Predictive value of bronchoalveolar lavage fluid leukotriene C4 for airway hyperresponsiveness in children with Mycoplasma pneumoniae pneumonia: a machine learning-based study.

BACKGROUND: Airway hyperresponsiveness (AHR) is a frequent sequela after acute Mycoplasma pneumoniae pneumonia (MPP) in children. This study assessed the association between bronchoalveolar lavage fluid (BALF) leukotriene C4 (LTC4) and MPP, evaluated its predictive value for AHR, and developed an early-identification machine-learning model. METHODS: We retrospectively studied 158 children with MPP admitted between January 2023 and December 2024; 20 children undergoing bronchoscopy for airway foreign bodies served as controls. BALF LTC4 was measured by enzyme-linked immunosorbent assay (ELISA). AHR was assessed in 87 patients at a 2-week follow-up (positivity 39.1%) using a composite of clinical and spirometric criteria rather than bronchial provocation testing. Features were selected by least absolute shrinkage and selection operator (LASSO) regression and recursive feature elimination (RFE). Nine machine-learning models, including a support vector classifier (SVC), were compared, and predictions were interpreted with SHapley Additive exPlanations (SHAP). RESULTS: In the matched cohort, BALF LTC4 was higher in MPP than in age- and sex-matched controls (46.14 vs. 10.46 pg/mL, P<0.001). Baseline LTC4 was higher in AHR-positive than AHR-negative patients (80.91 vs. 39.92 pg/mL, P<0.001). LTC4 correlated negatively with follow-up lung function [forced expiratory volume in 1 second (FEV1) percent predicted, FEV1/forced vital capacity (FVC); Spearman's rho =-0.44 to -0.48], whereas C-reactive protein (CRP) and interleukin-6 (IL-6) did not. An SVC using LTC4 and lactate dehydrogenase (LDH) performed best, with a leave-one-out cross-validation area under the curve (AUC) of 0.876 [95% confidence interval (CI): 0.791-0.943], accuracy 81.6%, sensitivity 73.5%, and specificity 86.8%. SHAP analysis indicated LTC4 was the stronger predictor, with a contribution approximately 1.92 times that of LDH. CONCLUSIONS: In this retrospective single-center cohort, BALF LTC4 was associated with post-MPP AHR in children and showed greater predictive value than systemic inflammatory markers. A parsimonious SVC based on BALF LTC4 and serum LDH showed encouraging performance, suggesting potential utility for early risk stratification. However, the 55% follow-up rate and selection toward clinically complex patients may have overestimated performance; prospective multicenter validation with more complete follow-up is warranted before clinical implementation.