Abstract
Objective: Existing predictive models for bronchopulmonary dysplasia (BPD) often lack external validation, limiting their clinical use. This study aimed to externally validate recent BPD prediction models using baseline variables, in a population-based cohort.
Design: This was an external validation study conducted on data collected from 2014 to 2021.
Setting: This was a retrospective, multicentre, population-level cohort with prospectively collected data.
Participants: Extremely low gestational age neonates recorded in the SwissNeoNet registry across all nine level III neonatal care units in Switzerland (n=1748) were included.
Interventions: Recent BPD prediction models estimating the risk of BPD or death at 36 weeks postmenstrual age, based on predictors available within the first 24 hours of life.
Main outcome measures: The primary outcome was survival without BPD. A systematic literature search identified five eligible models, which were externally validated and recalibrated for the Swiss cohort. The most performant model was further optimised to improve local applicability.
Results: Among 693 screened studies, five models based solely on perinatal variables were included. Without recalibration, models showed fair discrimination (area under the curve (AUC) 0.70-0.76) but variable calibration (observed/expected (O/E) 0.58-0.80). After recalibration, AUCs ranged from 0.69 to 0.76, and calibration improved (O/E 0.58-1.61). The optimised version of the best-performing model demonstrated improved calibration (O/E 1.03) and was validated in the Swiss population.
Conclusion: By comparing and externally validating existing BPD prediction models, we propose an optimised model using baseline variables at birth, enhancing its applicability to both the Swiss population and similar clinical contexts.