Bayesian point process modeling to quantify excess risk in spatial epidemiology: an analysis of stillbirths with a maternal contextual effect

Motivated by the paucity of high quality stillbirth surveillance data and the spatial analyses of such data, the current research sets out to quantitatively describe the pattern of stillbirth events that may lead to mechanistic hypotheses. We broaden the appeal of Bayesian Poisson point process modeling to quantify excess risk while accounting for unobserved heterogeneity. We consider a practical data analysis strategy when fitting the point process model and study the utility of parameterizing the intensity function governing the point process to include a maternal contextual effect to account for variation due to multiple stillbirth events experienced by the same mother in independent pregnancies. Simulation studies suggest that our practical data analysis strategy is reasonable and that there is a variance-bias trade-off associated with the use of a maternal contextual effect. The methodology is applied to the spatial distribution of stillbirth events in Iowa during the years 2005 through 2011 obtained using an active, statewide public health surveillance approach. Several localized areas of excess risk were identified and mapped based on model components that captured the nuanced and salient features of the data. A conditional formulation of the point process model is then considered, which has two main advantages: the ability to easily incorporate covariate information attached to both stillbirth and live birth, as well as obviate the need to estimate the background intensity. We assess the utility of the conditional approach in the presence of unobserved heterogeneity, compare two Bayesian estimation techniques, and extend the conditional formulation to adequately capture spatio-temporal effects. The motivating study comes from the Iowa Registry for Congenital and Inherited Disorders who has a committed interest in the surveillance and epidemiology of stillbirth in Iowa and whether the occurrence might be geographically linked.