Antenatal prediction models for short- and medium-term outcomes in preterm infants.

INTRODUCTION

In extremely and very preterm infants, predicting individual risks for adverse outcomes antenatally is challenging but necessary for risk-stratified perinatal management and parents' participation in decision-making about treatment. Our aim was to develop and validate prediction models for short-term (neonatal period) and medium-term (3 years of age) outcomes based on antenatal maternal and fetal factors alone.

MATERIAL AND

METHODS

A population-based study was conducted on 31 157 neonates weighing ≤1500 g and born between 22 and 31 weeks of gestation registered in the Neonatal Research Network of Japan during 2006-2015. Short-term outcomes were assessed in 31 157 infants and medium-term outcomes were assessed in 13 751 infants among the 31 157 infants. The clinical data were randomly divided into training and validation data sets in a ratio of 2:1. The prediction models were developed by factors selected using stepwise logistic regression from 12 antenatal maternal and fetal factors with the training data set. The number of factors incorporated into the model varied from 3 to 10, on the basis of each outcome. To evaluate predictive performance, the area under the receiver operating characteristics curve (AUROC) was calculated for each outcome with the validation data set.

RESULTS

Among short-term outcomes, AUROCs for in-hospital death, chronic lung disease, intraventricular hemorrhage (grade III or IV) and periventricular leukomalacia were 0.85 (95% CI 0.83-0.86), 0.80 (95% CI 0.79-0.81), 0.78 (95% CI 0.75-0.80), and 0.58 (95% CI 0.55-0.61), respectively. Among medium-term outcomes, AUROCs for cerebral palsy and developmental quotient of <70 at 3 years of age were 0.66 (95% CI 0.63-0.69) and 0.72 (95% CI 0.70-0.74), respectively.

CONCLUSIONS

Although the predictive performance of these models varied for each outcome, their discriminative ability for in-hospital death, chronic lung disease, and intraventricular hemorrhage (grade III or IV) was relatively good. We provided a bedside prediction tool for calculating the likelihood of various infant complications for clinical use. To develop these prediction models would be valuable in each country, and these risk assessment tools could facilitate risk-stratified perinatal management and parents' shared understanding of their infants' subsequent risks.