Geostatistical methods and applications in global health

Sub-saharan Africa shares a high portion of the global disease burden and has attracted the attention of several intervention programmes. Intervention programmes need an in-depth understanding of the spatial and temporal distribution of diseases and the underlying risk factors in order to plan effective control strategies. Geostatistical methods provide a means to map disease outcomes whilst explaining measured and unmeasured underlying risk factors. This thesis, made up of three papers, focuses of developing and applying geostatistical methods to understand the spatial (and temporal) distributions and risk factors of childhood undernutrition, malaria and Loa loa in sub-Saharan African countries. The relationship between the rate of infectious mosquito bites and the prevalence of malaria parasite in human hosts can highlight aspects of malaria epidemiology that are pertinent to malaria control. However, this relationship is poorly understood. In our first paper, we develop geostatistical models to study the spatio-temporal distributions of Plasmodium falciparum parasite prevalence and the rate of infectious mosquito bites. We then highlight key aspects of the malaria epidemiology relevant for intervention policies by using mechanistic and empirical statistical models to explore the relationship between infectious bites and parasite prevalence in a rural community in Malawi. The question of whether or not malaria is associated with growth in children has been studied for years, with different studies reporting contradictory results. However, none of these studies used spatial statistical methods. In the second paper, we develop a geostatistical model to investigate this association using 20 Demographic and Health Survey datasets from 13 sub-Saharan African countries. We then propose novel extensions of the modelling strategy to growth and malaria data collected as a spatial longitudinal study. Disease prevalence data are often obtained using different diagnostics, but in the absence of spatial statistical methods to jointly analyse such data, most studies report the results of separate analyses on the data from each diagnostic. A joint analysis can explain possible correlations between different diagnostics, which can then be exploited to make more precise and more reliable predictions. In the third paper, we developed a geostatistical framework for combining prevalence data from different diagnostics and apply the novel methodology to map malaria in the highlands of Western Kenya and Loa loa in sub-Saharan Africa.