Machine learning-based prediction of severe Mycoplasma pneumoniae pneumonia in pediatric patients

Introduction

Artificial intelligence (AI) has demonstrated significant performance in retrospective medical studies; however, its clinical translation remains limited due to challenges such as system efficiency, complexity, and human-AI collaboration. Randomized controlled trials (RCTs) and related studies are essential for validating its practical utility, and the U.S. Food and Drug Administration (FDA) is accelerating the approval of AI-based medical products. Deep learning has achieved notable progress in medical imaging, including radiology, pathology, gastroenterology, and ophthalmology, with applications in lung cancer risk prediction, cancer diagnosis, and colonic lesion assessment during colonoscopy. Current AI research primarily focuses on image classification, but future directions include analyzing non-imaging data such as text and genomic sequences, leveraging unsupervised learning for unlabeled datasets, and developing human-AI collaborative systems to enhance clinical applicability (1).

Mycoplasma pneumoniae pneumonia (MPP) is an inflammatory lung disease caused by Mycoplasma pneumoniae (MP) infection, affecting the bronchi, bronchioles, alveoli, and lung interstitium. MPP is a common lower respiratory tract infection in pediatric populations, particularly in children aged five years and older, and accounts for approximately 30% of pediatric community-acquired pneumonia (CAP) cases (2-4). Clinical manifestations are highly variable and typically include fever and paroxysmal, irritating dry cough, with some patients exhibiting shortness of breath and hypoxia. Mild MPP generally resolves within 10 days, is self-limiting with a favorable prognosis, and rarely results in sequelae.

Severe MPP (SMPP) represents a more critical form of MPP, characterized by rapid disease progression, respiratory failure, or life-threatening extrapulmonary complications, with a subset of cases requiring intensive life support. SMPP is associated with severe symptoms, a prolonged disease course, and complications such as pulmonary embolism, pleural effusion, atelectasis, necrotizing pneumonia, and lung abscess. Extrapulmonary manifestations may include skin and mucosal damage, myocardial injury, and in critically ill children, conditions such as acute respiratory distress syndrome and respiratory failure (5,6). The lack of specificity in clinical presentation makes early identification of SMPP challenging, highlighting the importance of timely risk assessment. This study retrospectively analyzed the clinical data of 407 pediatric patients with MPP to identify risk factors for progression to severe disease. We present this article in accordance with the TRIPOD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2025-aw-743/rc).

Methods

Study participants

A retrospective analysis was conducted on 407 pediatric patients hospitalized with MPP in the Department of Integrated Traditional Chinese and Western Medicine at Xi’an Children’s Hospital from June 2023 to November 2023. Patients were categorized into mild (n=123) and severe (n=284) MPP groups according to the “Diagnosis and Treatment Guidelines for Mycoplasma Pneumoniae Pneumonia in Children (2023 Edition)” based on disease severity.

Inclusion criteria: (I) age <18 years; (II) diagnosis of MPP; and (III) availability of complete clinical data. Exclusion criteria: (I) concurrent infection with other pathogens during the disease course; (II) presence of severe cardiac, hepatic, or renal disease, abnormal airway structure, or immunodeficiency; (III) inability of the child or family members to cooperate with clinical diagnosis and treatment, resulting in incomplete case data.

The diagnosis of MPP was based on the “Diagnosis and Treatment Guidelines for Mycoplasma Pneumoniae Pneumonia in Children (2023 Edition)” (7): (I) the main clinical manifestations included fever and cough, with some patients presenting with wheezing. Pulmonary signs may be subtle in the early stage; as the disease progresses, decreased breath sounds and dry or wet rales may appear. (II) Chest computed tomography (CT) confirmed the presence of pneumonia. (III) Laboratory evidence included: (i) a single serum MP antibody titer ≥1:160 [particle aggregation (PA) method]; (ii) positive MP-immunoglobulin M (MP-IgM) antibody; (iii) a fourfold or greater rise in paired serum MP antibody titers during the disease course; and (iv) detection of MP-DNA or RNA.

The distinction between mild and severe cases was based on clinical manifestations and guideline criteria.

