Increasing age and duration of sex work among female sex workers in South Africa and implications for HIV incidence estimation: Bayesian evidence synthesis and simulation exercise

Abstract

Introduction: In sub-Saharan Africa, accurate estimates of the HIV epidemic in female sex workers are crucial for effective prevention and care strategies. These estimates are typically derived from mathematical models that assume certain demographic and behavioural characteristics like age and duration of sex work to remain constant over time. We reviewed this assumption for female sex workers in South Africa.

Methods: We reviewed studies that reported estimates on either the age or the duration of sex work among female sex workers in South Africa. We used Bayesian hierarchical models to synthesize reported estimates and to study time trends. In a simulation exercise, we also investigated the potential impact of the "constant age and sex work duration"-assumption on estimates of HIV incidence.

Results: We included 24 different studies, conducted between 1996 and 2019, contributing 42 estimates on female sex worker age and 27 estimates on sex work duration. There was evidence suggesting an increase in both the duration of sex work and the age of female sex workers over time. According to the fitted models, over each decade the expected duration of sex work increased by 55.6% (95%-credible interval [CrI]: 23.5%-93.9%) and the expected age of female sex workers increased by 14.3% (95%-CrI: 9.1%-19.1%). Over the 23-year period, the predicted mean duration of sex work increased from 2.7 years in 1996 to 7.4 years in 2019, while the predicted mean age increased from 26.4 years to 32.3 years. Allowing for these time trends in the simulation exercise resulted in a notable decline in estimated HIV incidence rate among sex workers over time. This decline was significantly more pronounced than when assuming a constant age and duration of sex work.

Conclusions: In South Africa, age and duration of sex work in female sex workers increased over time. While this trend might be influenced by factors like expanding community mobilization and improved rights advocacy, the ongoing criminalisation, stigmatisation of sex work and lack of alternative employment opportunities could also be contributing. It is important to account for these changes when estimating HIV indicators in female sex workers.

Keywords: Female sex workers; HIV; Mathematical modelling; South Africa.

Fig. 1

(A) Estimated time trends in the expected duration of sex work (SW) from 1996 to 2019 and (B) corresponding time trends in the rate parameter $\lambda$ of the underlying exponential distribution we assumed for individual durations of SW. The solid lines reflect medians of posterior predictive distributions, while shaded areas correspond to 95% credible intervals (less transparent ones exclude random-effect variability while more transparent ones include it). The estimates and 95% credible intervals for each unique study population are shown in gray. Black dots correspond to the input data (adjusted mean duration of SW) with areas proportional to the study population sizes.

Fig. 2

Estimated time trends in the mean (A) and standard deviation (B) age of female sex workers (FSW) from 1996 to 2019, and (C) corresponding time trends in the shape parameter $\alpha$ and rate parameter $\beta$ of the underlying Gamma distribution we assumed for individual ages of FSW. The solid lines reflect medians of posterior predictive distributions, while shaded areas correspond to 95% credible intervals (less transparent ones exclude random-effect variability while more transparent ones include it). Estimates and 95% credible intervals for each unique study population are shown in gray. Black dots correspond to the input data (adjusted mean and standard deviation of FSW ages) with areas proportional to the study population sizes.

Fig. 3

Comparison of estimated time trends in the mean ages of female sex workers (FSW) during sex work (SW, main analysis) with estimated trends in the mean FSW ages at entry into SW (sensitivity analysis). Solid lines reflect the estimated medians of the posterior predictive distributions, while shaded areas correspond to 95% credible intervals (less transparent ones exclude random-effect variability while more transparent ones include it). For the gray area, no data was available and estimates were extrapolated. Estimates and 95% credible intervals for each unique study population obtained in the sensitivity analysis are shown in purple. The original data points (as used in the main analysis) are shown as red circles (data points included in sensitivity analyses) and crosses (data points not included in sensitivity analysis since the duration of SW was not reported for that study population).

Fig. 4

Estimated time trends in the HIV incidence rate $\rho$ from 1996 to 2019, assuming constant age of female sex workers (FSW) and duration of sex work (SW) (gray) or changing FSW age and duration of SW as estimated in the main analysis (color). (A) shows the results under the assumptions that the prevalence at entry into SW equals the prevalence of the general female population in the same age group. (B) shows results under the assumption that the prevalence at entry into SW is increased by a factor $r$ ($r$ = 1.25, 1.5, 1.75) compared to the general female population. Solid lines reflect medians of estimated posterior predictive distributions, while shaded areas correspond to 95% credible intervals (excluding random-effects variability).