Spatiotemporal models with confounding effects: application on under-five mortality across four sub-Saharan African countries

Abstract

Background: Different strategies have been developed to minimize under-five mortality (U5M) in sub-Saharan African (sSA) countries; however, it is still a major health concern for children in the region. Spatiotemporal modeling is important for areal data collected over time. However, when the number of time points and spatial areas is large and the areas are disconnected, fitting the model becomes computationally complex because of the high number of required parameters to be estimated. Therefore, the main aim of this study is to adopt a spatiotemporal dynamic model that includes the confounding effects between time, space, and their interactions with fixed covariates, with a special emphasis on U5M across disconnected sSA countries.

Method: We used nationally publicly representative Demographic and Health Survey (DHS) data for the period from 2000 to 2020. Bayesian spatiotemporal hierarchical modeling with an integrated nested Laplace approximation (INLA) program was used to model the spatiotemporal distribution of U5M among children across 37 districts located in four disconnected sSA regions: Ethiopia, Nigeria, Zimbabwe, and Ghana.

Results: A total of 170,356 under-five children from 37 districts were considered, and 15,467 died before the age of five. The relative risk of U5M in the first DHS was 2.02, which sharply decreased to 0.5 in the recent phase. The proportion of improved access to water, sanitation, clean fuel use, urbanization, and access to health facilities in the district had a significant negative association with U5M. The higher the proportion of these covariates, the lower is the prevalence of childhood mortality.

Conclusion: This study revealed evidence of strong spatial, temporal, and interaction effects that influence under-five mortality risk across districts. Improving the women's literacy index, access to improved water, the use of clean fuel, and the wealth index are associated with an improvement in the risk of mortality among under-five children across the districts. Districts in Nigeria and Ethiopia have the highest risk of U5M; hence, districts in these countries require special attention.

Keywords: confounding; space-time interactions; spatial random effects; spatiotemporal models; variance partitioning.

Figure 1

Map of eligible sub-Saharan African countries included in the study.

Figure 2

Locations of four disconnected countries with their clusters (Enumeration Areas) in the four phases of DHS that we consider, with boundaries of 37 districts in the four sSA countries.

Figure 3

Illustration of the four interaction types.

Figure 4

Standardized Mortality Rate and mean under-five mortalities per 1,000 live births and number surveyed (cases) of under-five children for each of the districts across the study period in four sSA countries.

Figure 5

Spatial and temporal effects in the U5M rate across districts of four sSA countries (A) mean posterior log odds of U5M for districts (B) provides the common posterior temporal mean log odds (trend) of U5M for the districts of each country.

Figure 6

Posterior log odds of U5M by districts over the study period: darker red (high risk) and darker green (lower risk).