Background: Necrotizing pneumonia is a severe complication of Mycoplasma pneumoniae pneumonia (MPP) in children. Early recognition remains challenging because initial clinical manifestations are often non-specific, highlighting the need for a practical tool for early risk stratification. Methods: We conducted a single-center retrospective study of hospitalized children with MPP. Data from 2015–2023 were used for model development, and patients enrolled in 2024 were reserved for temporal validation. We compared candidate machine-learning algorithms and selected a parsimonious random forest model using routinely available variables obtained during the early hospitalization period. Model performance was evaluated using discrimination, calibration, and decision curve analysis, and model interpretability was assessed using SHapley Additive exPlanations (SHAP). Results: The random forest model showed good discriminatory performance in internal validation and retained acceptable performance in the 2024 temporal cohort. Calibration indicated reasonable agreement between predicted and observed risks. Decision curve analysis suggested potential clinical value as a supportive tool for early risk stratification. SHAP analysis highlighted fever duration, C-reactive protein, pleural effusion, alanine aminotransferase, and gamma-glutamyl transferase as the main contributors to model prediction. Conclusions: We developed and temporally validated a clinical prediction model for necrotizing pneumonia in children hospitalized with MPP. The model may support early risk stratification using routinely available clinical data, but it is intended to complement rather than replace clinical judgment. External prospective validation is required before routine clinical implementation.
Lu et al. (Sun,) studied this question.