Prediction Models for AKI in ICU: A Comparative Study.

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Acute kidney injury (AKI) is a common and serious complication in Intensive Care Units (ICUs).
• Early prediction of AKI is crucial for timely intervention and improved patient outcomes.
• Existing prediction models require rigorous evaluation with advanced machine learning techniques.

Purpose of the Study:

• To assess the performance of various machine learning models for the early prediction of AKI in ICU patients.
• To identify the most effective model for predicting AKI onset within the first 72 hours of ICU admission.

Main Methods:

• Utilized data from the Medical Information Mart for Intensive Care (MIMIC)-III database, including 17,205 adult patients.
• Excluded patients with pre-existing kidney disease; selected 17 predictor variables based on KDIGO guidelines and literature.
• Evaluated six models: Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Decision Machine (LightGBM), and Convolutional Neural Network (CNN).
• Performance metrics included Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, precision, recall, and F1-score.

Main Results:

• LightGBM achieved the highest performance across all metrics (average AUC = 0.905, F1 = 0.897, recall = 0.836).
• XGBoost showed the second-best performance, while LR, RF, and SVM had similar AUCs.
• The Convolutional Neural Network (CNN) model yielded the lowest scores for accuracy, precision, F1, and AUC.
• Serum creatinine levels significantly influenced AKI prediction in LR, RF, SVM, and LightGBM models.

Conclusions:

• LightGBM demonstrated superior capability for predicting AKI within the initial 72 hours of ICU stay.
• LightGBM and XGBoost exhibit significant potential for clinical integration due to their high recall rates.
• Findings offer valuable insights for developing AI-powered clinical decision support systems for early AKI detection in ICUs.