Origin of the Thembisa model
Many mathematical models have been developed to describe different aspects of the South African HIV epidemic. Thembisa was developed in an effort to synthesize four previously-developed models, each of which has different strengths and limitations:
- The Actuarial Society of South Africa (ASSA) AIDS and Demographic model: This model was developed to evaluate the demographic impact of HIV in South Africa. Although one of the most detailed demographic models available, it has not been updated since 2011, and thus does not reflect many of the advances that have been made in the field of HIV prevention and treatment.
- The STI-HIV Interaction model: This model was developed to assess the role of different sexual risk behaviours in driving the transmission of HIV and other sexually transmitted infections (STIs) in South Africa. The model was also developed to assess the extent to which other STIs have promoted HIV transmission in South Africa. As with the ASSA model, the model has not been updated to reflect a number of recent advances in HIV prevention. In addition, it was not designed to be a demographic model.
- The UCT Paediatric HIV model: This model was developed to assess new strategies for the prevention of mother-to-child transmission in South Africa, as well as new strategies for diagnosing and treating paediatric HIV. A limitation of the model is that it is not integrated into a model of adult HIV transmission, which means that the effect of adult interventions on mother-to-child transmission cannot be evaluated dynamically.
- The National Strategic Plan ART Need model: This model was developed to forecast the number of people who would need antiretroviral treatment (ART), for South Africa’s 2012-2016 National Strategic Plan (NSP). A key advance over the ASSA and STI-HIV Interaction models is that it allows for ART initiation in earlier stages of disease and is based on a CD4 staging system rather than a clinical staging system. However, the model does not dynamically model HIV transmission and is not a demographic model.
Thembisa incorporates many of the elements of these previously-published models. In addition it also includes a number of new features not included in any of the previous models (for example, modelling of male circumcision and pre-exposure prophylaxis).
Versions of the Thembisa model
Two versions of the Thembisa model have been developed: a model programmed in C++ and a model programmed in Excel and Visual Basic for Applications (VBA). The Excel/VBA model can be downloaded for free (see Downloads). The Excel/VBA version has the advantage of being easy to use (especially for individuals without computer programming experience) but is slow to run. The C++ model is much faster to run and is therefore used for the purpose of uncertainty analysis and calculation of 95% confidence intervals around model outputs. However, the C++ model is not user-friendly in its current form, and is therefore not yet publicly available.
The Thembisa model has been applied to South Africa as a whole as well as each of the nine provinces. The default assumptions in the model relate to the country as a whole, but a file with province-specific assumptions can be downloaded and used to generate province-specific results.
A number of versions of the Thembisa model have been published:
- Thembisa version 1.0 was published in February of 2014. This model was only applied to the country as a whole, and was not made publicly available due to concern about the preliminary nature of the demographic assumptions. However, a report was published containing a detailed model description and model outputs.
- Thembisa version 2.4 was published in May 2016. This version includes several new features (e.g. rates of viral suppression, treatment interruptions, and the effect of ART duration during pregnancy on the mother-to-child transmission risk).
- Thembisa version 2.5 was published in August 2016. This includes updated demographic assumptions and province-specific calibrations.
- Thembisa version 3.2 was published in September 2017. This version includes HIV transmission between men who have sex with men and a number of other new features (birth PCR testing and changes in HIV virulence over time).
- Thembisa version 4.1 was published in August 2018. This version includes greater detail on HIV diagnosis in children.
- Thembisa version 4.2 was published in June 2019. This version is the first version to be calibrated to recorded mortality data in children.
- Thembisa version 4.3 was published in June 2020. This version includes several important revisions to assumptions about fertility rates in HIV-positive women. Changes in provincial calibration procedures also led to substantial increases in estimates of HIV prevalence in Northern Cape and Limpopo provinces.
- Thembisa version 4.4 was published in March 2021 as a 'lite' update to version 4.3 (mainly incorporating updated HIV programme data from 2019-20).
- Thembisa version 4.5 was published in April 2022. It included self-testing and index testing, and the model was revised to allow for higher rates of ART linkage after re-diagnosis. Demographic parameters were updated to reflect the impact of COVID-19.
Dr Leigh Johnson is an actuary and epidemiologist, based at the Centre for Infectious Disease Epidemiology and Research, at the University of Cape Town.
Professor Rob Dorrington is an actuary and demographer, based at the Centre for Actuarial Research, at the University of Cape Town.
The following organizations have funded the development of the Thembisa model:
- The Bill and Melinda Gates Foundation
- South African National AIDS Council (SANAC)
- Hasso Plattner Foundation
- National Institutes of Health (grant 1R01AI094586-01)