AI-Based Predictive Models for Cardiogenic Shock in STEMI: Real-World Data for Early Risk Assessment and Prognostic...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Cardiogenic shock (CS) is a severe complication of ST-elevation myocardial infarction (STEMI), leading to high in-hospital mortality.
• Early identification and intervention are crucial for improving patient outcomes in STEMI.
• Current reperfusion strategies have not significantly reduced CS-related mortality.

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) models in predicting the risk of CS during early care phases (prehospital, ED, cardiology-on-call).
• To assess the utility of ML for accurate triage and prioritization of STEMI patients requiring urgent angiography.
• To identify key clinical features that predict CS risk in STEMI patients.

Main Methods:

• Development and evaluation of various ML models, including Extra Trees, Support Vector Machine, and Random Forest classifiers.
• Assessment of model performance using metrics such as accuracy, precision, recall, F1-score, and MCC across different care phases.
• Identification of critical predictive features from routinely available clinical data.

Main Results:

• Extra Trees classifier showed high performance in the prehospital phase (ACC 0.9062).
• Support Vector Machine (ACC 78.12%) and Random Forest (ACC 81.25%) demonstrated strong predictive capabilities in the ED and cardiology-on-call phases, respectively.
• Killip class, ECG rhythm, creatinine, potassium, and renal dysfunction markers were key predictors; models showed greatest utility in prehospital and ED settings.

Conclusions:

• ML-based predictive models are valuable tools for early risk stratification of STEMI patients at risk for CS.
• Implementation of ML-driven tools can enhance decision-making in early STEMI care pathways.
• These tools have the potential to improve survival rates through faster and more accurate patient management, particularly in time-sensitive environments.