Identifying Risk Factors and Creating a Point-Based Risk Calculator for Postoperative Pneumonia in Thoracic Surgery Patients.

This secondary data analysis used the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP), logistic regression (Method 1), Xtreme Gradient Boosting (Method 2), and a 12-member expert panel (Method 3) to develop and validate a predictive model to identify patients undergoing thoracic surgery at risk for postoperative pneumonia (POP). Twenty-three covariates associated with POP were selected from the 2013-2022 ACS NSQIP dataset filtered for thoracic surgeries. Method 1 and Method 2 were assessed through area under the receiver operating characteristic curve (AUC ROC) using 10-fold cross-validation. Method 3 evaluated the 23 covariates for relevance to POP and relevant predictors were assessed through AUC ROC. Method 1 identified nine significant predictors (P < .05) with a 10-fold cross-validated AUC ROC = .72 (fair classifier). The significant preoperative predictors and their effect size were, sepsis (1.43), systemic inflammatory response syndrome (1.04), male gender (.77), bleeding disorder (.57), current smoker within 1 year (0.39), disseminated cancer (.39), hypoalbuminemia (.33), history of severe chronic obstructive pulmonary disease (.31), and anemia (.05). Method 2 achieved a 10-fold cross-validation AUC ROC = .75 (fair classifier). Method 3 had an AUC ROC = .6 (poor classifier). The nine significant predictors from Method 1 were used to develop a risk-based calculator.