Clinical features and risk factors of plastic bronchitis associated with Mycoplasma pneumoniae pneumonia in children
10.19405/j.cnki.issn1000-1492.2026.08.020
- VernacularTitle:儿童肺炎支原体肺炎并发塑形性支气管炎的临床特征及危险因素分析
- Author:
Junli DING
1
;
Jiao LI
1
;
Jiantao ZHOU
1
;
Junwei GU
1
;
Chenyang XU
1
;
Yulin ZHU
1
;
Shu HOU
1
;
Shujuan WU
1
;
Shenggang DING
1
Author Information
1. Department of Pediatrics， The First Affiliated Hospital of Anhui Medical University， Hefei 230022
- Publication Type:Journal Article
- Keywords:
Mycoplasma pneumoniae pneumonia;
plastic bronchitis;
propensity score matching;
machine learning;
prediction model;
GBM model;
SHAP;
risk factors
- From:
Acta Universitatis Medicinalis Anhui
2026;61(8):1478-1485
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo investigate the risk factors for plastic bronchitis （PB） in children with Mycoplasma pneumoniae pneumonia （MPP） and to construct a predictive model. MethodsPropensity score matching （1∶4 nearest neighbor matching） was used to balance baseline data between the two groups. Univariate and multivariate Logistic regression analyses were performed to identify independent risk factors. Various machine learning algorithms were applied to build predictive models， and model performance was evaluated using receiver operating characteristic curves， decision curves， and calibration curves. SHapley Additive exPlanations （SHAP） method was used for model interpretation. ResultsAfter matching， baseline characteristics were comparable between the two groups. Compared with the non-PB group， the PB group had significantly longer fever duration， higher incidence of pleural effusion and right upper lobe involvement， and higher levels of procalcitonin （PCT）， D-dimer， lactate dehydrogenase （LDH）， alanine aminotransferase （ALT）， aspartate aminotransferase （AST）， immunoglobulin E （IgE）， and red blood cell count in bronchoalveolar lavage fluid （BALF） （all P<0.05）， while lymphocyte and eosinophil counts in blood were significantly lower （P<0.05）. Multivariate Logistic regression showed that PCT>0.207 μg/L， LDH>386.25 U/L， right upper lobe involvement， and red blood cell count in BALF>1 500×10⁶ were independent risk factors. Among machine learning models， the gradient boosting machine （GBM） model achieved an area under the curve of 0.786 in the validation set， demonstrating the best generalization ability. SHAP analysis validated the importance of the above variables. ConclusionThe four indicators above have independent predictive value for PB occurrence， and the GBM model shows certain clinical risk stratification capability.
