Prognostic model for predicting survival in very preterm infants: an external validation study

Objective To perform a temporal and geographical validation of a prognostic model, considered of highest methodological quality in a recently published systematic review, for predicting survival in very preterm infants admitted to the neonatal intensive care unit. The original model was developed in the UK and included gestational age, birthweight and gender. Design External validation study in a population-based cohort. Setting Dutch neonatal wards. Population or sample All admitted white, singleton infants born between 23(+0) and 32(+6) weeks of gestation between 1 January 2015 and 31 December 2019. Additionally, the model's performance was assessed in four populations of admitted infants born between 24(+0) and 31(+6) weeks of gestation: white singletons, non-white singletons, all singletons and all multiples. Methods The original model was applied in all five validation sets. Model performance was assessed in terms of calibration and discrimination and, if indicated, it was updated. Main outcome measures Calibration (calibration-in-the-large and calibration slope) and discrimination (c statistic). Results Out of 6092 infants, 5659 (92.9%) survived. The model showed good external validity as indicated by good discrimination (c statistic 0.82, 95% CI 0.79-0.84) and calibration (calibration-in-the-large 0.003, calibration slope 0.92, 95% CI 0.84-1.00). The model also showed good external validity in the other singleton populations, but required a small intercept update in the multiples population. Conclusions A high-quality prognostic model predicting survival in very preterm infants had good external validity in an independent, nationwide cohort. The accurate performance of the model indicates that after impact assessment, implementation of the model in clinical practice in the neonatal intensive care unit could be considered. Tweetable abstract A high-quality model predicting survival in very preterm infants is externally valid in an independent cohort