Modeling to Inform the Delivery of HIV Pre-Exposure Prophylaxis in Sub-Saharan Africa

Daily oral tenofovir disoproxil fumarate and emtricitabine (TDF/FTC) as HIV pre-exposure prophylaxis (PrEP) is a safe and effective method for HIV prevention and offers potential to substantially reduce HIV incidence in sub-Saharan Africa. Mathematical models are commonly used to project the cost-effectiveness of investments in PrEP in comparison to alternative resource allocation strategies. Predictive modeling can also identify individuals at elevated risk
who may benefit most from PrEP. The studies contained in this dissertation address fundamental issues in estimating the cost and potential impact of PrEP implementation in sub-Saharan Africa.
First, we estimated the cost of routine PrEP delivery through maternal and child health (MCH) and family planning (FP) clinics in western Kenya (Chapter 1). PrEP delivery through MCH and FP leverages existing service delivery platforms that reach a large fraction of women at elevated HIV risk. Using data from over 20,000 PrEP encounters through 16 clinics, we estimated that the cost per client-month of PrEP dispensed to be $26.52 (2017 USD), with personnel (43%), drugs (25%), and laboratory testing (14%) accounting for the majority of costs. Postponing creatinine testing from PrEP initiation to the first follow-up visit could save 8% of total program costs. Under Ministry of Health implementation, we projected costs would decrease by 38%, but estimates were sensitive to changes in PrEP uptake and retention.
Second, we used an individual-based transmission model calibrated to Eswatini to evaluate the sensitivity of model projections of PrEP impact and efficiency to specification HIV exposure heterogeneity (Chapter 2). A common method for introducing HIV exposure heterogeneity into a model is to stratify the population into “risk group” categories with different average sexual behavior parameters, allowing PrEP coverage to vary by risk group without having to explicitly represent individual partnerships. We found that this specification leads to a sharp tradeoff between total impact and efficiency depending on PrEP coverage levels in each risk group. In comparison, PrEP use among the general population is projected to be two times more efficient if PrEP use is prioritized during partnerships and over six times more efficient if use is further prioritized among individuals with HIV-positive partners. In addition, large incidence reductions can be achieved at low levels of PrEP coverage if PrEP use in the general population is concentrated when HIV exposure is more likely, but high levels of PrEP coverage are needed if time-varying individual risk is ignored.
Third, we developed and validated HIV risk prediction models incorporating individual level and geospatial covariates using data from nearly 20,000 individuals in a population-based cohort in rural KwaZulu-Natal, South Africa (Chapter 3). Individual-level predictors included demographic, socioeconomic, and sexual behavior measures, while geospatial covariates included local estimates of community HIV prevalence and viral load. We compared full models to simpler models restricted to only individual-level covariates or only age and geospatial covariates. Models using only age group and geospatial covariates had similar performance (women: area under the receiver operating characteristic curve (AUROC) = 0.65, men: AUROC = 0.71) to the full models (women: AUROC = 0.68, men: AUROC = 0.72). In addition, geospatial models more accurately identified high incidence regions than individual-level models; the 20% of the study area with the highest predicted risk accounted for 60% of the high incidence areas when using geospatial models but only 13% using models with only individual-level covariates.
These findings have implications for PrEP policies. Our primary costing study identified service delivery bottlenecks and cost drivers that can inform efforts to streamline PrEP delivery. By ignoring the alignment of PrEP use with time-varying individual HIV exposure, models using a risk group specification may overestimate the cost and underestimate the impact of widespread PrEP availability. Finally, local estimates of HIV prevalence can help identify individuals and areas to prioritize for PrEP services to maximize impact.