Shiny App to Predict the Risk of Death in Very Low Birth Weight Newborns Through a New Classifier

Abstract

Neonatal mortality is a major health concern worldwide, and one of the groups of babies that has contributed most to this outcome is the very low birth weight (VLBW) newborns. Knowing the factors that contribute to the death of these babies and having an idea of the probability of their survival can help health professionals to make important decisions. The objective of this study was to develop a classifier that predicts whether a particular newborn weighing less than 1500 g will survive or die. For this, the tools R and RStudio were used, as well as a database provided by the National Registry of Newborns of Very Low Weight from Portugal. Construction of logistic regression models were performed, and a shiny web application was created to facilitate the use of the final model by health professionals. The application is available in https://rodrigues.shinyapps.io/VLBWnewborn/. Multivariate analysis showed that gestational age, length at birth, prenatal corticosteroids, sex, average of the three Apgar scores (1st, 5th and 10th min), insufflator resuscitation, major congenital malformation, diagnosis of necrotizing enterocolitis and administration of Ibuprofen for treatment of persistent ductus arteriosus are factors responsible for neonatal mortality. This model predicts with 0.926 of certainty the real state of a VLBW newborn. Its area under the ROC curve was 0.891 and 0.797 for internal and external validation, respectively. The fact that it presents better predictive results than the CRIB and SNAPPE II indexes, when using test data, makes it a possible alternative index to be used.