Do standard model assumptions realistically represent HIV dynamics in sex workers? A modelling analysis of South African data

Female sex workers (FSW) in sub-Saharan Africa experience disproportionately high risks of HIV infection. Mathematical models are widely used to assess the contribution of sex workers and other key populations to HIV transmission dynamics and to inform targeted programmes. However, many rely on simplifying assumptions, such as stable sex worker characteristics and constant HIV transmission risk over time. These assumptions may be unrealistic and could bias modelled estimates. We used the South African Thembisa model to assess how alternative assumptions about FSW age, duration of sex work, and client-to-FSW transmission risk affect modelled HIV outcomes. We compared six scenarios that combined constant and increasing FSW age and sex work duration with constant and early-epidemic declining (exponentially or exposure-dependent) transmission risk. Each scenario was calibrated to HIV prevalence data from population-based and sex worker-specific surveys. Scenarios that allowed both FSW characteristics and transmission risk to vary over time showed the best agreement with external data, most closely reproducing HIV incidence, prevalence, and viral suppression estimates from a 2019 national sex worker survey (incidence [~]5 per 100 person-years, prevalence 61-62%, viral suppression [~]60%), and producing incidence rate ratios more consistent with estimates from the broader eastern and southern Africa region. By contrast, the scenario assuming constant FSW characteristics and transmission risk overestimated HIV incidence and underestimated prevalence and viral suppression. At the same time, this time-invariant specification attributed a much larger share of new HIV infections to sex work, with commercial sex work accounting for more than 20% of new infections in 2025, compared with 9-13% under time-varying assumptions. Overall, our findings show that HIV model estimates for sex workers are highly sensitive to modelling assumptions. Incorporating time-varying FSW parameters yields estimates that are more consistent with empirical data and support more reliable programme planning and evaluation. Author SummaryFemale sex workers in sub-Saharan Africa face much higher risks of HIV infection than other women. Mathematical models are often used to understand why and to guide prevention programmes. Yet many of these models make simple assumptions about sex workers - for example, that their average age stays the same over time, that they spend a fixed number of years in sex work, or that the chance of HIV passing from a client to a sex worker never changes. In reality, these factors changed over time. In this study, we used South Africas national HIV model to test how changing these assumptions affects the results. We compared different versions of the model and checked which ones best matched national sex worker survey data. We found that the model worked better when we allowed sex workers to become older over time, to spend longer in sex work, and the risk of passing on HIV to decline. Our findings show that mathematical models can give very different answers depending on how they represent the lives and experiences of sex workers. More realistic assumptions lead to more accurate estimates and can help ensure that programmes focus support where it is most needed.