Background: HPTN071(PopART) is a three-arm cluster-randomized trial in 12 communities in Zambia and 9 in South Africa, measuring the impact of a combination prevention intervention including universal testing and treatment (UTT) on population-level HIV incidence. Data were collected on the baseline characteristics, and uptake of the intervention. Mathematical modelling informed the trial planning and was developed into an efficient stochastic individual-based simulation model (IBM) of heterosexual HIV transmission in generalised epidemics. Gains in algorithmic efficiency mean that IBM of HIV can be used in fitting to trial data and understanding which determinants are expected to influence the trial outcome.
Methods: Model output was calibrated to historical prevalence and community-level UTT data, identifying the most likely parameter values. Stochastic IBM uncertainty and heterogeneities across communities were assessed by running simulations repeatedly with community-specific parameters. Sensitivity analyses identified the most important determinants of predicted trial impact and long-term effectiveness.
Results: The IBM adequately replicates heterogeneities across sex and age-groups in the trial data. The mean predicted reduction in incidence in the intervention arm was 39% (range 17% to 52%) compared to control communities. The most important determinant of outcome was the uptake of intervention. Conditioning on this, increasing the proportion of infections from individuals in the acute phase, and increasing the number of infections from outside a trial community (both from 10% to 20%) causes on average a 4% decrease in the trial impact. The parameterisation of the sexual contact network explains differences of up to 2%. On average the trial outcome varies by 3% across communities, attributable to heterogeneities in demography, baseline HIV prevalence, and linkage to care. Differences between Zambia and South Africa can be explained by country-specific differences in the proportion of partners from outside a trial community and the baseline prevalence of male circumcision.
Conclusions: The choice of IBM parameters (some obvious and some less so) has a substantial effect on the predicted trial outcome. Detailed modelling of the trial allows identification of determinants of effectiveness, that will in turn improve long-term modelling projections, offering insight into developments far beyond the trial horizon, relevant for decisions to be taken after trial completion.

Download the e-Poster