Abstract
OBJECTIVE: To identify key risk factors for postoperative pulmonary infections (PPIs) in lung cancer (LC), patients undergoing radical surgery and construct a multiparametric nomogram model to improve PPI risk prediction accuracy, guiding individualized interventions. METHODS: A retrospective analysis was conducted on LC patients treated at Yidu Central Hospital of Weifang from March 2020 to May 2023. Among the 1,084 LC cases reviewed, patients were divided into an infected group (n = 131) and an uninfected group (n = 953) based on infection status. Key factors for PPIs were screened using machine learning techniques, including least absolute shrinkage and selection operator (LASSO) regression, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). A nomogram prediction model was developed, and its stability and clinical utility were evaluated using calibration curves and decision curve analysis, with internal validation through random case selection. RESULTS: Thirteen factors - including tumor stage, diabetes history, chronic obstructive pulmonary disease (COPD), operation duration, mechanical ventilation duration, age, C-reactive protein, procalcitonin, high-mobility group box 1, interleukin-6, neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, and systemic immune-inflammation index - were identified as significantly associated with PPIs. The nomogram model demonstrated high predictive accuracy in internal validation (C-index = 0.935), strong calibration, and substantial clinical benefit. For two randomly selected cases, the model predicted a 63% infection probability for the infected patient and a 32% probability for the uninfected patient, affirming the model's predictive effectiveness. CONCLUSIONS: The multiparametric nomogram model developed in this study provides a reliable method for PPI risk prediction in LC patients, supporting clinical decision-making and improving postoperative management.