Modelling Tuberculosis in Ethiopia: Spatiotemporal Transmission Dynamics and Effects of Public Health Interventions

Abstract

Tuberculosis (TB) is now the world’s leading infectious killer with an estimated 10million cases and 1.6 million deaths in 2016. A small number of countries bear the majority of the burden of disease, with two-thirds of cases occurring in only seven countries. TB transmission occurs in both households and the local community, leading to focal disease hotspots which perpetuate TB spread within and across community groups. Integrating spatial analysis with mathematical transmission dynamic models can help in evaluating the role of these hotspots in the spread of TB and understanding the potential impact of geographically targeted interventions.

The first part of this thesis evaluates whether TB exhibits spatial heterogeneity in rural and remote regions of Ethiopia using data from all TB patients treated in a remote administrative region of the country. This study demonstrated considerable spatial heterogeneity in TB distribution in this resource-limited setting. However, most of these heterogeneities were accounted for by health facility availability, implying differential case detection between areas with better and poorer access to health care. Thus, this Chapter cautions that spatial analysis of TB and the identification of geographical hotspots using programmatic data alone can be misleading, as it may be strongly influenced by undetected cases, which is in turn dependent on local programmatic performance.

Building on the findings outlined above, Chapter three presents a systematic review of methods used in published spatial analyses of TB. From this review, this Chapter elucidates limitations in the current approaches to spatial analysis of TB. Of particular importance is the consistent failure to account for unreported or undetected cases, despite notification data being used in 95 percent of the reviewed studies. The Chapter also describes methodological flaws to many of the studies, in particular the use of conventional regression analysis to draw spatial conclusions. In addition, most spatial analyses of TB distribution used residential information to define the location of patients, which potentially understates the importance of other community settings, despite more than 80% of all transmission events occurring outside households.

The study in Chapter 4 proposes a method to address the limitations outlined in the previous chapters – particularly the lack of methods to account for undetected cases. The model estimates both incidence and case detection rates simultaneously across space and time, providing a useful platform for regularly tracking spatial patterns and temporal trends. In addition, this technique is general and can be applied to any disease in any setting. Applied to the Ethiopian setting, this model identifies previously unrecognized areas of high TB burden in locations with no available health care facilities. With the aim of quantifying the role of TB hotspots in community transmission as well as evaluating the potential impact of targeting spatial hotspots, Chapter 5 utilises incidence data generated by the novel method described above to identify spatial TB hotspots. At this point, the thesis constructs spatially structured mathematical models and quantifies the extent to which these hotspots account for the spatial spread of TB. Findings from this work suggest that TB transmission in the same study region in rural Ethiopia is localised and the role of spatial hotspots in the spatial spread is limited, although their impact is considerable in adjacent locations due to very high relative incidence in the hotspot compared to the other regions. Finally, Chapter 6 uses the same model introduced in Chapter 5 to evaluate the impact of various TB intervention strategies before concluding the thesis.

Overall, this thesis advances current approaches to spatial analysis and provides a means to account for the problem of undetected cases. It also provides a platform to estimate both incidence and case detection rate simultaneously, and hence could provide as an alternative approach to the spatial interpretation of TB epidemiology. This is of particular importance to high endemic settings, where considerable number of TB cases are missed, and case notification is biased to areas with better access to health care. Importantly, the study also concludes that the impact of spatial TB hotspots on the spatial spread of disease in remote regions of Ethiopia is limited and transmission is predominantly locally driven. Hence, interventions strategies that are spatially targeted may not achieve anticipated outcomes, although the overall effect of these interventions remains considerable due to extremely high incidence in the hotspot regions.