Frequently asked questions

 

 

How does Thembisa differ from Goals, Spectrum and EPP?

 

Goals, Spectrum and EPP are models that can be applied in any HIV epidemic setting. Thembisa is a model that has been developed specifically for South Africa, and no attempts have been made to apply the model to settings outside of South Africa. South Africa is a data-rich country, which means that it is possible to develop much more detailed models of the South African HIV epidemic than would be possible in many other African countries.

 

Goals, Spectrum and EPP are part of a suite of models, each of which serves a different function. EPP is used to obtain estimates of HIV incidence rates from HIV prevalence data, while Spectrum uses these incidence estimates to evaluate the demographic impact of HIV, and Goals is used to evaluate the impact of HIV interventions. The Thembisa model integrates these different functions into a single model.

 

The model-fitting procedures in EPP and Thembisa both aim to identify the model parameters that yield HIV prevalence estimates consistent with survey data. However, the Thembisa model fitting procedure works with age-specific HIV prevalence data, while the EPP model fitting procedure uses aggregated HIV prevalence estimates for the 15-49 age group. In addition, the Thembisa model is calibrated to mortality data.

 

The demographic components of the Spectrum and Thembisa models are similar. Both stratify the population by age and sex. Both stratify the HIV-positive population by CD4 count and antiretroviral treatment (ART) status.

 

The intervention components of the Goals and Thembisa model differ in several respects. Goals includes a number of sub-populations that are not currently in the Thembisa model (e.g. people who inject drugs (PWID)), and can thus evaluate the impact of interventions specific to PWID. On the other hand, Goals is not an age-structured model, and is therefore not appropriate for considering age-specific interventions. Cross-sectional measures of coverage tend to be specified as inputs in the Goals model, whereas in the Thembisa model, coverage is usually an output and the inputs are specified either as the rate of engaging in a particular HIV service or the number of individuals engaging in a particular service for the first time. The Goals model comes with a built-in ‘impact matrix’ for determining the impact of individual interventions on key epidemiological parameters, while the Thembisa model requires that users specify the key epidemiological parameters directly, taking into account the expected impact of interventions.

 

How do the Thembisa estimates differ from the UNAIDS estimates for South Africa?

 

Since 2017, the UNAIDS estimates for South Africa have been based on the Thembisa model. Although UNAIDS still uses the Spectrum model to produce its estimates for South Africa, the Spectrum model has been adjusted to match the Thembisa model as closely as possible. This involves using the same annual HIV incidence rates as estimated by Thembisa, aligning other inputs (such as numbers of patients on ART) to be consistent with Thembisa, and making various other adjustments to yield similar outputs to Thembisa. The reason for creating a Spectrum version of Thembisa is that there are certain UNAIDS indicators (for example, numbers of AIDS orphans) that are estimated by Spectrum but not by Thembisa.

 

Although the UNAIDS estimates for South Africa are very similar to the Thembisa estimates, there are some minor differences. These differences are particularly notable in the case of the paediatric HIV estimates, and the Thembisa developers are currently engaging with the UNAIDS modelling team to better understand these differences and achieve greater alignment.

 

How does Thembisa differ from the ASSA model?

 

Thembisa inherits a number of features from the Actuarial Society of South Africa (ASSA) AIDS and Demographic model, especially in the demographic structure of the model. The key differences are:

• The Thembisa model simulates a much wider range of interventions than the ASSA model (most importantly, male circumcision, pre-exposure prophylaxis (PrEP) and WHO option B+ for preventing mother-to-child transmission).
• The Thembisa model allows for ART initiation at any stage in the course of HIV disease, unlike the ASSA model, which allows for ART initiation only in advanced disease (i.e. at the time of progressing to AIDS).
• The Thembisa model is based on a CD4 staging system (which makes it appropriate for evaluating the impact of changes in ART initiation criteria), while the ASSA model is based on a clinical staging system.
• The Thembisa model works with monthly time steps, while the ASSA model works with annual time steps. The Thembisa model is therefore considerably slower to run.
• Sexual behaviour risk groups are defined differently in the two models. The ASSA model divides the population into four risk groups and assumes no movement between these risk groups over time. The Thembisa model divides the population into two broad risk groups (high risk and low risk, with no movement between these groups over time) and further sub-divides these groups into sub-groups that are defined in terms of marital status and commercial sex, allowing for movement between these sub-groups over time.

 

What are the limitations of Thembisa?

 

The Thembisa model simulates heterosexual transmission, transmission between men who have sex with men, and mother-to-child transmission, but it does not consider PWID as a separate risk group. This means that interventions that are specific to PWID cannot be evaluated by Thembisa in its current form.

 

The modelling of ART interruptions and ART discontinuation is currently quite simplistic. ART discontinuation and ART resumption are not modelled dynamically; instead the model works with an assumed fraction of ART-experienced adults who are actively on treatment at each treatment duration. Research is currently being conducted to estimate rates of ART discontinuation and resumption in South Africa, and will be used to improve the model in future.

 

The Thembisa model is a frequency-dependent model. This means that the rate at which an individual acquires HIV depends on the HIV prevalence in the pool of individuals with whom that individual could potentially partner. Individuals are not assigned specific sexual partners, and this means that the model cannot evaluate couple-based HIV prevention strategies (e.g. PrEP for discordant couples and couple-based HIV testing). It also means that the modelling of HIV transmission is not fully realistic.

 

Thembisa is a compartmental model which divides the population into discrete risk groups. In reality, sexual behaviour is too diverse to be ‘compartmentalized’. This means that the model of sexual behaviour is simplistic, and it may not adequately reflect the true level of heterogeneity in risk behaviour that exists in reality.

 

How should I cite the outputs of the Thembisa model?

 

If you are citing the national model, please cite the following journal article:

Johnson LF, May MT, Dorrington RE, Cornell M, Boulle A, Egger M and Davies MA. (2017) Estimating the impact of antiretroviral treatment on adult mortality trends in South Africa: a mathematical modelling study. PLoS Medicine. 14(12): e1002468.

 

If you are citing the provincial model, please cite the following journal article:

Johnson LF, Dorrington RE, Moolla H. (2017) Progress towards the 2020 targets for HIV diagnosis and antiretroviral treatment in South Africa. Southern African Journal of HIV Medicine.18(1): a694

 

In some cases it may be more appropriate to cite other papers or reports (see the Publications page), depending on the context.