Critically ill patients transported to the emergency department by ambulance represent a high-risk and clinically heterogeneous population. Early identification of patients requiring hospitalization is essential for effective resource allocation and timely clinical decision-making. This study focused on clinical and laboratory parameters routinely obtained during the early emergency department course. This retrospective single-center cohort study included 2,338 adult patients brought to the emergency department by ambulance. Demographic, clinical, and early laboratory parameters obtained shortly after emergency department arrival were used to develop machine learning models predicting hospitalization. Logistic regression, random forest, and gradient boosting algorithms were trained and evaluated using five-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (ROC AUC), accuracy, sensitivity, and specificity. Among the included patients, 37.0% required hospitalization. The random forest model demonstrated the highest predictive performance, with a cross-validation ROC AUC of 0.850 and a test ROC AUC of 0.776. Independent predictors of hospitalization included troponin level, altered mental status, lactate level, age, creatinine, leukocyte count, and pH. Machine learning models, particularly random forest, demonstrated clinically meaningful discriminative performance in predicting hospitalization among critically ill emergency department patients transported by ambulance using early clinical and laboratory data.
Durmuş et al. (Fri,) studied this question.