Spatio-Temporal Modeling of sub-national under-five mortality Rates in a Developing Country Context

Abstract

The mortality indicator used, the Standardized Mortality Rate (SMR) depends to a large degree on the size of the population; its variance is inversely proportional to the expected values and therefore areas with a small population result in estimates that vary greatly. Furthermore, the variability in the observed cases is usually higher than expected, which produces over dispersion. The availability of spatial data is important to distinguish between two sources of extra variability, which are due to ‘spatial dependence’ and the correlation between the spatial unit and contiguous spatial units, generally the adjacent geographical area. The variations in mortality rates are more compounded in health outcomes when it varies over time (years). Bayesian spatio-temporal modeling strategies can be applied to a large number of rare causes of mortality outcomes to enable examination of spatio-temporal variations on smaller geographic scales such as counties (districts) as suggested in [2]. This method allows examination of spatiotemporal variation across states (districts) in a developing country.The hierarchical Bayesian spatiotemporal models were implemented with spatially structured and unstructured random effects, correlated time effects, time varying confounders and space-time interaction terms in the software R-INLA to produce smoothed state level SMRs. The approach was applied to childhood mortality data from DHS between 2003 – 2013 to explore spatio-temporal variation in SMRs. Model-based estimates of SMRs were mapped to explore geographic variation. The model performance and predictions were evaluated using predictive measures such as Deviance information criterion (DIC) Conditional Predictive ordinates (CPO) and Probability Integral Transforms ( PIT)