Predicting mortality among septic patients presenting to the emergency department-a cross sectional analysis using...

Area of Science:

• Emergency Medicine
• Data Science
• Critical Care

Background:

• Sepsis is a leading cause of global mortality, responsible for nearly 20% of all deaths worldwide.
• Early identification of sepsis in the emergency department (ED) is critical for timely intervention and improved patient survival.

Purpose of the Study:

• To identify variables predictive of 7- and 30-day mortality in emergency department sepsis patients.
• To evaluate the utility of machine learning models for sepsis mortality prediction.

Main Methods:

• Retrospective analysis of 445 sepsis patients presenting to the ED.
• Utilized a Balanced Random Forest Classifier with 91 variables reflecting patient presentation.
• Performed 10-fold cross-validation to assess model accuracy using AUC, sensitivity, and specificity.

Main Results:

• Six variables accurately predicted 7-day mortality: fever, abnormal verbal response, low saturation, EMS arrival, abnormal consciousness, and chills (AUC=0.83).
• Six variables also predicted 30-day mortality, substituting breathing difficulties for abnormal consciousness (AUC=0.80).
• The models demonstrated good accuracy in predicting both short-term and 30-day mortality.

Conclusions:

• Specific clinical signs, symptoms, and mode of arrival are strong predictors of sepsis mortality.
• These variables, alongside vital signs, can inform the development of future sepsis mortality prediction tools.
• Random Forests are a suitable machine learning method for sepsis-related research.