• Mild cases: patients who did not meet the criteria for severe disease, with a disease course typically lasting 7–10 days, generally had a favorable prognosis and no sequelae.
• Severe cases: patients meeting any of the following criteria were classified as severe: (i) persistent high fever ≥39 ℃ for ≥5 days, or fever lasting ≥7 days without a decreasing trend in peak body temperature; (ii) presence of one or more respiratory symptoms, including wheezing, shortness of breath, dyspnea, chest pain, or hemoptysis; (iii) occurrence of extrapulmonary complications not reaching the threshold for critical illness; the clinical decision threshold was determined based on the Youden index (sensitivity + specificity − 1) from the receiver operating characteristic (ROC) curve of the training set. (iv) finger pulse oxygen saturation ≤0.93 at rest while breathing ambient air; (v) imaging findings, including single-lobe involvement ≥2/3 with uniform high-density consolidation, involvement of two or more lobes with high-density consolidation (regardless of affected area), possible moderate to large pleural effusion, localized bronchiolitis, diffuse bronchiolitis in a single lung or bilateral involvement of ≥4/5 lobes, potentially combined with bronchitis and mucus plugs causing atelectasis; (vi) progressive worsening of clinical symptoms, with imaging demonstrating lesion expansion >50% within 24 to 48 hours; (vii) significant elevation of at least one laboratory parameter: C-reactive protein (CRP), lactate dehydrogenase (LDH), or D-dimer (D-D).

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by Institutional Ethics Committee of Xi’an Children’s Hospital (No. 2025-051-01). Due to its retrospective design, exemption from informed consent was granted.

Data collection

Demographic information, including age, sex, and month of admission, was recorded, along with clinical manifestations such as duration of fever and cough, pulmonary signs, and complications. All patients meeting the inclusion criteria underwent chest imaging examinations. Within 24 hours of hospitalization, pharyngeal swab cultures and laboratory tests were performed, including routine blood tests, MP etiological detection, myocardial enzyme spectrum, and ten liver function parameters. Laboratory variables included white blood cell (WBC) count, high-sensitivity CRP (hs-CRP), erythrocyte sedimentation rate (ESR), procalcitonin (PCT), fibrinogen (FIB), alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatine kinase-MB (CK-MB), creatinine (Cr), LDH, and D-D. Missing data were addressed using imputation with 5 iterations, and the imputation process was based on the correlation between variables. All predictor variables were collected within 24 hours of hospital admission. The proportions of missing data were as follows: FIB, 50.12%; D-D, 55.77%; and electrocardiogram (ECG), 87.47%.

Pathogen diagnostic criteria

MP infection was confirmed if the single serum MP-IgM antibody titer was ≥1:160 or if MP-DNA/RNA was detected in nasopharyngeal secretions or bronchoalveolar lavage fluid.

Model development

The dataset was split into training set (70%) and test set (30%) using stratified random sampling to maintain the case mix ratio. Grid search was implemented with 5-fold cross-validation on the training set, and the hyperparameter ranges for each model are listed in Table S1 {e.g., random forest: n_estimators [100, 200, 300], max_depth [5, 10, 15], min_samples_split [2, 5]}. The optimal hyperparameters were selected based on the highest cross-validated area under the curve (AUC). Model calibration was performed using the Hosmer-Lemeshow test, and calibration curves were added to visualize the consistency between predicted and actual values.

Statistical analysis

Data were analyzed using SPSS 24.0 software. Normally distributed measurement data are expressed as mean ± standard deviation, and comparisons between groups were performed using independent-samples t-tests. Non-normally distributed measurement data are expressed as median (interquartile range) and compared using the rank-sum test. Categorical data are presented as counts and percentages (%) and compared using the Chi-squared test. A two-sided P value <0.05 was considered statistically significant. Eight machine learning (ML) algorithms were applied to construct predictive models, and all analyses and computations were performed using R V4.1.2 and Python V3.7.0. All ML models used grid search for hyperparameter tuning. The detailed training procedure has been added as follows: (I) the dataset was split into a training set (70%) and a test set (30%) using stratified random sampling to preserve the case distribution; (II) grid search with 5-fold cross-validation was performed on the training set, with the hyperparameter ranges for each model provided in Table S1 {e.g., Random Forest: n_estimators [100, 200, 300], max_depth [5, 10, 15], min_samples_split [2, 5]}; and (III) the optimal hyperparameters were selected based on the highest cross-validated AUC.

