Development and validation of a multivariate predictive model for the risk of refractory mycoplasma pneumoniae pneumonia in children based on quantitative computed tomography features.

Background: Refractory mycoplasma pneumoniae pneumonia (RMPP) in children often responds poorly to treatment and is prone to causing serious complications; therefore, early identification is clinically crucial. The aim of this study is to evaluate quantitative imaging features from pediatric chest computed tomography (CT) scans, develop a multivariate predictive model, perform internal validation of the model, and investigate the predictive value of quantitative CT measures for the risk of RMPP in children. Methods: This study retrospectively enrolled hospitalized pediatric patients who underwent their first chest CT scan between January 2019 and December 2021 and were clinically diagnosed with mycoplasma pneumonia, and selected participants according to established criteria. Cohorts were stratified by hospital: 116 cases from hospital A served as the training set, while 81 cases from hospitals B, C, and D served as the external validation set, for a total of 197 pediatric patients, including 60 with RMPP and 137 with conventional Mycoplasma pneumoniae pneumonia (MPP). Missing data were handled using the complete-cases method. Clinical, laboratory, and quantitative CT imaging features were uniformly collected as candidate predictors. Independent predictors were identified using stepwise logistic regression, and a nomogram model was constructed. Model calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test, and the model's discriminatory performance in the training and validation sets was evaluated via receiver operating characteristic (ROC) curve analysis to complete the internal validation of the model. Results: A total of 60 children with RMPP and 137 children with general mycoplasma pneumoniae pneumonia (GMPP) were included (training set, 116; validation set, 81). Baseline data for all pediatric patients included age, sex, duration of fever, complete blood count, inflammatory markers, and chest CT findings; there were no statistically significant differences in age, sex, or baseline clinical indicators between the training and validation sets (P>0.05). Compared to the general mycoplasma pneumonia group, children with RMPP in the training set exhibited significantly higher levels of C-reactive protein (CRP), lactate dehydrogenase (LDH), and lesion percentage to lung volume (LeV%), alongside a correspondingly lower percentage of ground-glass opacity (GGO%). Multivariate analysis revealed that CRP, LDH, LeV%, and GGO% were independent predictors of RMPP. The developed nomogram had high accuracy, with the area under the curve (AUC) of the ROC values of 0.933 [95% confidence interval (CI), 0.888-0.977] in the training set and 0.922 (95% CI, 0.859-0.985) in the validation set, respectively. The predicted probabilities from the nomogram, as indicated by the calibration curve and Hosmer-Lemeshow test, were in good agreement with the observed probabilities. Conclusions: The nomogram that combines clinical and CT quantitative features may effectively predict the risk of RMPP in children. It has certain predictive value for the risk of RMPP in children and can serve as a reference for the development of clinical management strategies.