Predictors of early-onset neonatal sepsis or death among newborns born at <32 weeks of gestation

Objective: To develop a predictive model for early-onset neonatal sepsis or death among infants born at less than 32 weeks of gestation. Study design: This was a case-control study of all deliveries <32 weeks between 2011 and 2015 in a single tertiary care center. Cases were defined as neonates diagnosed with early-onset sepsis based on a blood or cerebrospinal fluid culture or neonates who expired during the first week of life. Controls consisted of neonates without these outcomes. Variables previously identified to be associated with neonatal sepsis or death were abstracted from the medical record. Bivariable analyses and multivariable logistic regression were used to determine independent risk factors for early-onset neonatal sepsis or death. An ROC curve was created and AUC calculated to estimate the predictive capacity of these associations. Results: Of 779 eligible neonates, early-onset neonatal sepsis or death occurred in 73 (9.4%). In bivariable analyses, mothers whose neonates were diagnosed with early-onset sepsis or death were more likely to be obese, have an intrapartum fever, and have meconium-stained amniotic fluid, and were less likely to have received betamethasone or antepartum/intrapartum antibiotics. Gestational age at delivery and birth weight was significantly lower among neonates diagnosed with neonatal sepsis or death. In multivariable analyses, factors remaining independently associated with neonatal sepsis or death were earlier gestational age at the time of delivery (specifically <28 weeks), intrapartum fever, presence of meconium-stained amniotic fluid, and lower birth weight. The AUC for this regression was 0.81 (95% confidence interval 0.77–0.83). Conclusion: Earlier gestational age at the time of delivery, intrapartum fever, meconium, and lower birth weight are independently associated with early-onset neonatal sepsis or death among deliveries occurring at <32 weeks of gestation; these factors can be used to create a model with fair predictive capability.