Disease (AKD) in Sepsis-Associated Acute Kidney Injury (SA-AKI): A Multicenter Cohort Study with External Validation

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Sepsis-associated acute kidney injury (SA-AKI) represents a critical challenge in the management of critically ill patients, significantly contributing to morbidity and mortality in intensive care units (ICUs). This study aims to enhance the understanding and prediction of SA-AKI progression to acute kidney disease (AKD) by utilizing a comprehensive machine learning approach. Data from the MIMIC-IV and eICU-CRD databases were analyzed, incorporating 14 key clinical features identified through rigorous feature selection methods, including Boruta and LASSO regression. Eleven machine learning models were developed, with Gradient Boosting demonstrating the highest accuracy (78.94%) and optimal calibration characteristics. The external validation cohort revealed a decrease in model performance, emphasizing the risk of overfitting in complex models. Notably, the use of ACE inhibitors/ARBs was associated with a reduced risk of AKD progression, while nephrotoxic agents significantly increased this risk. Prognostic scoring systems, including SOFA and LODS, were found to correlate significantly with AKD outcomes, facilitating better risk stratification. Furthermore, a web-based risk calculator was developed to provide clinicians with an accessible tool for predicting the risk of SA-AKI progression to AKD based on individual patient data. In conclusion, this study underscores the importance of timely interventions and tailored treatment strategies in the management of SA-AKI, while also paving the way for future research to refine predictive models and improve clinical outcomes in critically ill patients.

Article activity feed