Grant: Doctoral Award - Frederick Banting and Charles Best Canada Graduate Scholarships

BACKGROUND: Every year, over half a million people become infected with HIV in Southern Africa, with key populations like female sex workers (FSW) and men who have sex with men (MSM) bearing the heaviest burden. Mathematical modelling of epidemics plays an important role in policy response, from helping to understand dynamics of epidemics, to finding optimal intervention strategies (e.g. increase condom use or increase treatment). Such models are often complex, and need to be solved using code. One important question is just how complex they need to be, since recent work has shown that our interpretation of the results can be different depending on how complex the model is - for example, which intervention turns out to be "optimal". Is it important to include key populations like FSW and MSM in the model? What about differences among FSW and among MSM? What about other characteristics like age, race, and geography? OBJECTIVES: In this project, we hope to clarify which population characteristics it is important to include in models of HIV transmission, and in which contexts. Since inclusion of these characteristics requires data about the population, we will also clarify exactly which pieces of data are needed to build a good model. Finally, when such data are unavailable, as they sometimes are, models may be limited in their ability to simulate an epidemic accurately; therefore, we will also describe potential biases in the simulated epidemics when certain data are not included, and what to do about them. METHODS: First, we will create a "gold-standard" model, and use detailed data from Southern African countries to ensure it simulates historical epidemic features accurately. Next, we will systematically remove modelled characteristics (like age), or change how they are represented, one-at-a-time, and then in combination, to see how the simulated epidemic changes. We will also compare the impact of these changes on which intervention is predicted to be optimal.