Predicting Bacteremia among Septic Patients Based on ED Information by Machine Learning Methods: A Comparative Study.

Area of Science:

• Infectious Diseases
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Bacteremia is a critical, life-threatening infection lacking clear risk assessment guidelines.
• Predictive models are needed to determine bacteremia risk and blood culture urgency in emergency departments.

Purpose of the Study:

• To develop and evaluate predictive models for bacteremia in septic patients using emergency department big data.
• To compare the performance of logistic regression, support vector machine (SVM), and random forest (RF) models.

Main Methods:

• Retrospective cohort study of adult patients with systemic inflammatory response syndrome receiving blood cultures.
• Development of four models: two based on logistic regression (LR) and two incorporating machine learning (SVM, RF).
• Evaluation using area under the receiver operating curve (AUROC) and Akaike information criterion (AIC).

Main Results:

• The study included 40,395 patients, with 7.7% diagnosed with bacteremia.
• Machine learning models (SVM, RF) demonstrated performance comparable to logistic regression (AUROC ~0.73).
• No statistical evidence indicated logistic regression superiority over SVM or RF models.

Conclusions:

• Developed models effectively discriminate bacteremia risk in septic patients.
• SVM and RF models offer flexibility beyond linear relationships, aiding clinical decision-making.
• These machine learning approaches can help identify high-risk patients and optimize antibiotic use.