Analysis of Clinical Factors Associated with Preterm Infant Death with Clustered Data Using Marginal Models

Abstract

Background: Preterm infant death is the most sensitive indicator of population health. A birth occurring before the 37th week of pregnancy is a preterm birth. Ethiopia is among the few countries that bear the highest burden of preterm infant deaths. Methods: A retrospective study design was used to collect the data from the neonatal chart of preterm infants admitted to the neonatal intensive care unit in Jimma University Specialized Hospital Neonatology Clinic, SouthWest of Ethiopia from January, 2013 to December, 2015. Marginal model families; generalized estimating equation and alternating logistic regression model have been used to analyze the effects of selected variables on preterm infant death by taking gestational age as a clustering effect. Results: From the descriptive analysis the results showed that among those eligible premature infants, 171 (34.9%) died. Based on the model comparison analysis, alternating logistic regression was the best model to fit the data. The clustering effect parameter alpha (α) in the model is statistically significant with p-value = 0.0001, indicating that based on their gestational age, there is a strong correlation among the preterm infants in regard to death. Analysis from alternating logistic regression models leads to the decision that respiratory distress syndrome (OR=2.8811), sepsis (OR=2.2702), jaundice (OR=8.9289), hyaline membrane disease (OR=3.5145), birth weight (OR=0.4005), antenatal care visit (OR=0.5847), prenatal asphyxia (OR=7.1306) and multiple pregnancy of the mother (OR=1.7976) were significant risk factors of preterm infant death. Furthermore, sex of neonate, mother’s residence, mode of delivery, age at admission, hypoglycemia, and hypothermia were not significant risk factors of preterm infant death in this study. Conclusions: This study shows that as there is a strong correlation within categories of preterm infants based on their cluster (gestational age) and they share similar risks toward death. The study also shows that the alternating logistic regression model was the best model that fits the preterm data than generalized estimating equation. More importantly, this study contributes to the understanding of clinical risk factors influencing preterm infant death.