Scholarlink

Background <p><i>Mycoplasma pneumoniae</i>&#xa0;(MP) is a major pathogen of community-acquired pneumonia (CAP) in children. Severe&#xa0;<i>Mycoplasma pneumoniae</i>&#xa0;pneumonia (SMPP) poses a significant threat to pediatric health. Current diagnostic approaches rely primarily on imaging and clinical signs and lack objective and quantitative tools for early risk prediction. This study aimed to develop an explainable machine learning (ML) model for the early prediction of SMPP risk in children.</p> Methods <p>A retrospective analysis was conducted on data from 507 pediatric inpatients with&#xa0;<i>M. pneumoniae</i>&#xa0;pneumonia (MPP) admitted to the Affiliated Hospital of Yan’an University between January 2023 and December 2024. Clinical characteristics were compared between the MPP and SMPP groups. Key predictors were identified using the Mann–Whitney <i>U</i> test, chi-square test, Pearson correlation analysis, and least absolute shrinkage and selection operator (LASSO) regression. Seven ML models were constructed. Hyperparameters were optimized through fivefold cross-validation and grid search. Model performance was evaluated using the following metrics: accuracy, precision, recall, F1 score, and the area under the curve (AUC). Model interpretability was analyzed using the SHapley Additive exPlanations (SHAP) method.</p> Results <p>Following feature selection, 21 variables were retained. Seven ML models were constructed and evaluated for their predictive efficacy. Among them, the Categorical Boosting (CatBoost) model demonstrated optimal performance, achieving an AUC of 0.882 and an accuracy of 0.850. Decision curve analysis (DCA) was employed to assess the clinical utility of these models. Within the threshold probability range of 0.3 to 0.8 in the test set, the CatBoost model exhibited a superior net benefit compared to the other ML models. In the external validation, the model achieved an AUC of 0.769. Furthermore, the SHAP method was applied to interpret the CatBoost model. The SHAP analysis identified the following five key features as contributing most to the model’s predictions, listed in order: peak temperature, duration of fever, lactate dehydrogenase (LDH), platelet count (PLT), and D-dimer.</p> Conclusions <p>This study developed an explainable machine learning model based on the CatBoost algorithm, providing a potential quantitative tool for the early risk warning of SMPP in children. The model demonstrated favorable discriminatory performance and a certain degree of generalizability within the retrospective cohort. Furthermore, interpretability analysis revealed key clinical risk factors, including peak temperature, duration of fever, LDH, PLT, and D-dimer, among which peak temperature and fever duration contributed the most to the predictions. These findings provide a transparent and quantitative basis for early clinical identification and decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of explainable machine learning in early diagnosis models for risk prediction of severe Mycoplasma pneumoniae pneumonia in children

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月7日

查看全文
  • Shuai Yu,
  • Yaya Ren,
  • Jiangang Song,
  • Yuxin Zhu,
  • Hua Jiang,
  • Yuanxia Li

摘要

Background

Mycoplasma pneumoniae (MP) is a major pathogen of community-acquired pneumonia (CAP) in children. Severe Mycoplasma pneumoniae pneumonia (SMPP) poses a significant threat to pediatric health. Current diagnostic approaches rely primarily on imaging and clinical signs and lack objective and quantitative tools for early risk prediction. This study aimed to develop an explainable machine learning (ML) model for the early prediction of SMPP risk in children.

Methods

A retrospective analysis was conducted on data from 507 pediatric inpatients with M. pneumoniae pneumonia (MPP) admitted to the Affiliated Hospital of Yan’an University between January 2023 and December 2024. Clinical characteristics were compared between the MPP and SMPP groups. Key predictors were identified using the Mann–Whitney U test, chi-square test, Pearson correlation analysis, and least absolute shrinkage and selection operator (LASSO) regression. Seven ML models were constructed. Hyperparameters were optimized through fivefold cross-validation and grid search. Model performance was evaluated using the following metrics: accuracy, precision, recall, F1 score, and the area under the curve (AUC). Model interpretability was analyzed using the SHapley Additive exPlanations (SHAP) method.

Results

Following feature selection, 21 variables were retained. Seven ML models were constructed and evaluated for their predictive efficacy. Among them, the Categorical Boosting (CatBoost) model demonstrated optimal performance, achieving an AUC of 0.882 and an accuracy of 0.850. Decision curve analysis (DCA) was employed to assess the clinical utility of these models. Within the threshold probability range of 0.3 to 0.8 in the test set, the CatBoost model exhibited a superior net benefit compared to the other ML models. In the external validation, the model achieved an AUC of 0.769. Furthermore, the SHAP method was applied to interpret the CatBoost model. The SHAP analysis identified the following five key features as contributing most to the model’s predictions, listed in order: peak temperature, duration of fever, lactate dehydrogenase (LDH), platelet count (PLT), and D-dimer.

Conclusions

This study developed an explainable machine learning model based on the CatBoost algorithm, providing a potential quantitative tool for the early risk warning of SMPP in children. The model demonstrated favorable discriminatory performance and a certain degree of generalizability within the retrospective cohort. Furthermore, interpretability analysis revealed key clinical risk factors, including peak temperature, duration of fever, LDH, PLT, and D-dimer, among which peak temperature and fever duration contributed the most to the predictions. These findings provide a transparent and quantitative basis for early clinical identification and decision-making.