We modified EMOD-HIV v2.5, an age-stratified and individual-based network model of HIV of South Africa, to incorporate HIV vaccination according to pox-protein HIV vaccine regimens (such as the regimen currently being tested in HVTN 702). Because EMOD is an individual-based model, interventions such as a time-varying course of vaccine efficacy can be applied to each individual according to his or her own timing of vaccination and adherence to the booster series. This renders the model well suited for a nuanced analysis of the anticipated time-dependent efficacy of the pox-protein HIV vaccine regimen.
The parameters, model input values, sources, projections, and sensitivities of the epidemic projection without vaccine, used as the reference group for comparison, have been described previously (Bershteyn et al.
2012,
2013; Klein et al.
2014; Selinger et al.
2019). A detailed model description, default parameters, user tutorials, model installer, and source code are available for download at
http://idmod.org/software. EMOD-HIV is an individual-based model that simulates transmission of HIV using an explicitly defined network of heterosexual relationships that are formed and dissolved according to age- and risk-dependent preference patterns (Klein
2012). The synthetic population was initiated in 1960, and population recruitment and mortality were assumed to be proportional following age- and gender-stratified fertility and mortality tables and projections from the 2012 UN World Population Prospects (UnitedNations
2016). Since the population size of South Africa exceeds the computational limit of simulated agents, we assumed that one simulated agent corresponds to 300 real-world individuals. The model was calibrated to match retrospective estimates of age- and gender-stratified, national-level prevalence, incidence, ART coverage and all-cause mortality from nationally representative HIV surveys in South Africa (Shisana et al.
2010,
2014; Eaton et al.
2015; Rehle et al.
2007; Shisana and Simbayi
2002) and demographic data (Statistics South Africa
2014) (see Figure 1 in Supplementary Materials for an overlay between model calibration outputs and source data). We also refer to a comparative calibration study comprising ten models (including EMOD) of HIV transmission in South Africa (see supplementary material in Eaton et al.
2015). For the purpose of this modeling study, we emphasized calibration of model parameters concerning risk assortativity during partnership formation, duration, and concurrency of partnerships as well as condom usage. For each simulated vaccine scenario, we used the 50 most likely parameter sets obtained from the gradient-descent-based calibration process (see Table 1 in Supplementary Material). The age patterns of sexual mixing were configured to match those observed in the rural, HIV-hyperendemic province of KwaZulu-Natal, South Africa (Ott et al.
2011). Recently, a validation study showed that self-reported partner ages in this setting are relatively accurate, with 72% of self-reported estimates falling within 2 years of the partners' actual date of birth (Harling et al.
2015). In addition, a recent field study on age-mixing patterns in Cape Town came to similar conclusions concerning the age gap between men and women (Beauclair et al.
2018), suggesting that the age-mixing data used in our model is not specific to KwaZulu-Natal but applies more generally to South Africa. Further, the transmission patterns observed in EMOD (Bershteyn et al.
2013) are consistent with those revealed in a recent phylogenetic analysis of the age/gender patterns of HIV transmission in this setting (de Oliveira et al.
2017).