AI-SOFA

Accurate and timely mortality prediction is essential for nursing clinical decision-making in intensive care units (ICUs). Although the Sequential Organ Failure Assessment (SOFA) score is widely used to evaluate organ dysfunction, its manual calculation limits routine application in fast-paced clinical environments. This study aimed to enhance ICU system-level safety and workflow efficiency by refining and evaluating an automated Electronic Medical Record (EMR)-integrated SOFA scoring system (AI-SOFA) to evaluate: (1) its predictive performance for mortality compared to traditional manual scoring; and (2) its clinical utility as a nursing informatics initiative. A retrospective cohort study was conducted using EMR data from 2559 ICU admissions at a tertiary hospital in South Korea. Automated SOFA scores were generated using 11 routinely collected clinical parameters. Logistic regression, random forest, and XGBoost models were trained, and model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, accuracy, and F1 score. ICU mortality increased markedly with higher SOFA scores, exceeding 50% at scores ≥13. Among the machine learning (ML) models, XGBoost demonstrated the highest predictive performance (AUROC=0.9005), outperforming random forest (0.8975) and logistic regression (0.8722). In contrast, mortality prediction based on manual SOFA scoring showed substantially lower accuracy (AUROC=0.64). The AI-SOFA system serves as a nursing informatics tool that supports nursing workflows by enabling real-time risk stratification, reducing documentation burden, and facilitating timely clinical decision-making in ICU settings.