An Ensemble Machine Learning Model for Early Prediction of Vancomycin-Induced Acute Kidney Injury in ICU Patients

Area of Science:

• Nephrology
• Artificial Intelligence
• Clinical Informatics

Background:

• Vancomycin treatment can cause Acute Kidney Injury (AKI), a serious complication with limited predictive research.
• Early detection of vancomycin-induced AKI is crucial for patient outcomes.

Purpose of the Study:

• To develop and validate a stacking ensemble machine learning model for predicting AKI in vancomycin-treated patients.
• To identify key clinical predictors of AKI in this high-risk population.

Main Methods:

• A stacking ensemble machine learning model was developed using data from 314 ICU patients.
• SHapley Additive exPlanations (SHAP) were employed for model interpretability.
• Key predictors like serum creatinine, glucose variability, and age were identified.

Main Results:

• The model achieved 92% accuracy, 93% precision, 92% sensitivity, and an AUC of 0.94.
• Serum creatinine, glucose variability, and patient age were identified as critical predictors of AKI.
• The model outperformed existing predictive approaches.

Conclusions:

• Machine learning and interpretable AI offer a scalable, cost-effective solution for early AKI detection.
• The developed model facilitates proactive clinical decision-making to reduce AKI risk.
• Integrating AI into healthcare settings can improve patient outcomes in vancomycin therapy.