Table 1. Comparison of EPP/Spectrum and Thembisa: Model inputs and assumptions for adult HIV incidence estimation.

	EPP/Spectrum	Thembisa
Data sources used in calibration	HIV prevalence data from ANCs from 1990–2012 are included in the model to develop a prevalence curve over time. Results from the 2005, 2008, 2012 HSRC surveys are included to calibrate the ANC results and inform trends.	Age-specific HIV prevalence data from 1991–2011 ANC surveys, age- and sex-specific prevalence data from 2005, 2008 and 2012 HSRC surveys, self-reported HIV testing data from the same three HSRC surveys, reported death data by age and sex (1997–2010).
Uncertainty analysis	Uncertainty in EPP is quantified using a Bayesian approach, with prior distributions for each of the EPP parameters, a likelihood function based on the listed data sources, and a posterior distribution simulated using IMIS. Uncertainty in Spectrum is calculated by 1000 runs, whereby each run randomly selects an item from the posterior incidence estimate and a set of parameters from the possible ranges of those parameters.	Uncertainty is quantified using a Bayesian approach. Prior distributions are specified to represent ranges of uncertainty around the sexual behaviour, HIV survival, HIV transmission and HCT parameters. A likelihood function is specified based on the data sources listed. The posterior distribution is simulated using IMIS.
Sexual behaviour	Sexual behaviour modelled in EPP by dividing sexually active population into high risk and “not at risk” populations. HIV incidence estimates from EPP are entered into Spectrum, so there are no sexual behaviour assumptions in Spectrum.	Model includes assumptions about % in high risk group, sexual debut, non-marital sex, marriage, divorce, widowhood, commercial sex, mixing between age groups and risk groups, coital frequency, effect of HIV status knowledge and ART on sexual behaviour.
HIV transmission	ART reduces probability of transmission by 70%.	Transmission probability depends on HIV disease stage, type of relationship, age and risk groups of both partners (implicit allowance for STI cofactors). ART reduces HIV transmission probability by 80%.
Untreated HIV survival	HIV-infected adults progress through a 7-stage model of CD4 decline, with HIV-related mortality occurring in all HIV states. Rates of CD4 decline increase with age.	After an initial acute phase, HIV-infected adults progress through a 4-stage model of CD4 decline, with HIV-related mortality occurring in the final two HIV states (CD4 200–349 and CD4 <200). Rates of CD4 decline increase with age.
Condom usage	Not modelled, though there is implicit allowance in EPP for changes in sexual risk behaviour over time.	Probability of condom use depends on age, sex and relationship type. Condom usage in HIV-negative individuals increases over time but reduces after 2008. Condom usage in HIV-positive individuals increases after diagnosis.
Male circumcision	Not modelled.	Model allows for traditional as well as medical male circumcision. Probability of female-to-male transmission is reduced by 60% if male is circumcised.
ART	Coverage estimates based on estimates of total numbers receiving ART in each year. Eligibility criteria change over time, with CD4 350 criterion being applied from mid-2011. Survival assumptions are based on data from IeDEA collaboration (mortality specified separately for 0–6 months, 6–12 months and >12 months after ART initiation).	Coverage estimates based on published estimates of total numbers starting ART in each year, combining public sector data and private sector data. Eligibility criteria change over time, with CD4 350 criterion being applied from mid-2011. Survival assumptions are based on models fitted to data from IeDEA Southern Africa (mortality specified separately for each of first five years of ART).

Abbreviations: EPP: Estimates and Projection Package; ART: antiretroviral therapy.