Predicting Admission to Neonatal Care Unit at Mid-Pregnancy and Delivery Using Data from a General Obstetric Population.

Abstract

Objectives: Development and validation of risk prediction models at mid-pregnancy and delivery to predict admission to the neonatal care unit.

Methods: We used data from all singleton deliveries at Cork University Maternity Hospital (CUMH), Ireland during 2019. Admission to the neonatal care unit was assumed if length of stay in the unit was > 24 h. Multivariable logistic regression with backward stepwise selection was used to develop the models. Discrimination was assessed using the ROC curve C-statistic, and internal validation was assessed using bootstrapping techniques. We conducted temporal external validation using data from all singleton deliveries at CUMH during 2020.

Results: Out of 6,077 women, 5,809 (95.6%) with complete data were included in the analyses. A total of 612 infants (10.54%) were admitted to the neonatal care unit for > 24 hours. Six variables were informative at mid-pregnancy: male infants, maternal smoking, advancing maternal age, maternal overweight/obesity, nulliparity and history of gestational diabetes (C-statistic: 0.600, 95% CI: 0.567, 0.614). Seven variables were informative at delivery: male infants, nulliparity, public antenatal care, gestational age < 39 weeks', non-spontaneous vaginal delivery, premature rupture of membranes and time of birth between 17:01-07.59 h (C-statistic: 0.738, 95% CI: 0.715, 0.760). Using these predictors, we developed nomograms to calculate individualised risk of neonatal care unit admission. Bootstrapping indicated good internal performance and external validation suggested good reproducibility.

Discussion: Our nomograms allow the user to quickly estimate individualised risk of neonatal care unit admission. Future research should aim to improve accuracy in early pregnancy to better assist counselling of parents.