Development of a clinical and bronchoalveolar lavage fluid biomarker model for predicting post-infectious bronchiolitis obliterans in hospitalized children

Objective To evaluate previously reported clinical factors and explore whether acute-phase bronchoalveolar lavage fluid (BALF) inflammatory biomarkers measured during hospitalization provide additional prognostic information for subsequent post-infectious bronchiolitis obliterans (PIBO). Methods This single-center retrospective cohort included 220 hospitalized children with lower respiratory tract infection who underwent bronchoscopy and BALF testing for clinical indications. BALF biomarkers were measured during the index hospitalization, before PIBO ascertainment at follow-up approximately 2 months after infection. Patients were randomly divided into a model-development cohort ( n = 155) and an internal holdout cohort ( n = 65). Logistic regression was used to construct clinical, BALF biomarker, and combined models. Model development followed the TRIPOD + AI framework, with evaluation of discrimination, calibration, clinical net benefit, bootstrap optimism, and penalized-regression sensitivity analyses. Results During follow-up, 49/220 children (22.3%) met PIBO diagnostic criteria. In the training cohort, PIBO was associated with adenovirus or mixed infection, longer fever duration, PICU admission, mechanical ventilation, and higher BALF matrix metalloproteinase-9, interleukin-8, and myeloperoxidase (MPO) levels (all P < 0.001). In multivariable analysis, each 1-SD increase in fever duration ( OR = 1.72, 95% CI , 1.03–3.01; P = 0.045) and BALF MPO ( OR = 1.93, 95% CI , 1.13–3.61; P = 0.028), adenovirus infection ( OR = 4.68, 95% CI , 1.36–20.03; P = 0.021), and mechanical ventilation ( OR = 4.04, 95% CI , 1.30–13.08; P = 0.016) remained associated with PIBO. Clinical, BALF biomarker, and combined models yielded AUCs of 0.846, 0.815, and 0.860 in training and 0.718, 0.703, and 0.723 in the holdout cohort. Bootstrap-corrected AUC for the combined model was 0.822. In ridge sensitivity analysis, holdout AUCs were 0.742 for the combined model and 0.706 for the clinical model (DeLong P = 0.475). Conclusion Adenovirus infection, prolonged fever, mechanical ventilation, and higher acute-phase BALF MPO were associated with subsequent PIBO. However, adding BALF biomarkers produced only small, non-significant improvements in holdout discrimination. These findings are exploratory and require prospective external validation in larger multicenter cohorts before clinical application.