Machine Learning To Stratify Methicillin-Resistant Staphylococcus aureus Risk among Hospitalized Patients with...

Area of Science:

• Infectious Diseases
• Medical Informatics
• Pulmonology

Background:

• Methicillin-resistant Staphylococcus aureus (MRSA) is a serious cause of community-acquired pneumonia (CAP).
• Lack of validated risk factors for MRSA CAP leads to overuse of broad-spectrum antibiotics.
• Accurate risk prediction models are needed to guide antibiotic stewardship.

Purpose of the Study:

• To develop and validate machine learning models for predicting MRSA CAP risk.
• To identify key clinical factors associated with MRSA CAP.
• To improve empirical antibiotic prescribing for CAP.

Main Methods:

• Utilized a population-based sample of hospitalized CAP patients from academic and community teaching hospitals.
• Employed machine learning, specifically Classification Tree Analysis, for model development.
• Defined cases as CAP with MRSA isolated within 72 hours of admission; controls lacked MRSA CAP.

Main Results:

• The machine learning model achieved high accuracy (ROC area = 0.775).
• MRSA CAP risk was higher in ICU patients requiring mechanical ventilation (OR, 8.3).
• Risk was lower in ward patients (OR, 0.21) and ICU patients without recent antibiotic use (OR, 0.03).

Conclusions:

• A simple, accurate machine learning model can predict MRSA CAP risk within 72 hours postadmission.
• Model facilitates targeted antibiotic use, potentially reducing unnecessary broad-spectrum prescriptions.
• Identified key predictors like ICU admission, mechanical ventilation, and recent antibiotic history.