In this dissertation, I used various methods to model the transmission of two infectious diseases, Ebola in an epidemic setting and GU Chlamydia in an endemic setting within in Sub-Saharan Africa. Since 2015, there have been five outbreaks of Ebola Virus Disease (EVD) in several dierent countries in Sub-Saharan Africa, one of which became the second largest EVD outbreak in history in the setting of a longstanding conflict zone. It is suspected that after violent events occur, EVD transmission will increase; however, empirical studies to understand the impact of violence on transmission are lacking. In my first chapter, I used spatial and temporal trends of EVD case counts to compare transmission rates between health zones that have versus have not experienced recent violent events during the outbreak. In my second chapter, I also sought to use modeling to make outbreak projections, looking at the 2020 outbreak in The Democratic Republic of Congo. I made short and long-term projections for the outbreak in an eort to assess the potential to provide more accurate forecasting for an ongoing outbreak. I also evaluated how the outbreak’s timing and course affected the accuracy of such forecasts. Lastly in my third chapter, I focused on trachoma endemic areas of Sub-Saharan Africa and modeling the impact of annual Trachoma Mass Drug Administration (MDA) with azithromycin upon the prevalence of genitourinary (GU) chlamydia using a compartmental model. Communities that are especially hard hit with Trachoma are almost exclusively poor commu- nities with poor access to sanitation, screening and antibiotics to treat the infection; conditions that may allow for STDs to maintain a high chain of transmission. The dosing of azithromycin for the Trachoma MDA is consistent with dosing given clinically to treat GU chlamydial (GUC) disease, and recent evidence has suggested it reduces the population prevalence.
In my first chapter investigating the potential impact of violent events upon local instability and increased EVD transmission, I collected daily EVD case counts from DRC Ministry of Health for the 2018 outbreak in the Democratic Republic of Congo (DRC). A time-varying indicator of recent violence in each health zone was derived from events documented in the WHO situation reports. I used the Wallinga-Teunis technique to estimate the reproduction number R for each case by day per zone in the 2018–2019 outbreak. I fit an exponentially decaying curve to estimates of R overall and by health zone, for comparison to past outbreaks.
As of 16 April 2019, the mean overall R for the entire outbreak was 1.11. I found evidence of an increase in the estimated transmission rates in health zones with recently reported violent events versus those without (p = 0.008). The average R was estimated as between 0.61 and 0.86 in regions not aected by recent violent events, and between 1.01 and 1.07 in zones aected by violent events within the previous 21 days, leading to an increase in R between 0.17 and 0.53. Within zones with recent violent events, the mean estimated quenching rate was lower than for all past outbreaks except the 2013–2016 West African outbreak. The difference in the estimated transmission rates between zones affected by recent violent events suggests that violent events contributed to increased transmission and the prolonged nature of the second largest EVD outbreak in history.
In my second chapter performing EVD outbreak projections, several mathematical models were used to predict the final outbreak size and weekly incidence for the 2020 DRC outbreak. Projections were commenced prospectively mid-way through the outbreak, and retrospectively applied for the early out- break. Short-term forecasts were made using two different models: (i) a particle-filter branching-process model and (ii) a naive auto-regression. Final outbreak size predictions were made using four different models: (i) the particle-filter branching-process model, (ii) Theil-Sen regression, (iii) Gott’s Law and (iv) a novel Bayesian branching process model parameterized using prior outbreak sizes and contingent on the current outbreak size. The Bayesian model examined final size distributions across a range of current outbreak sizes, allowing for an examination of parameter fits.Overall, there were reasonable amounts of variability in the forecasts created by different models. For short-term, auto-regression models showed relatively stable steady-state growth in the outbreak, with somewhat larger confidence intervals while the particle-filter branching model projected an outbreak slowly ending in the same period. Final outbreak size predictions increased overall as the outbreak continued. The median expectation among models increased between 2.5–4.0 fold in September over initial expectations from June as the outbreak grew from 34 to 128 cases. The branching-process model was overall the most stable consistent performer, though the Bayesian model was a close second. Including the West Africa outbreak, easily the largest to date, increased the range of predicted outcomes for the DRC outbreak between 40–50%.
In predicting the 2020 Ebola outbreak, the most consistent performing model was the branching process particle-filter model though the Bayesian model did nearly as well, despite being agnostic to the trajectory of the outbreak. Our short-term models consistently predicted the outbreak would grow, though models disagreed over the slowing pace; it will be important to evaluate the performance of these models in future outbreaks to understand these uncertainties. The growth of the outbreak to well over a hundred cases underscores the real risk EBOV poses to the region and the need for improved understanding of outbreak trajectories even with the presence of three approved vaccines.
In my third chapter, I analyzed the impact of Trachoma MDA upon GUChlamydia prevalence using an extended compartmental SIS model, accounting for the natural history of GUC, risk structure, and gender. The model includes slowly developing partial immunity. MDA was modelled as an impulsively forced treatment with varying coverage and efficacy.
My model showed that three years of MDA at current levels reduced the prevalence of GUC in all populations by at least 15%. Between annual MDA, the prevalence partially rebounded to pre-treatment levels. With Coverage x Efficacy ≥ 0.80, the time between MDA treatments was insufficient to sustain transmission, allowing for GUC burden to be suppressed below 1 in 10,000 after 5 rounds for starting prevalence less than 9.2%. When serial non-compliance is increased from 20% to 80%, this target is achieved for starting prevalences below 4.7%, down from 9.2%. Targeting azithromycin treatment only to high-risk individuals reduces the starting prevalences for which target is reached to 1.8%.
My model suggests that MDA could reduce the prevalence of GUC to less than 1 in 10,000 within 5 years time. This reinforces the suggestions of potential additional health benefits of trachoma MDA and points to potential value of screening and disease treatment even in impoverished areas, and suggests testable hypotheses regarding prevalence in endemic areas under treatment.
