Construction of a machine learning model based on serum indicators and lung ultrasound features for the prediction of respiratory failure in children with severe pneumonia
10.3760/cma.j.cn101070-20240822-00527
- VernacularTitle:基于血清指标和肺部超声特征构建机器学习模型预测小儿重症肺炎并呼吸衰竭
- Author:
Fan ZHOU
1
;
Xueping NIE
1
Author Information
1. 九江市妇幼保健院儿童重症医学科，九江 332000
- Publication Type:Journal Article
- Keywords:
Child;
Respiratory failure;
Serum indicator;
Ultrasonic features of lung;
Machine learning model;
Severe pneumonia
- From:
Chinese Journal of Applied Clinical Pediatrics
2025;40(3):194-200
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a machine learning model based on serum indicators and lung ultrasound features for the prediction of respiratory failure in children with severe pneumonia.Methods:Case-control study.The clinical data of 208 children with severe pneumonia admitted to the Pediatric Intensive Care Unit of Jiujiang Maternal and Child Health Hospital from January 2019 to January 2024 were analyzed.According to whether respiratory failure occurred 3 months after admission, the patients were divided into a respiratory failure group (131 cases) and a non-respiratory failure group (77 cases).Logistic regression, classification regression tree (CRT) and back propagation neural network (BPNN) algorithms based on machine learning were used to construct the prediction models of respiratory failure in children with severe pneumonia by SPSS software.The predictive value of models was analyzed by the receiver operating characteristic (ROC) curve.Results:The univariate analysis showed that there were significant differences in age, history of pulmonary infection, pediatric critical illness score (PCIS), congenital airway malformation, congenital heart disease, lung ultrasound score (LUS), Krebs von den Lungen-6 (KL-6) and soluble triggering receptor expressed on myeloid cells 1 (sTREM-1) between the two groups (all P<0.05).The multivariate Logistic regression analysis showed that age, PCIS score, LUS, congenital heart disease, KL-6 and sTREM-1 were independent risk factors for respiratory failure in children with severe pneumonia (all P<0.05).The analysis results of the CRT model showed that age, congenital heart disease, KL-6 and congenital airway malformation were factors influencing the development of respiratory failure in children with severe pneumonia.The analysis results of the BPNN model indicated that KL-6 was the most important factor that affected the occurrence of respiratory failure in severe pneumonia, followed by age, sTREM-1, LUS, PCIS score, congenital heart disease, congenital airway malformation, and history of pulmonary infection successively.Among the models constructed by the three machine learning algorithms, the Logistic model had the best prediction performance, with an area under the ROC curve of 0.981, a sensitivity of 0.962, and a specificity of 0.909. Conclusions:The risk factors of respiratory failure in children with severe pneumonia include age, LUS, KL-6, sTREM-1, etc.Machine learning models based on serum indicators and lung ultrasound features are effective in predicting respiratory failure in children with severe pneumonia, especially the Logistic model, which has the best prediction efficiency.
