天津医药

天津医药 ›› 2026, Vol. 54 ›› Issue (4): 369-373.doi: 10.11958/20251913

• 临床研究 • 上一篇    下一篇

基于BP神经网络构建儿童肺炎支原体混合腺病毒感染的重症肺炎预测模型

姚国华1(), 刘杰1, 张雯1, 马翠安1△(), 魏博涛1, 高娜2   

  1. 1 天津市儿童医院， 天津大学儿童医院感染科（邮编300134）
    2 天津职业技术师范大学自动化与电气工程学院
  • 收稿日期:2025-05-09 修回日期:2025-11-21 出版日期:2026-04-15 发布日期:2026-04-14
  • 通讯作者: △E-mail：macuian@126.com
  • 作者简介:姚国华（1982），女，副主任医师，主要从事儿童感染性疾病诊治方面研究。E-mail：moyan1111@163.com

The construction of a prediction model for severe pneumonia caused by mycoplasma pneumoniae mixed with adenovirus infection in children based on BP neural network

YAO Guohua1(), LIU Jie1, ZHANG Wen1, MA Cuian1△(), WEI Botao1, GAO Na2   

  1. 1 Department of Infectious Diseases, Tianjin Children’s Hospital, Tianjin University, Tianjin 300134, China
    2 School of Automation and Electrical Engineering, Tianjin University of Technology and Education
  • Received:2025-05-09 Revised:2025-11-21 Published:2026-04-15 Online:2026-04-14
  • Contact: △E-mail：macuian@126.com

RichHTML

0

PDF (PC)

8

摘要：

目的 基于反向传播法（BP）神经网络构建儿童肺炎支原体（MP）混合腺病毒（ADV）感染的重症肺炎的临床预测模型。方法 回顾性分析138例MP混合ADV感染的社区获得性肺炎患儿的临床、实验室及影像学资料，按7∶3将研究对象随机分为训练集（96例）和测试集（42例），构建BP神经网络预测模型。训练集用沙普利加法解释量化临床特征贡献度，筛选出MP混合ADV的重症肺炎的预测因子。通过测试集的准确率、损失值、混淆矩阵对其进行验证。结果 重症组发热持续天数、最高体温、中性粒细胞百分比（N%）、天冬氨酸转氨酶（AST）、乳酸脱氢酶（LDH）、白细胞介素-6（IL-6）、大片炎性实变、住院天数高于非重症组，淋巴细胞百分比（L%）、白蛋白低于非重症组（P<0.05）。基于BP神经网络研究的结果显示发热持续天数、AST、N%、最高体温、大片炎性实变、IL-6、L%、LDH是MP混合ADV感染所致重症肺炎的关键预测因子。在构建儿童重症MP混合ADV临床预测模型上，测试集显示准确率90.48%、损失值0.233 2。结论 基于BP神经网络成功构建的儿童MP混合ADV感染重症肺炎的预测模型筛选出8项关键预测因子，可为临床早期识别重症病例提供参考。

关键词: 肺炎, 支原体, 腺病毒, 人, 同时感染, 模型, 统计学, 儿童, BP神经网络, 沙普利加法解释

Abstract:

Objective To construct a clinical prediction model for severe pneumonia caused by mycoplasma pneumoniae (MP) and adenovirus (ADV) in children based on the backpropagation method (BP) neural network. Methods A retrospective analysis was conducted on the clinical, laboratory and imaging data of 138 children with severe pneumonia caused by MP mixed with ADV infection. The research subjects were randomly divided into the training set and the test set (7:3), and a BP neural network prediction model was constructed. The contribution of clinical features in the training set was quantified by shapley additive explanations (SHAP). The predictors of severe pneumonia for MP mixed with ADV were screened out. It is verified through the accuracy rate, loss value and confusion matrix in the test set. Results In the severe group, the duration of fever, the highest body temperature, neutrophil percentage (N%), aspartate transaminase (AST), lactate dehydrogenase (LDH), interleukin-6 (IL-6), extensive inflammatory consolidation and length of hospital stay were higher than those in the non-severe group, while lymphocyte percentage (L%) and albumin levels were lower than those in the non-severe group（P<0.05）. Further research results through BP neural network showed that the duration of fever, AST, N%, maximum body temperature, large areas of inflammatory consolidation, IL-6, L% and LDH were the key predictors of severe pneumonia caused by MP and ADV infection. In constructing the clinical prediction model of severe MP mixed with ADV in children, the test set showed an accuracy rate of 90.48% and a loss value of 0.233 2. Conclusion The prediction model for severe pneumonia caused by mixed MP-ADV infections is successfully constructed in children using a BP neural network. The eight key predictors are identified from the model that can serve as a reference for early clinical identification of severe cases.

Key words: pneumonia, mycoplasma, adenoviruses, human, coinfection, models, statistical, child, BP neural network, shapley additive explanations

中图分类号: