A machine learning approach to predict mortality and pulmonary hypertension severity in newborns with congenital diaphragmatic hernia

Abstract

Prenatal prediction of postnatal outcomes in newborns with congenital diaphragmatic hernia (CDH) remains challenging, especially for mortality and neonatal persistent pulmonary hypertension (PPHN). Despite the increasing utilization of advanced artificial intelligence (AI) technologies in the neonatal field, this study is pioneering in exploring AI methodologies in the context of CDH. It represents an initial attempt to implement a Machine Learning (ML) system to predict postnatal mortality and PPHN severity, using prenatal and early postnatal data as input variables. We enrolled 50 patients with isolated left-sided CDH from singleton pregnancies and retrospectively collected clinical and imaging variables from fetal ultrasound (US) and shape features extracted from magnetic resonance imaging (MRI), combined with gestational age and birth weight. A supervised ML model for predicting mortality and PPHN severity was developed, achieving good accuracy (88% for mortality prediction and 82% for PPHN) and sensitivity (95% for mortality and 85% for PPHN). The area under the curve (AUC) of the ROC curve was 0.88 for mortality and 0.82 for PPHN predictions. Our results may lead to novel AI applications in the neonatal field, focusing on predicting postnatal outcomes based on prenatal data, ultimately improving prognostic assessments and intervention strategies for such a complex disease.