Epidemiological Modelling for Eswatini HIV programme planning
The Challenge
Eswatini has faced one of the most severe HIV epidemics in the world. By 2016, the country had the highest HIV prevalence globally among adults aged 15 to 49, reaching 27.2%.
Over the past 15 years, Eswatini made significant strides in addressing the HIV crisis, and in 2023, Eswatini was the first African country to achieve the UNAIDS 95-95-95 targets, with 94% of adults living with HIV aware of their status, 97% of those individuals on antiretroviral therapy (ART), and 96% of those on ART achieving viral suppression.
Despite the above, there is still a need for further reduction in HIV incidence. In particular, adolescent girls and young women (AGYW) acquire new HIV infections at a very high rate, and gaps remain in achieving viral suppression (VS) among men aged 25-34 years.
With many behavioural and biomedical HIV prevention methods available, the challenge is to provide the Eswatini Ministry of Health with guidance on the optimal combination of interventions for maximal impact.
Our Approach
Under the leadership of the World Health Organization, we joined an alliance of epidemiological modelling experts and developed a detailed model, designed to simulate how different choices in the expansion of the Eswatini national HIV prevention and treatment programme would lead to different health outcomes and budgetary consequences.
Our model tracked the past and planned rollout of Voluntary Medical Male Circumcision (VMMC); Pre-Exposure Prophylaxis (PrEP) via oral treatment, injectables and dapivirine vaginal ring; HIV prevention programmes targeted to AGYW; and interventions focused on the increased uptake of various HIV testing modalities.
To calibrate our model to country-specific data, we first identified the model parameters that were most influential for the sexual behaviour and therapeutic dynamics of the model population.
Next, we used an Approximate Bayesian Computation algorithm to tune these parameters such that the model output aligned with the demographic, epidemiological, and programme data (target features) from Eswatini.
Once the model was calibrated, we produced projections of the expected impact (averted HIV infections and deaths) and associated cost over the next 15 years, for multiple scenarios of the national HIV prevention and treatment programme.
Case and Point
When we introduced Simpact, our tool for modelling HIV epidemics in 2018, we applied it to the HIV epidemic in Eswatini.
At the time, we had data up to 2017, and we estimated that in the age group between 15 and 50 years old, per 1000 people, 13 would get newly infected with HIV in the year 2017. Importantly, our model predicted that the rate of new infections would drop by about 40%, to 8 per 1000 people in 2021.
Now, looking back, we have actual measurements from the Swaziland HIV Incidence Measurement Survey (SHIMS) that was done in 2021. And that study confirmed that that we were right. The HIV incidence in the age group between 15 and 50 years old was 7.7 per 1000 people in the SHIMS study.
.png)
What We Learned
A recent Executive Order on Reevaluating and Realigning United States Foreign Aid has increased the importance and urgency for African countries, including Eswatini, to develop and deploy regional and local expertise in planning, monitoring and evaluating HIV prevention and treatment programmes.
However, the art and science of evidence-based policy making in public health is challenged by the multitude of factors that must be taken into account when developing and adapting public health policies. Besides the question of efficacy, other important considerations include real-world effectiveness, budget impact, equity, cost-effectiveness and sustainability.
Epidemiological modelling provides a unifying framework in which many “what if” questions can be answered simultaneously in silico, i.e in a virtual environment, thus saving precious time and financial resources, and bypassing potential ethical dilemmas.
“Building on the aphorism that all models are wrong, but some are useful, it is valuable to consider that models are not only useful when they are able to produce a clear and precise answer. A model that leaves a lot of residual uncertainty, expressed as wide prediction intervals, can be just as useful, as it prompts and gives direction to further biomedical, behavioural and health economic research to resolve the knowledge gap.“
