Analysis of influencing factors and construction of a risk prediction model for severe Mycoplasma pneumoniae lobar pneumonia in children based on CT imaging and laboratory indicators

To develop and validate an early risk stratification model for severe Mycoplasma pneumoniae lobar pneumonia in children, integrating computed tomography (CT) imaging features and laboratory indicators. A total of 289 children with Mycoplasma pneumoniae lobar pneumonia were retrospectively enrolled and divided into severe and non-severe groups. Baseline data, laboratory metrics, and CT imaging features were collected. Multivariate logistic regression was used to construct the prediction model, and the SHapley Additive exPlanations (SHAP) framework was applied post hoc to interpret the model predictions. Model discrimination, calibration, and clinical utility were evaluated through internal validation. The final comprehensive model, comprising eight clinical and imaging variables, achieved AUCs of 0.75 and 0.77 in the training and validation sets, respectively. Bootstrap internal validation (1000 resamples) demonstrated robust performance, with an optimism-corrected C-index of 0.723 and a mean absolute calibration error of 0.036. Among the eight variables in the final model, SHAP analysis identified lactate dehydrogenase (LDH), multilobar involvement (≥ 3 lobes), and consolidation CT attenuation value (HU) as the top three contributors to the model's predictions. We developed an interpretable early risk stratification model with moderate discriminative ability for severe childhood Mycoplasma pneumoniae lobar pneumonia. LDH and multilobar involvement were the major predictors, with CT attenuation providing supplementary imaging value beyond clinical and laboratory indicators. Limited by its retrospective single-center design and lack of external validation, the model should serve as an admission-based screening aid rather than a standalone clinical decision tool. Prospective, multicenter external validation is required before routine clinical implementation.