Results

Comparison of general information and clinical characteristics

The general information and clinical characteristics of the 407 pediatric patients with MPP are presented in Table 1. Among patients in the severe group, 43 cases (15.14%) presented with pleural effusion, and 7 (2.46%) had pericardial effusion. Respiratory and systemic manifestations in the severe group included wheezing in 38 cases (13.38%), rash in 12 (4.22%), chest pain in 1 (0.35%), and diarrhea in 1 (0.35%).

Development of prediction models

Least absolute shrinkage and selection operator (LASSO) regression analysis was applied to identify the most relevant predictive features from numerous clinical variables. LASSO effectively reduced model complexity and mitigated overfitting by shrinking coefficients while retaining variables with significant predictive value. The features selected for inclusion in the final model were month of admission, duration of fever, Cr, WBC count, hs-CRP, PCT, FIB, gender, ECG findings, lung CT, and other complications, as presented in Figure 1.

Figure 1 Graph of LASSO path and cross-validation plot for LASSO regression. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CK-MB, creatine kinase-MB; CT, computed tomography; CV, cross-validation; ESR, erythrocyte sedimentation rate; FIB, fibrinogen; hs-CRP, high-sensitivity C-reactive protein; LASSO, least absolute shrinkage and selection operator; LDH, lactate dehydrogenase; PCT, procalcitonin; WBC, white blood cell.

Eight ML algorithms—random forest (RF), Categorical Boosting (CatBoost), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), support vector machine (SVM), decision tree, Gradient Boosting, and artificial neural network (ANN)—were applied to construct predictive models. The performance of these models was evaluated using the AUC-ROC, which indicated that the RF model achieved the highest value (Figure 2). RF achieved an AUC of 0.98 [95% confidence interval (CI): 0.96–0.99] with an accuracy of 0.92 (95% CI: 0.89–0.95), while LightGBM achieved an AUC of 0.97 (95% CI: 0.95–0.99) with an accuracy of 0.94 (95% CI: 0.91–0.96). In comparison, the logistic regression model showed an AUC of 0.89 (95% CI: 0.85–0.93) and an accuracy of 0.84 (95% CI: 0.80–0.88), which were markedly lower than those of the RF model. This comparison confirms that, relative to conventional logistic regression, advanced ML models provide superior discriminative performance and incremental value, particularly in capturing the nonlinear relationships between predictors and severe MP.

Figure 2 ROC curves of eight ML models for predicting the individual risk of severe Mycoplasma pneumoniae pneumonia. ANN, artificial neural network; CatBoost, Categorical Boosting; LightGBM, Light Gradient Boosting Machine; ML, machine learning; ROC, receiver operating characteristic; SVM, support vector machine; XGBoost, eXtreme Gradient Boosting.

Model performance metrics, including accuracy, precision, recall, specificity, and F1 score, were calculated. The RF model demonstrated an accuracy of 0.92, precision of 0.91, recall of 0.91, and F1 score of 0.91, as presented in Table 2.

Explanation of the model

SHapley Additive exPlanations (SHAP) was used to interpret the predictive models, quantifying the contribution of each feature to the model output. SHAP is based on the Shapley value concept from game theory and provides insight into how individual features influence model predictions. Across the eight ML models, the factors most closely associated with the individual risk of severe MPP included month of admission, Cr, and WBC count, with minor variations in feature importance rankings.

Analysis of predictive patterns indicated that admission month clustered between July and September, indicating that seasonal factors may modulate disease severity, potentially due to interactions between climatic conditions and pathogen epidemiology.

At the biomarker level, WBC exhibited a negative contribution, indicating that elevated WBC levels may reflect an effective host immune response. FIB in the ANN model supported a pathogenic mechanism involving coagulation dysfunction. PCT demonstrated a negative effect, which, together with clinical observations, suggests that viral coinfection may suppress PCT release, a hypothesis that requires further mechanistic validation.

Model heterogeneity was observed across algorithm types. Tree-based models (decision tree and RF) prioritized duration of fever, consistent with their sensitivity to temporal features. ANN displayed heightened sensitivity to electrocardiographic abnormalities, suggesting that myocardial injury may serve as a potential marker for severe disease. Boosting algorithms (Gradient Boosting and XGBoost) effectively captured interactions between hs-CRP and FIB, revealing synergistic mechanisms within the inflammation-coagulation network.

