A machine learning-based predictive model for multilobar pulmonary consolidation induced by macrolide-resistant
Yan Guo1,2,3, Yonghan Luo3,4
1Department of Reproductive Gynecology, NHC Key Laboratory of Healthy Birth and Birth Defect Prevention in Western China First People's Hospital of Yunnan Province, Kunming, Yunnan, China.
Abstract:
This study aims to develop a machine learning (ML)-based predictive model for assessing the risk of multilobar pulmonary consolidation in children with macrolide-resistant Mycoplasma pneumoniae pneumonia (MRMP) caused by the 23S rRNA A2063G mutation, a subgroup underrepresented in prior studies. A total of 404 MRMP cases diagnosed between October 2024 and February 2025 were included in this study. Key clinical characteristics, including laboratory test results, symptoms, and treatment outcomes, were extracted from electronic medical records. Six ML models, including Logistic Regression, Naive Bayes, K-Nearest Neighbors, Multilayer Perceptron, Random Forest, and XG-Boost, were developed to predict multilobar pulmonary consolidation. Least absolute shrinkage and selection operator (LASSO) regression was used to select relevant variables. Model performance was then evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA). Finally, Sharpley Additive Explanations was used for model interpretability. XG-Boost demonstrated the highest predictive performance with an area under the ROC curve of 0.976 and 0.904 in the training and validation sets, respectively, showing a high sensitivity of 0.97, specificity of 0.81, accuracy of 0.94, and an F1 score of 0.95. Key predictors identified for multilobar pulmonary consolidation included the top 10 variables: C-reactive protein, lactate dehydrogenase, fibrinogen, platelet count, albumin, hemoglobin, creatinine, aspartate aminotransferase, interleukin-6, and oxygen therapy. DCA showed that the model also exhibited strong clinical utility. The XG-Boost predictive model offers a robust tool for identifying high-risk children with MRMP caused by the 23S rRNA A2063G mutation. By integrating clinical features, the model enhances early risk stratification and can support clinical decision-making, improving the accuracy and efficiency of treatment plans.IMPORTANCEMacrolide-resistant Mycoplasma pneumoniae pneumonia caused by the 23S rRNA A2063G mutation poses a significant threat to pediatric health, often leading to severe multilobar pulmonary consolidation. This study develops a high-performance machine learning model (XG-Boost) that accurately predicts this complication using key clinical indicators such as C-reactive protein, lactate dehydrogenase, and IL-6. With an area under the ROC curve of 0.976, the model enables early risk stratification, guiding clinicians in optimizing treatment for high-risk children. By improving diagnostic precision and intervention timing, this tool can reduce disease severity, minimize hospital stays, and enhance patient outcomes. The interpretability of the model via Sharpley Additive Explanations analysis further ensures its clinical applicability, making it a valuable advancement in managing antibiotic-resistant pediatric pneumonia.
Insights
This study developed an XG-Boost machine learning model to predict multilobar pulmonary consolidation in children with macrolide-resistant Mycoplasma pneumoniae pneumonia (MRMP). The model accurately identifies high-risk children, improving early risk stratification and treatment decisions.
Area of Science:
- Pediatric Pulmonology
- Infectious Diseases
- Medical Informatics
Background:
- Macrolide-resistant Mycoplasma pneumoniae pneumonia (MRMP), particularly due to the 23S rRNA A2063G mutation, is a growing concern in pediatric health.
- Multilobar pulmonary consolidation is a severe complication of MRMP, necessitating accurate risk prediction for timely intervention.
Purpose of the Study:
- To develop and validate a machine learning (ML)-based predictive model for assessing the risk of multilobar pulmonary consolidation in children with MRMP caused by the 23S rRNA A2063G mutation.
- To identify key clinical predictors associated with multilobar pulmonary consolidation in this specific pediatric subgroup.
- To enhance early risk stratification and support clinical decision-making for improved treatment outcomes.
Main Methods:
- Utilized a dataset of 404 pediatric MRMP cases diagnosed between October 2024 and February 2025.
- Developed and compared six ML models (Logistic Regression, Naive Bayes, K-NN, MLP, Random Forest, XG-Boost) for predicting multilobar pulmonary consolidation.
- Employed LASSO regression for variable selection, ROC curves and DCA for performance evaluation, and SHAP for model interpretability.
Main Results:
- The XG-Boost model achieved the highest predictive performance, with an AUC of 0.976 (training) and 0.904 (validation).
- The model demonstrated high sensitivity (0.97), specificity (0.81), accuracy (0.94), and F1 score (0.95).
- Key predictors identified included C-reactive protein, lactate dehydrogenase, fibrinogen, platelet count, albumin, hemoglobin, creatinine, AST, IL-6, and oxygen therapy.
Conclusions:
- The XG-Boost predictive model provides a robust and interpretable tool for identifying children at high risk of multilobar pulmonary consolidation due to MRMP (23S rRNA A2063G mutation).
- Integrating clinical features into this ML model enhances early risk stratification, aiding clinicians in optimizing treatment plans and improving patient outcomes.
- This model represents a valuable advancement in managing antibiotic-resistant pediatric pneumonia, potentially reducing disease severity and hospital stays.
