DATASET AI

This study evaluates the predictive performance of 11 machine learning (ML) classifiers on clinical data from 158 STEMI-CS patients treated between 2019 and 2022 at the Cardiology Department, University Emergency Hospital, Bucharest, Romania. The analysis is structured across five distinct critical phases in the patient's evaluation: the prehospital phase, emergency department phase, cardiology consultation phase in the emergency department, interventional cardiologist phase, and cardiac intensive care unit (CICU/ICI) phase. The predictive performance was assessed using several key performance metrics such as Accuracy, Precision, Recall, F1-score and Matthews Correlation Coefficient, along with confusion matrices.The analysis highlights that model performance varies depending on the specific characteristics of each dataset. The Extra Trees (ET) classifier achieved the highest accuracy (90.62%) in the prehospital phase, while Support Vector Machine (SVM) performed best in the emergency department setting. Random Forest (RF) consistently delivered strong results in both the cardiology (cardiology consultation and interventional cardiologist) phase and ICU phases, with accuracies of 81.25%, 87.5% and 84.37%, respectively. At the same time, Quadratic Discriminant Analysis (QDA) demonstrated the most stable and reliable performance across all five clinical phases, indicating a high capacity for generalization. Clinically relevant predictors included Killip classification at presentation, ECG rhythm, CKI, creatinine, potassium, reperfusion type, and pre-PCI TIMI score. The greatest clinical utility was found in prehospital and emergency settings, allowing patient prioritization for angiography, essential in centers with limited resources. In the cardiac intensive care unit (CICU), models effectively identified patients at increased risk of post-reperfusion complications, optimizing therapeutic decisions.