Interpretable machine learning model for predicting acute kidney injury in critically ill patients

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Critical Care Medicine

Background:

• Acute Kidney Injury (AKI) poses a significant threat to intensive care unit (ICU) patients.
• Prompt prediction of AKI is crucial for timely intervention and improved patient outcomes.
• Existing prediction methods may lack interpretability, hindering clinical adoption.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for the early prediction of AKI in ICU patients.
• To apply explainable AI techniques to understand the factors contributing to AKI prediction.
• To provide a tool that assists physicians in identifying high-risk patients for proactive management.

Main Methods:

• Utilized data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (2008-2019).
• Developed six ML models, including eXtreme Gradient Boosting (XGBoost), to predict AKI.
• Employed Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Included 53,150 severely ill patients, with 80% in the training set and 20% in the validation set.
• The XGBoost model demonstrated superior AKI prediction performance with an Area Under the Curve (AUC) of 0.816.
• Key predictors identified by XGBoost include SOFA score, weight, mechanical ventilation, and SAPS II.

Conclusions:

• Clinical feature-based ML models, particularly XGBoost, show high accuracy in predicting AKI in ICU settings.
• Interpretable AI facilitates understanding of AKI risk factors, supporting clinical decision-making.
• Early identification of AKI risk enables timely interventions, potentially improving patient prognosis.