These findings informed actionable clinical thresholds: intensive care unit monitoring is recommended for patients with hospitalization exceeding nine days, and CRP testing is advised for patients admitted in July, translating model insights into practical clinical decision-making (Figure 3).

Figure 3 Ranking of importance scores for the features of eight prediction models. ANN, artificial neural network; CatBoost, Categorical Boosting; CT, computed tomography; FIB, fibrinogen; hs-CRP, high-sensitivity C-reactive protein; LightGBM, Light Gradient Boosting Machine; PCT, procalcitonin; RF, random forest; SVM, support vector machine; XGBoost, eXtreme Gradient Boosting; WBC, white blood cell.

The SHAP summary plot demonstrates the distribution of the impact of each feature on the model output. The horizontal axis represents SHAP values, indicating the degree of influence of each feature, while the vertical axis lists the feature names. The color scale reflects the magnitude of feature values, with red representing high values and blue representing low values. Features such as admission month show wide distributions, indicating significant influence on the model’s predictions. For example, Cr demonstrates distinct distributions of SHAP values for high and low levels, indicating that Cr specifically affects the model’s assessment of severe disease risk. Similarly, WBC values correlate with changes in model output, revealing whether higher or lower values promote or inhibit predicted outcomes and clarifying the mode of action of different features in the model.

The waterfall plot depicts the model decision-making process for a single patient. The baseline output, E[f(x)] =0.731, represents the average output for all samples, and the final output, f(x) =0.998, represents the individualized prediction. Features are ordered by their influence, with red blocks indicating factors that increase the output and blue blocks indicating those that decrease it. For example, admission month =10 increased the output by +0.09, while PCT =1.548 ng/mL and WBC =12.21×10⁹/L each decreased it by −0.03. This visualization allows the contribution of each feature to be traced step by step, providing clear insight into the model’s logic for specific cases and supporting the clinical interpretation and application of model predictions. The feature importance plot demonstrates that, for a female patient admitted in October with 8 days of hospitalization and values of FIB =4.44 g/L, Cr =25 µmol/L, PCT =1.548 ng/mL, and WBC =12.21×10⁹/L, the predicted probability of severe pneumonia was 99.8%.

In summary, the SHAP analyses comprehensively illustrate the relationship between clinical features and model output, capturing both the overall distribution of feature impacts and the decision-making pathway for individual cases. These visualizations facilitate the exploration of feature contributions in disease-related models and provide an interpretable basis for medical decision-making and research. Future studies should further investigate in-depth associations between features and disease prognosis, diagnosis, and outcomes based on these insights (Figure 4).

Figure 4 SHAP explanation of the RF model. CatBoost, Categorical Boosting; CT, computed tomography; FIB, fibrinogen; hs-CRP, high-sensitivity C-reactive protein; LightGBM, Light Gradient Boosting Machine; PCT, procalcitonin; RF, random forest; SHAP, SHapley Additive exPlanations; SVM, support vector machine; XGBoost, eXtreme Gradient Boosting; WBC, white blood cell.

Discussion

Mycoplasma is a common pathogen responsible for respiratory tract infections and exhibits a relatively high detection rate in CAP. Annual outbreaks of M. pneumoniae have been reported globally, with regional periodic epidemics observed. Since 2021, the incidence of SMPP has been increasing. Compared with mild cases, SMPP presents with more pronounced clinical symptoms and a higher likelihood of complications, including pleural effusion and pulmonary embolism. The early clinical manifestations of SMPP are non-specific, making early diagnosis challenging, and delayed recognition can lead to serious sequelae. Consequently, analyzing the risk factors for SMPP in children and developing predictive models has significant clinical importance.

The clinical characteristics of 407 children with MPP were retrospectively analyzed in this study, and factors associated with SMPP were investigated using ML algorithms. Among the eight predictive models developed, the RF model exhibited the highest performance, with an AUC of 0.98, surpassing the other seven models. SHAP was employed to interpret the models, identifying twelve key predictive features and enhancing their clinical interpretability. The SHAP summary and force plots of the RF model provide clinicians with a clear and intuitive understanding of feature contributions. Key factors associated with SMPP included admission month, length of hospital stay, Cr, hs-CRP, FIB, days of fever, and other clinical indicators, whereas WBC count and PCT demonstrated negative associations with disease severity.

