Machine learning-based scoring system to predict cardiogenic shock in acute coronary syndrome

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Cardiogenic shock (CS) is a life-threatening complication of acute coronary syndrome (ACS), associated with nearly 50% mortality.
• Early identification of patients at high risk for CS is crucial for implementing timely, life-saving interventions.
• Current risk stratification methods may not fully capture the complexity of CS development in ACS patients.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based scoring system to predict the risk of developing CS during hospitalization in ACS patients.
• To create a simple tool utilizing variables readily available at first medical contact.
• To improve early risk assessment for CS in the context of ACS.

Main Methods:

• An observational, multicenter study involving large derivation and external validation cohorts of ACS patients.
• Development of nine ML models using 13 readily available clinical variables.
• Logistic regression with elastic net regularization was selected for its superior predictive performance.

Main Results:

• The logistic regression model demonstrated the highest externally validated predictive performance, with a c-statistic of 0.844 (95% CI, 0.841-0.847).
• The STOP SHOCK score, derived from this model, effectively predicts CS development in ACS patients.
• The tool incorporates key variables such as heart rate, blood pressure, and oxygen saturation.

Conclusions:

• The STOP SHOCK score is a validated, machine learning-based tool for predicting CS risk in ACS patients at first medical contact.
• This accessible scoring system can facilitate preemptive strategies to prevent CS and improve patient outcomes.
• A web application for the STOP SHOCK score is available for clinical use.