Development and external validation of a machine learning model to predict bronchopulmonary dysplasia using dynamic factors.

We hypothesized that incorporating postnatal dynamic factors would enhance the prediction accuracy of bronchopulmonary dysplasia in preterm infants. This retrospective cohort study included neonates born before 32 weeks of gestation at Seoul National University Hospital between 2013 and 2022. The primary outcome was moderate or severe bronchopulmonary dysplasia. We assessed both static perinatal risk factors and dynamic factors, such as respiratory support type, inspired oxygen concentration, and blood gas analysis results within the first 7 days. The model was developed using data from 546 infants born between 2013 and 2021, with internal validation on 75 infants born in 2022. External validation was based on 105 infants recruited at the Boramae Medical Center. The integrated prediction model, combining static and dynamic factors, showed superior predictive performance, with an area under the receiver operating characteristic curve (AUROC) of 0.841 in the development set, outperforming the static perinatal factor model. Internal validation confirmed the robustness of the integrated model (AUROC: 0.912 vs. 0.805, p < 0.0001). The performance was maintained in the external validation (AUROC: 0.814). Incorporating early respiratory support and blood gas analysis into predictive models substantially improved the accuracy of bronchopulmonary dysplasia prediction in preterm infants.