Scholarlink

Objective <p>The incidence of&#xa0;Mycoplasma pneumoniae&#xa0;pneumonia (MPP) in children has increased in recent years, accompanied by a trend toward more severe clinical presentations. Early and accurate diagnosis is essential for timely targeted therapy and improved prognosis. This study aimed to develop and validate a machine learning-based diagnostic model integrating clinical characteristics, laboratory indicators, and chest CT findings for the early prediction of MPP in children.</p> Methods <p>A total of 449 children diagnosed with community-acquired pneumonia (CAP) who were hospitalized at Ordos Central Hospital between January 2019 and April 2025 were retrospectively enrolled, including 311 MPP cases and 138 non-MPP cases. The training set included 314 patients (217 MPP and 97 non-MPP), while the test set included 135 patients (94 MPP and 41 non-MPP). Clinical features, laboratory indicators obtained within 24&#xa0;h of admission, and chest CT findings were collected. Predictive factors were identified using univariate and multivariate logistic regression analyses. Ten machine learning algorithms were used to construct a clinical model (based on clinical and laboratory data only) and a combined model (incorporating chest CT features). Model performance was evaluated using AUC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score, and SHAP was applied to interpret the models.</p> Results <p>Ten predictive factors were identified: neutrophil percentage (NEU%), C-reactive protein (CRP), neutrophil-to-lymphocyte ratio (NLR), body temperature at the initial visit, segmental consolidation, patchy consolidation, pleural effusion, bilateral involvement, lobar consolidation, and multilobar involvement. Among all algorithms, the LightGBM-based clinical model achieved an AUC of 0.754 (95% CI: 0.648–0.837) and a sensitivity of 0.883 in the test set, whereas the LightGBM-based combined model achieved an AUC of 0.825 (95% CI: 0.733–0.904) and a sensitivity of 0.872 in the test set. SHAP analysis demonstrated that NEU%, pleural effusion, and segmental consolidation were positive contributors to MPP prediction.</p> Conclusion <p>This study successfully developed and validated a LightGBM-based diagnostic model for the early identification of pediatric MPP, demonstrating good predictive performance and interpretability. By integrating readily accessible clinical and imaging variables, the model provides pediatricians with a valuable decision-support tool for the early diagnosis and individualized treatment of MPP.</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

Early prediction of mycoplasma pneumoniae pneumonia in pediatric patients: an interpretable machine learning model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Hui Liu,
  • Yang Xin,
  • Xiaojuan Xu,
  • Rina Sa,
  • Hai Du

摘要

Objective

The incidence of Mycoplasma pneumoniae pneumonia (MPP) in children has increased in recent years, accompanied by a trend toward more severe clinical presentations. Early and accurate diagnosis is essential for timely targeted therapy and improved prognosis. This study aimed to develop and validate a machine learning-based diagnostic model integrating clinical characteristics, laboratory indicators, and chest CT findings for the early prediction of MPP in children.

Methods

A total of 449 children diagnosed with community-acquired pneumonia (CAP) who were hospitalized at Ordos Central Hospital between January 2019 and April 2025 were retrospectively enrolled, including 311 MPP cases and 138 non-MPP cases. The training set included 314 patients (217 MPP and 97 non-MPP), while the test set included 135 patients (94 MPP and 41 non-MPP). Clinical features, laboratory indicators obtained within 24 h of admission, and chest CT findings were collected. Predictive factors were identified using univariate and multivariate logistic regression analyses. Ten machine learning algorithms were used to construct a clinical model (based on clinical and laboratory data only) and a combined model (incorporating chest CT features). Model performance was evaluated using AUC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score, and SHAP was applied to interpret the models.

Results

Ten predictive factors were identified: neutrophil percentage (NEU%), C-reactive protein (CRP), neutrophil-to-lymphocyte ratio (NLR), body temperature at the initial visit, segmental consolidation, patchy consolidation, pleural effusion, bilateral involvement, lobar consolidation, and multilobar involvement. Among all algorithms, the LightGBM-based clinical model achieved an AUC of 0.754 (95% CI: 0.648–0.837) and a sensitivity of 0.883 in the test set, whereas the LightGBM-based combined model achieved an AUC of 0.825 (95% CI: 0.733–0.904) and a sensitivity of 0.872 in the test set. SHAP analysis demonstrated that NEU%, pleural effusion, and segmental consolidation were positive contributors to MPP prediction.

Conclusion

This study successfully developed and validated a LightGBM-based diagnostic model for the early identification of pediatric MPP, demonstrating good predictive performance and interpretability. By integrating readily accessible clinical and imaging variables, the model provides pediatricians with a valuable decision-support tool for the early diagnosis and individualized treatment of MPP.