The findings of this study are consistent with previous reports indicating that children with refractory MPP exhibit longer fever duration, prolonged hospitalization, and higher rates of extrapulmonary complications compared with non-refractory cases (8,9). In this cohort, severe cases predominated between August and October, reflecting the seasonal characteristics of MPP and highlighting the need for heightened surveillance during epidemic periods, particularly in school-aged children. Prolonged hospitalization reflects the clinical severity of SMPP, often resulting from extended treatment requirements for complications such as pleural effusion and atelectasis. Its high SHAP value supports the clinical understanding that hospital stay serves as a composite indicator of disease severity. Extended hospitalization may also increase the risk of nosocomial infections, emphasizing the need to optimize management strategies for critically ill patients.

MP infection activates both the endogenous and exogenous coagulation systems through multiple pathways, resulting in abnormal coagulation and promoting thrombosis (10). Regarding coagulation indicators, the FIB level in children with SMPP was higher than in those with MPP, indicating an increased risk of embolism in severe cases. CRP is a non-specific acute-phase marker synthesized by the liver, that rises significantly in response to severe tissue damage. Given the extensive lung tissue injury and intense inflammatory response in SMPP, CRP levels are markedly elevated in affected patients (11). Evidence indicates that CRP, as a pro-inflammatory factor, can act on endothelial cells and influence FIB concentration, potentially leading to acute thrombosis. CRP may also affect coagulation through multiple pathways, though the precise mechanisms remain unclear. Studies have demonstrated that CRP is independently associated with SMPP, and clinicians should consider the possibility of severe disease when CRP exceeds 36 mg/L (12). Because CRP has a short half-life, its levels decrease rapidly as the patient’s condition improves, making it a valuable indicator for disease prognosis. Monitoring changes in serum CRP levels in patients with SMPP has practical significance for assessing disease progression (13).

PCT is an acute-phase reactant protein synthesized in the liver and serves as the precursor peptide of calcitonin without hormonal activity. Under normal physiological conditions, PCT is secreted primarily by thyroid C cells and neuroendocrine cells in the lungs, and its circulating levels are extremely low. During infection, particularly when stimulated by bacterial endotoxins, parenchymal tissue cells in multiple organs synthesize PCT, leading to a substantial increase in plasma concentrations (14). The application of PCT testing in MPP remains controversial. Clinicians generally report that PCT levels are only slightly elevated or remain within the normal range in MPP, and PCT testing is less frequently performed in outpatient or emergency settings for suspected MP infections. A multicenter study evaluating 1,735 hospitalized patients with CAP found that median PCT levels were higher in patients with typical bacterial pneumonia compared with atypical bacterial or viral pneumonia (2.50, 0.20, and 0.09 μg/L, respectively) (15).

However, several limitations should be acknowledged. First, this was a single-center retrospective study with a relatively small sample size, which may limit the generalizability of the findings. Second, the absence of external validation restricts the ability to confirm the robustness and applicability of the results across different clinical settings. Third, a certain degree of case mix imbalance existed in the study population, which may have influenced the stability of the predictive performance. To address these issues, future research should focus on conducting multi-center prospective studies with larger and more balanced samples. External validation across different regions and epidemic periods is also necessary to assess the reproducibility of the findings. In addition, comparing the predictive performance of the proposed model with conventional clinical scoring systems, such as CURB-65, would further clarify its incremental clinical value and practical applicability.

Conclusions

An interpretable ML model was developed and validated to predict the individual risk of SMPP among children with MPP. This model can rapidly identify patients at risk of developing severe disease using readily available clinical variables. Key factors identified as risk contributors included the month of hospitalization, duration of fever, hs-CRP, and FIB levels. SHAP was applied to interpret the model, enhancing its transparency and clinical applicability. The model may assist clinicians in prioritizing care, optimizing patient management, and reducing hospitalization duration and medical costs.

