AI-SOFA: An EMR-integrated Nursing Informatics-driven Decision Support System for Mortality Risk-informed ICU...

Area of Science:

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

Background:

• Accurate mortality prediction is crucial for intensive care unit (ICU) clinical decisions.
• Manual Sequential Organ Failure Assessment (SOFA) scoring is time-consuming and limits routine use in fast-paced ICUs.

Purpose of the Study:

• To refine and evaluate an automated Electronic Medical Record (EMR)-integrated SOFA scoring system (AI-SOFA).
• To compare the mortality prediction performance of AI-SOFA against traditional manual SOFA scoring.
• To assess the clinical utility of AI-SOFA as a nursing informatics initiative.

Main Methods:

• Retrospective cohort study using EMR data from 2559 ICU admissions.
• Automated SOFA scores generated from 11 clinical parameters.
• Machine learning models (logistic regression, random forest, XGBoost) trained and evaluated using AUROC, sensitivity, specificity, accuracy, and F1 score.

Main Results:

• ICU mortality increased significantly with higher SOFA scores (over 50% at scores ≥13).
• XGBoost model achieved the highest predictive performance (AUROC=0.9005), outperforming other ML models.
• AI-SOFA demonstrated substantially higher accuracy (AUROC=0.64) than manual SOFA scoring.

Conclusions:

• The AI-SOFA system offers superior mortality prediction compared to manual SOFA scoring.
• AI-SOFA functions as a valuable nursing informatics tool, supporting real-time risk stratification and reducing documentation burden.
• This automated system facilitates timely clinical decision-making, enhancing safety and workflow efficiency in ICUs.