Early sepsis prediction in the emergency department using machine learning

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Early sepsis recognition is vital for timely intervention and reducing organ injury or death.
• Current clinical scores (qSOFA, NEWS) have limited predictive accuracy for sepsis-related outcomes.
• Existing scores require manual input and cannot integrate diverse data sources.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting organ injury and 30-day mortality in early sepsis.
• To compare the performance of ML models against conventional scoring systems (qSOFA, NEWS).
• To assess the utility of ML models in improving risk stratification for sepsis patients.

Main Methods:

• A post-hoc analysis of prospectively collected data from 1431 ED patients with suspected sepsis.
• Extraction of 149 routinely collected clinical parameters from electronic health records and bedside measurements.
• Development of ML models using 10-fold cross-validation to predict organ injury and 30-day mortality.

Main Results:

• ML models achieved an AUROC of 0.798 for organ injury and 0.808 for 30-day mortality.
• Performance significantly outperformed qSOFA (AUROC 0.557-0.577) and NEWS (AUROC 0.576-0.610).
• Key contributing parameters included serum creatinine, SOFA score, ISAR-HP score, and clinical impression.

Conclusions:

• Machine learning models demonstrate superior predictive performance for organ injury and mortality in early sepsis.
• Routinely available Emergency Department data can be effectively utilized by ML models.
• ML models show potential as valuable clinical decision support tools for sepsis management.