Developing and validating an updated neonatal mortality score on admission to neonatal intensive care units in Ethiopia

In low-resource settings, neonatal intensive care units (NICUs) provide care for a high-acuity population of critically ill neonates and therefore capture a substantial proportion of neonatal deaths. Existing early warning scores to identify neonates at risk of dying in NICUs are overly complex, poorly calibrated, or lack generalizability. We conducted a retrospective cohort study in Ethiopia to derive and validate an updated Neonatal Mortality Score (uNMS). Data from a hospital in Addis Ababa of neonates admitted between September 2016 and January 2018 were randomly split into a derivation dataset ( n = 2004, l709 lived and 295 died), and a split-sample internal validation dataset ( n = 669, 558 lived and 111 died). Data from a hospital in Gondar of neonatal admissions between January 2016 to December 2016 were used for geographic external validation ( n = 812, 605 lived and 207 died). To derive the score, we performed univariate then multivariate logistic regression of predictor variables. Discrimination and calibration of the score were evaluated using the area under the curve (AUC), and calibration curves. The uNMS resulted in five variables associated with neonatal mortality in the NICU: gestational age, birthweight, and admission respiratory rate, respiratory distress, and level of consciousness. The uNMS ranges from 0 to 24 and is calculated using simple arithmetic. Score discrimination is excellent, with an AUC of 0.85 (95% CI 0.81 to 0.89) on derivation, 0.87 (95% CI 0.80 to 0.92) on split-sample internal validation, and 0.87 (95% CI 0.83 to 0.90) on geographic external validation. The score is well-calibrated, with a calibration intercept of 0 (95% CI -0.15 to 0.15) and slope of 1 (95% CI 0.89 to 1.11) on derivation, an intercept of 0.11 (95% CI -0.15 to 0.37) and slope of 1.06 (95% CI 0.87 to 1.24) on split-sample internal validation, and an intercept of 0.24 (95% CI 0.03 to 0.45) and slope of 0.90 (95% CI 0.76 to 1.03) on geographic external validation. The uNMS is a simple prognostic early warning score with excellent discrimination and calibration for predicting the risk of neonatal mortality on admission using clinical characteristics. Additional external validation and prospective evaluation are warranted.