External validation of models for predicting pneumonia after cardiac surgery

Area of Science:

• Medical research
• Infectious disease epidemiology
• Cardiothoracic surgery

Background:

• Post-cardiac surgery infections, particularly pneumonia, affect approximately 20% of patients and significantly increase mortality risk.
• Identifying high-risk patient groups is crucial for effective infection prevention strategies.
• External validation of prediction models is essential for reliable identification of at-risk populations.

Purpose of the Study:

• To externally validate and compare the performance of two established prediction models for post-cardiac surgery pneumonia: Classification and Regression Tree (CART) and Logistic Regression Model (LRM).

Main Methods:

• Prospective analysis of 527 adult cardiac surgery patients to identify pneumonia predictors.
• Development and application of CART and LRM using a derivation set.
• Validation of models using a separate set of 333 adult cardiac surgery patients.
• Comparison of model performance using global, discrimination, and calibration measures.

Main Results:

• The LRM identified emergency surgery, COPD, ventricular dysfunction, and age as independent pneumonia predictors.
• The CART model identified emergency surgery, age, unstable angina, BMI, COPD, weight, and ventricular dysfunction.
• Emergency surgery emerged as the strongest predictor in both models.
• The LRM demonstrated superior performance across global, discrimination, and calibration metrics compared to CART.

Conclusions:

• The LRM exhibited superior predictive performance for pneumonia after cardiac surgery.
• While CART offers graphical interpretability, the LRM provides a more robust tool for risk adjustment and efficacy evaluation in clinical practice.
• The LRM is recommended for infection control practitioners for risk stratification and performance monitoring.