Development and validation of a CT radiomics–clinical machine-learning model for predicting the need for repeat bronchoalveolar lavage in children with mycoplasma pneumoniae pneumonia complicated by plastic bronchitis

Plastic bronchitis (PB) is a severe complication of Mycoplasma pneumoniae pneumonia (MPP) in children, often requiring repeated bronchoalveolar lavage (BAL). However, there is no uniform standard for whether a second BAL is needed. This study aimed to develop and validate a combined radiomics-clinical machine-learning model to predict the need for secondary BAL in children with MPP complicated with PB. This multi-center retrospective study included pediatric patients (≤ 14 years) with MPP and PB admitted to two tertiary hospitals in China from October 2019 to October 2024. Clinical variables and computed tomography (CT) radiomics features were extracted, preprocessed, and selected using LinearSVC. Significant clinical predictors were identified using univariate logistic regression. Six machine learning algorithms were trained and validated using independent datasets. Performance was assessed using the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, and decision curve analysis (DCA). Shapley Additive exPlanations (SHAP) were used for model interpretation. A total of 219 patients were analyzed (training: 104; internal validation: 27; external testing: 88). Lactate dehydrogenase (LDH), C-reactive protein (CRP), and D-dimer levels were significantly associated with secondary BAL. The combined ExtraTrees_gini model achieved the most stable performance (AUROC: 0.889, 0.842, and 0.826 across datasets). DCA showed consistently higher net clinical benefit than radiomics-only or clinical-only models. SHAP analysis identified wavelet_HHL_glszm_GrayLevelVariance, LDH, and CRP levels as the most influential features, with elevated LDH and CRP levels increasing the predicted probability of secondary BAL, and texture features showing variable directional effects. The findings showed that the combined model could be used to predict the risk of secondary BAL in MPP patients complicated with PB. These findings could significantly enhance clinical decision-making and improve personalized management strategies.