Spatio-Temporal Modelling Spatially Aggregated Data, In case of Malaria in Southern Ethiopia

Abstract

Malaria is a major public health concern worldwide, particularly in Sub-Saharan Africa. Ethiopia accounts for 1.7% of cases and 1.5% of global deaths, with seasonal and unstable transmission patterns. The COVID-19 pandemic has increased morbidity, making early de- tection crucial for mitigating the negative effects of malaria. Thus disease risk mapping is important to identify areas with elevated risks to assist police decisions. However, in such areas, the data sets are mostly available in aggregated form. Spatially aggregated data is often expressed as disease cases or average measurements from districts, often focused on administrative convenience rather than knowledge of the aetiology of the disease. When the underlying results in the process of disease outcomes are thought to be spatially con- tinuous, a spatially continuous model should be discovered rather than typically employed spatially discrete models. This thesis is composed of three papers; in the first paper, We have developed an early warning system for malaria that provides a signal whenever the in- cidence of malaria exceeds certain thresholds. Then, we identified areas with some elevated risks based on temporal trends. Our findings could assist public health offices in identifying areas for early intervention and guiding healthcare resource allocation, allowing its limited resources to be utilized to the greatest extent possible. In addition, we explored the sig- nificance of incorporating environmental variables in forecasting malaria risk in Southern Ethiopia. In the second paper, the researcher examined a spatiotemporal model to highlight disease risk change over time. Four models were considered to be the most appropriate for the problem of spatially aggregated data in a small area. According to the findings of the vii study, the spatiotemporal spatially discrete approximation to the log-Gaussian Cox process (SPSDLGCP) gives credible and computationally efficient estimates of disease risk on both spatially continuous and aggregated scales. Furthermore, we have discovered that including spatial correlation is critical when modelling the spatiotemporal variation of disease risk in the region. In the third paper, taking into account areal units as the vertices of a graph and neighbour interactions as a collection of edges, we proposed a graph-based optimization approach that could be used to estimate either a static or a temporally changing neighbour- hood matrix to better capture the spatial correlation in the data set. When compared to the commonly utilized border-sharing rule, the strategy yielded better inference. In particular, using a temporal varying waiting matrix for modelling the aggregated count of two domi- nating Plasmodium species; P. falciparum and P. vivax ; a clearer picture of the distribution of the incidence in the region was provided. We have also highlighted regions where either species poses unusual threats. Finally, we identified major climate variables linked with malaria risk in the region, used a spline function to integrate the non-linear relationship between climatic factors and malaria risk, and investigated the delayed climatic impact on malaria risk. From August 2013 to June 2019, we considered aggregated malaria counts in 149 districts located in Southern Ethiopia. The findings of our study have the potential to assist the SNNPRS public health department and other stakeholders in defining areas for early intervention as well as regulating the distribution of limited resources for the healthcare facility to the greatest extent possible. Furthermore, by looking for a more toiled approach for estimating the waiting matrix for multivariate cases, including additional risk factors that were not identified, risk discontinuities, and the problem of zero inflation, one can improve the existing methods.