Development and validation of a risk score to predict neonatal mortality among NICU admissions in Southern Ethiopia: a retrospective follow-up study

TL;DR

This study developed a risk score to predict neonatal mortality in Ethiopia's NICU, aiming to improve healthcare outcomes for newborns.

Contribution

A novel risk score model was developed and validated for predicting neonatal mortality in Southern Ethiopia.

Findings

The model achieved an AUC of 0.781 with 80% sensitivity and 66% specificity.

Seven key variables were identified as significant predictors of neonatal mortality.

The model showed good calibration and net benefit across threshold probabilities.

Abstract

The World Health Organization reported 2.6 million neonatal deaths in 2016, accounting for nearly 46% of all under-five deaths globally. Ethiopia is among the top 10 countries with the highest neonatal mortality, with an estimated 122,000 newborn deaths annually. This study aimed to develop and validate a risk score to predict neonatal mortality.

We conducted a retrospective follow-up study among 845 neonates admitted tot Hawassa University Comprehensive Specialized Hospital, Southern Ethiopia. Data were entered into EpiData version 4.6 and analyzed using R version 4.0.5. Variables with p < 0.25 in the bivariable analysis were entered into the multivariable model. A stepwise backward elimination technique with p < 0.1 for the likelihood ratio test to fit the reduced model. Finally, variables with p < 0.05 were considered statistically significant.

Of the 845 neonates included in the study, 130 died, resulting in a neonatal mortality incidence proportion of 15.4% (95% CI: 13%, 17%). Seven variables, namely, residence, primigravida, low birth weight, amniotic fluid status, Apgar score, perinatal asphyxia, and breastfeeding, were included in the model. The AUC of the final reduced validated model was 0.781 (95% CI: 0.73, 0.82). The accuracy of the model was also assessed by calibration and resulted in a p-value of 0.781. The model had a sensitivity and specificity of 80% and 66%, respectively. Decision curve analysis of the model provides a higher net benefit across ranges of threshold probabilities.

We constructed and internally validated a prediction model with good performance. This model is feasible and applicable in healthcare settings to reducing neonatal mortality and improving overall maternal and child healthcare.

Figures