Abstract
BACKGROUND: Mycoplasma pneumoniae pneumonia (MPP) can impose substantial healthcare burden in children. Because retrospective severity definitions can be heterogeneous, we used total hospitalization cost as a complementary high–resource-use phenotype and aimed to identify admission-time predictors of high cost and develop an early prediction model. METHODS: We retrospectively enrolled 402 hospitalized children with MPP at Shantou Central Hospital, China (Jan–Dec 2024). A Gaussian mixture model (GMM) was fitted to total hospitalization cost to derive a data-driven threshold for a high-cost (high–resource-use) subgroup. Using variables available within the first 24 h of admission, we developed a logistic model including age, Tmax, lactate dehydrogenase (LDH), and radiographic consolidation, and then evaluated an age × consolidation interaction. Discrimination was assessed by ROC AUC; age-stratified analyses were used to interpret joint age–consolidation patterns. RESULTS: A data-driven cost threshold identified a high-cost subgroup (> 7,139 RMB; 41/402, 10.2%), defined a posteriori from the observed hospitalization-cost distribution. High-cost cases had longer length of stay, more frequent intensive management (bronchoalveolar lavage and systemic corticosteroids), and a higher prevalence of consolidation. The four-variable model showed good discrimination (AUC 0.790), which improved after adding the age × consolidation interaction (AUC 0.838). Inflammatory markers were higher in consolidation-positive patients, with the largest differences observed in younger children; correspondingly, the interaction model identified younger children with consolidation as having the highest predicted high-cost risk. CONCLUSIONS: In this single-center retrospective cohort, hospitalization cost operationalized a high–resource-use phenotype in pediatric MPP. An admission-time model incorporating an age × consolidation interaction may support early risk stratification and resource planning in similar settings; external validation is needed before clinical use. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12887-026-06691-z.