Application of Machine Learning for Patients With Cardiac Arrest: Systematic Review and Meta-Analysis

Area of Science:

• Cardiology and Artificial Intelligence
• Clinical Prediction Modeling
• Health Informatics

Background:

• Current early assessment tools for cardiac arrest (CA) are insufficient.
• Machine learning (ML) models show potential for predicting CA occurrence and prognosis.
• Systematic evidence is needed to validate ML model efficacy in CA prediction.

Purpose of the Study:

• To evaluate the predictive value of ML for CA occurrence, neurological prognosis, mortality, and return of spontaneous circulation (ROSC).
• To provide evidence for developing and refining clinical CA prediction tools.
• To assess ML model performance using meta-analysis.

Main Methods:

• Systematic literature search of PubMed, Embase, Cochrane Library, and Web of Science until May 17, 2024.
• Included 93 studies with over 5.7 million patients (in-hospital and out-of-hospital).
• Assessed risk of bias using the Prediction Model Risk of Bias Assessment Tool.

Main Results:

• ML models demonstrated high predictive accuracy for CA occurrence (C-index 0.90), good neurological prognosis (C-index 0.86), and mortality (C-index 0.85).
• Key predictors for CA included respiratory rate, blood pressure, age, and temperature.
• Significant predictors for neurological outcomes varied between in-hospital and out-of-hospital CA, including rhythm, age, and medication use.

Conclusions:

• ML is a promising approach for predicting CA and its outcomes.
• ML models show superior performance compared to traditional scoring tools for specific CA outcomes.
• Future research should focus on integrating ML for AI-driven enhancements to traditional CA scoring tools.