Abstract Introduction Extremely preterm infants’ post-birth fluid status impacts the pathogenesis of bronchopulmonary dysplasia (BPD) with limited data guiding fluid administration strategies. Objective (1) Describe trajectories of fluid balance between infants with and without BPD or death, (2) temporally analyze the weights and interactions of fluid balance indicators over the first postnatal week and (3) integrate these models into optimal daily fluid prescriptions based on input parameters. Design/Methods Multicenter cohort study in infants born at 29 weeks’ gestation at the University of Alabama at Birmingham (discovery cohort) and Yale New Haven (validation) hospitals. The primary outcome was BPD (based on any respiratory support at 36 weeks’ postmenstrual age) or in-hospital death. Models were developed using daily measures from the first postnatal week including total fluid intake, urine output, serum sodium, and weight with further incorporation of birth weight and gestational age (GA). Each variable was characterized by level and shape over the first postnatal week using daily values, moving averages, linear slopes, polynomial trajectories, and pairwise interactions (e.g. total fluids x sodium). Linear mixed models were used to differentiate parameter trajectories by outcome. A binary classification model for BPD or death was developed using gradient-boosted decision tree models (LightGBM, version 4.5) with the determination of feature importance to daily models. From these models a fluid estimation tool was developed that incorporates patient-specific inputs such as birth weight and urine output to produce individualized fluid prescription recommendations. Results Data from 425 infants (2,939 infant-days) were available in the discovery cohort (average GA: 26; SD 1.7) and 234 infants (1,515 infant-days) in the validation cohort (average GA: 26; SD 1.8). The outcome of BPD or death occurred in 55% and 71% of the discovery and validation cohorts, respectively. Infants that developed BPD or death were exposed to more total fluid, regained more weight, and had higher urine output. Gestational age, birth weight, and parameters of urine output and total fluid were the top predictors of BPD or death (Figure 1: Panel A, B). From these models a fluid estimator was successfully constructed (Figure 1: Panel C). Conclusions Compared to survivors without BPD, extremely preterm infants with BPD or death exhibit distinct characteristics of fluid balance from which variable, temporal feature importance can be integrated into an estimator for daily fluid prescriptions. This abstract is funded by: NIH
Gentle et al. (2026) studied this question.