Using machine learning for accurate prediction: unlocking a new chapter in risk assessment for postoperative pneumonia in osteosarcoma

Abstract Objective This study aimed to develop and validate risk prediction models for postoperative pneumonia (POP) in osteosarcoma (OS) patients using multiple machine learning (ML) algorithms, ultimately selecting the model with optimal predictive performance for risk assessment. Methods A retrospective analysis was performed on the clinical data from 290 patients with OS who underwent surgical intervention. The dataset was preprocessed and feature-selected using statistical methods such as the T-test, Fisher's exact test, and Chi-square test. Subsequently, ML algorithms—Univariate Logistic Regression, Multivariate Logistic Regression, Least Absolute Shrinkage and Selection Operator (LASSO) regression, Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were employed to construct risk prediction models. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). The DeLong test was used to compare models and identify the optimal one. Results Among the evaluated models, XGBoost demonstrated superior performance with an area under the receiver operating characteristic curve (ROC-AUC) of 0.948, area under the precision–recall curve (PR-AUC) of 0.866, achieving an accuracy of 0.938, sensitivity of 0.933, and specificity of 0.939. The XGBoost model identified intraoperative blood loss, surgical duration, prophylactic antibiotic administration, age, targeted therapy, lymph node metastasis, and biomarkers (CA-125, CRP, FIB, Tn, SF, GGT) as significant predictors of postoperative pneumonia risk. A nomogram was developed based on these variables. POP was an independent predictor of poor overall survival (adjusted hazard ratio = 1.96, 95% confidence interval [CI]: 1.20–3.20; P = 0.008), with validated proportional hazards (global P = 0.428). Conclusion This study demonstrated that the XGBoost model incorporating 12 routinely available variables exhibited superior predictive performance for POP in OS patients. Additionally, POP was validated as an independent adverse prognostic factor for overall survival in this population.