External Validation of a Model for Predicting Outcomes in Preterm Newborns.

Abstract

Predicting outcomes in preterm newborns is crucial for guiding therapeutic decisions and personalized rehabilitation. Prognostic models are a promising approach but rigorous validation is required to ensure their reliability and generalizability. To provide external validation of the PRETERM-POM model, a multimodal prognostic model developed for predicting outcomes in extremely preterm newborns at age 2 years. In this prognostic study conducted at Amiens-Picardie University Hospital (France), the original parameters of the model, developed through multivariate analysis and decision-tree algorithms, were applied to a temporal validation group of preterm newborns born between January 2018 and January 2021 that did not overlap with the development population. The study included preterm newborns born at 23 to 28 weeks of gestational age who had undergone cranial ultrasound and conventional electroencephalogram within the first 14 days after delivery and follow-up until a corrected age of 2 years. Neurodevelopment, assessed with the Denver Developmental Screening Test-II, was classified as favorable (no severe neurodevelopmental impairment [NDI]) or adverse (death or severe NDI). The area under the curve (AUC) for decision-tree classifications of adverse vs favorable outcomes was determined for the validation group and compared with the model development group using the DeLong test. Calibration and model fit were assessed using calibration-in-the-large, calibration curves, the Hosmer-Lemeshow test, and Brier score. For the validation group (104 participants; median [IQR] gestational age at delivery, 26.3 [25.4-27.7] weeks; 46 [44.2%] male participants), the model achieved an AUC of 85.9% (95% CI, 79.0%-92.8%) for predicting outcomes, closely mirroring the performance observed in the development cohort. The model fit was good although it tended to underestimate the risk of adverse outcomes. In this external validation study, the high performance of the model for predicting NDI in preterm newborns was confirmed. This multimodal approach, with a transparent display of risk factor contributions, enhances the potential of this model for clinical applications guiding timely management and decision-making.