Acknowledgments

We would like to acknowledge the hard and dedicated work of all the staff that implemented the intervention and evaluation components of the study.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-aw-743/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-aw-743/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-aw-743/prf

Funding: This study was supported by the Shaanxi Provincial Natural Science Foundation (No. 2022JM-534) and the Project in the Field of Social Development of the Shaanxi Provincial Department of Science and Technology (No. 2024SF-YBXM-509). The funding body had no role in the design of the study and collection, analysis, and interpretation of data or in writing the manuscript.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2025-aw-743/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by Institutional Ethics Committee of Xi’an Children’s Hospital (No. 2025-051-01). Due to its retrospective design, exemption from informed consent was granted.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rajpurkar P, Chen E, Banerjee O, et al. AI in health and medicine. Nat Med 2022;28:31-8. [Crossref] [PubMed]
2. Fu Y, Zhang TQ, Dong CJ, et al. Clinical characteristics of 14 pediatric mycoplasma pneumoniae pneumonia associated thrombosis: a retrospective study. BMC Cardiovasc Disord 2023;23:1. [Crossref] [PubMed]
3. Miao Y, Li J, Huang L, et al. Mycoplasma pneumoniae detections in children with acute respiratory infection, 2010-2023: a large sample study in China. Ital J Pediatr 2025;51:11. [Crossref] [PubMed]
4. Sharplin L, Goyal V. Mycoplasma pneumoniae respiratory tract infections in children: when and how to diagnose and treat. Breathe (Sheff) 2025;21:250046. [Crossref] [PubMed]
5. Chilet F, Iqbal AM, Mitzov N, et al. Acute saddle pulmonary embolism: A rare complication of mycoplasma pneumonia. Respir Med Case Rep 2020;30:101033. [Crossref] [PubMed]
6. Sheng CQ, Yang CF, Ao Y, et al. Mycoplasma pneumoniae pneumonia with pulmonary embolism: A study on pediatric cases in Jilin province of China. Exp Ther Med 2021;21:201. [Crossref] [PubMed]
7. Zhao S, Qian S, Chen Z, et al. Guidelines for the diagnosis and treatment of Mycoplasma pneumoniae in children (2023 edition). Infectious Disease Information 2023;36:291-7.
8. Wu X, Lu W, Liu W, et al. Predictive value of an early comprehensive assessment model for refractory mycoplasma pneumoniae pneumonia and internal validation. BMC Infect Dis 2025;25:744. [Crossref] [PubMed]
9. Zhao J, Ji X, Wang Y, et al. Clinical Role of Serum Interleukin-17A in the Prediction of Refractory Mycoplasma pneumoniae Pneumonia in Children. Infect Drug Resist 2020;13:835-43. [Crossref] [PubMed]
10. Li T, Yu H, Hou W, et al. Evaluation of variation in coagulation among children with Mycoplasma pneumoniae pneumonia: a case-control study. J Int Med Res 2017;45:2110-8. [Crossref] [PubMed]
11. Wang J, Mao J, Chen G, et al. Evaluation on blood coagulation and C-reactive protein level among children with mycoplasma pneumoniae pneumonia by different chest imaging findings. Medicine (Baltimore) 2021;100:e23926. [Crossref] [PubMed]
12. Zhang C, Zhang Q, Du JL, et al. Correlation Between the Clinical Severity, Bacterial Load, and Inflammatory Reaction in Children with Mycoplasma Pneumoniae Pneumonia. Curr Med Sci 2020;40:822-8. [Crossref] [PubMed]
13. Chen L, Liu J, Zhao S, et al. Clinical features and treatment of refractory Mycoplasma pneumoniae pneumonia unresponded to conventional dose methylprednisolone in children. Zhonghua Er Ke Za Zhi 2014;52:172-6. Chinese.
14. Khanna DK, Jayalakshmi KMD, Arun D, et al. A short-term cross-sectional retrospective study on procalcitonin as a diagnostic aid for various infectious diseases. Futur J Pharm Sci 2025;11:29.
15. Meyer Sauteur PM, Krautter S, Ambroggio L, et al. Improved Diagnostics Help to Identify Clinical Features and Biomarkers That Predict Mycoplasma pneumoniae Community-acquired Pneumonia in Children. Clin Infect Dis 2020;71:1645-54. [Crossref] [PubMed]