A dynamic discrimination model for predicting respiratory distress at birth based on the mass volume ratio in fetuses with congenital lung malformations.

Objectives The congenital lung malformation volume ratio (CVR) is a prenatal ultrasound measurement that parameterizes congenital lung malformation (CLM) size. The aims of this study were to use serial measurements to create estimated growth curves of fetal CVR for asymptomatic and symptomatic neonates with CLM and to investigate whether a discriminant prognostic model based on these measurements could predict accurately which fetuses with CLM will require invasive respiratory support at delivery and should therefore be delivered at a tertiary-care facility. Methods This was a retrospective study of fetuses diagnosed prenatally with CLM at three tertiary-care children's hospitals between 2009 and 2016. Those with two or more sonographic measurements of CVR were included. Serial fetal CVR measurements were used to create estimated growth curves for neonates with and those without respiratory symptoms at delivery, defined as requiring invasive respiratory support for the first 24 h after delivery. A discriminant model based on serial CVR measurements was used to calculate the dynamic probability of the need for invasive respiratory support. The performance of this model overall and in preterm and term neonates was compared with those using maximum CVR thresholds of 1.0 and 1.6. Results Of the 147 neonates meeting the inclusion criteria, 16 (10.9%) required postnatal invasive respiratory support. The estimated CVR growth curvemodels showed different growth trajectories for asymptomatic and symptomatic neonates, with significantly higher CVR in symptomatic neonates, and values peaking late in the second trimester at around 25 weeks' gestation in asymptomatic neonates. All prognostic methods had high accuracy for the prediction of the need for invasive respiratory support in term neonates, but the discriminant model had the best performance overall (area under the receiver-operating characteristics curve (AUC) = 0.88) and in the preterm population (AUC = 0.85). Conclusions The estimated CVR growth curves showed different growth patterns in asymptomatic and symptomatic neonates with CLM. The dynamic discriminant model performed well overall and particularly in neonates that were carried to term. Development of an externally validated clinical tool based on this analysis could be useful in determining the site of delivery for fetuses with CLM. Copyright (C) 2019 ISUOG. Published by John Wiley & Sons Ltd.