Abstract
PURPOSE: To evaluate the postoperative pneumonia (POP) risk of patients with non-small cell lung cancer (NSCLC), identify influencing factors, develop a LASSO regression-based model to predict POP risk and identify critical influencing factors. METHODS: This retrospective analysis included patients with NSCLC who underwent surgery at our hospital from 2021 to 2024. Potential predictors spanning demographics, comorbidities, and preoperative biomarkers were evaluated. LASSO regression screened variables, followed by logistic regression to construct the model. Model performance was assessed via Area Under the Curve (AUC), Sensitivity (SEN), Specificity (SPE), Accuracy, Positive predictive value (PPV), F1-score, calibration curves, and decision curve analysis (DCA). RESULTS: A total of 457 patients were included, with 64 (14%) developing POP. Patients were randomly allocated in a 7:3 ratio to training (n=323) and validation (n=134) cohorts. The model demonstrated acceptable yet moderate discriminatory power, with an AUC of 0.832 in the validation set. However, it exhibited high specificity (0.941) at the cost of low sensitivity (0.438), indicating a limitation in identifying all POP cases. The validation accuracy was 0.881. A nomogram was developed to visualize the model for clinical use. CONCLUSION: The developed model shows potential for identifying a subset of patients at very high risk of POP due to its high specificity. However, its clinical utility is currently limited by its low sensitivity and is threshold-dependent. It may serve as a component of a broader risk assessment strategy rather than a stand-alone clinical screening tool.