Machine learning-based prediction model for patients with recurrent Staphylococcus aureus bacteremia.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Infectious Disease Epidemiology

Background:

• Staphylococcus aureus bacteremia (SAB) presents significant challenges due to high recurrence and mortality rates.
• The incidence of SAB, particularly Methicillin-resistant Staphylococcus aureus (MRSA) infections, has increased post-COVID-19.
• Effective prediction models are crucial for timely clinical intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting recurrent SAB.
• To assist physicians in prompt patient assessment and proactive treatment strategies.
• To identify key clinical features associated with SAB recurrence.

Main Methods:

• Utilized the MIMIC-IV database (v2.2) with a 7:3 train-test split.
• Employed Recursive Feature Elimination (RFE) and LASSO for feature selection.
• Built and compared prediction models including XGBoost, Random Forest, LR, SVM, and ANN, validated using ROC, DCA, and PRC.
• Applied SHAP values for model interpretability and feature significance analysis.

Main Results:

• Key predictors identified include MRSA, PTT, RBC, RDW, Neutrophils_abs, Sodium, Calcium, Vancomycin concentration, MCHC, MCV, and Prognostic Nutritional Index (PNI).
• The XGBoost model exhibited the best performance with an AUC of 0.76 (ROC) and 0.56 (PRC).
• A website was developed based on the XGBoost model, with SHAP values explaining feature importance.

Conclusions:

• XGBoost is a suitable algorithm for developing medical prediction models.
• The developed prediction model for recurrent SAB aids physicians in timely diagnosis and treatment.
• The model enhances clinical decision-making for patients with SAB.