Identifying Causes of Neonatal Mortality from Observational Data: A Bayesian Network Approach

Despite improvements in access to birth facilities, neonatal mortality remains a critical health issue in many developing countries and causes are not fully understood. The Global Network Maternal Newborn Health Registry provides a rich source of data of neonatal mortality risk factors and outcomes to identify direct causes and higher-level determinants, however performing causal inference using observational data is difficult and remains an open problem in epidemiology. In this paper we sought to determine whether Bayesian networks can be used to identify the complex causal pathways leading to neonatal mortality outcomes and to quantify the effect of each cause on mortality. Our analysis identified a complex network of causes that contribute to neonatal mortality, including maternal death, pre-term birth, movement and breathing at birth. For variables identified as direct causes we estimated the average causal effect using logistic regression models that controlled for known confounders.