Projected outcomes of universal testing and treatment in a generalised HIV epidemic in Zambia and South Africa (the HPTN 071 [PopART] trial): a modelling study

Abstract

Background: The long-term impact of universal home-based testing and treatment as part of universal testing and treatment (UTT) on HIV incidence is unknown. We made projections using a detailed individual-based model of the effect of the intervention delivered in the HPTN 071 (PopART) cluster-randomised trial.

Methods: In this modelling study, we fitted an individual-based model to the HIV epidemic and HIV care cascade in 21 high prevalence communities in Zambia and South Africa that were part of the PopART cluster-randomised trial (intervention period Nov 1, 2013, to Dec 31, 2017). The model represents coverage of home-based testing and counselling by age and sex, delivered as part of the trial, antiretroviral therapy (ART) uptake, and any changes in national guidelines on ART eligibility. In PopART, communities were randomly assigned to one of three arms: arm A received the full PopART intervention for all individuals who tested positive for HIV, arm B received the intervention with ART provided in accordance with national guidelines, and arm C received standard of care. We fitted the model to trial data twice using Approximate Bayesian Computation, once before data unblinding and then again after data unblinding. We compared projections of intervention impact with observed effects, and for four different scenarios of UTT up to Jan 1, 2030 in the study communities.

Findings: Compared with standard of care, a 51% (95% credible interval 40-60) reduction in HIV incidence is projected if the trial intervention (arms A and B combined) is continued from 2020 to 2030, over and above a declining trend in HIV incidence under standard of care.

Interpretation: A widespread and continued commitment to UTT via home-based testing and counselling can have a substantial effect on HIV incidence in high prevalence communities.

Funding: National Institute of Allergy and Infectious Diseases, US President's Emergency Plan for AIDS Relief, International Initiative for Impact Evaluation, Bill & Melinda Gates Foundation, National Institute on Drug Abuse, and National Institute of Mental Health.

Figure 1

Timing of different data sources used in model fitting Demographic and Health Surveys (DHS) were used in Zambian communities, surveys from the Human Sciences Research Council (HSRC) were used in South Africa.

Figure 2

Predicted intervention impact, measured as relative reduction in HIV incidence, projected over a 12–36-month period and from 2020 to 2030 Data are geometric mean reductions in incidence, with 95% credible intervals taken across all communities and across 1000 parameter sets from the calibration framework. Projections are made for the PC age range (18–44 years) and the whole population for the period of 12–36 months after the start of the trial (PC12–36) with both the pre-unblinding model and the post-unblinding model. Projections for 2020–30 are under a continuation of the PopART intervention to 2030 in the trial community (scenario 1). Observed estimates of intervention effect from the statistical analysis of trial data are provided for the PC12–36 period with 95% CIs. The pre-unblinding model projections compare intervention communities to counterfactual simulations of the same intervention communities; all other projections compare intervention communities to their matched arm C community in the same triplet (and 95% credible intervals of model output). PC=population cohort

Figure 3

Predicted average HIV incidence rate using post-unblinding model versus observed HIV incidence over population cohorts 12–36 (A) Arm A vs arm C communities. (B) Arm B vs arm C communities. Projections are stratified by sex across seven triplets, across 1000 parameter sets. Horizontal lines within violin plots show medians of model projections.

Figure 4

Projected mean HIV incidence across total population of arm A and B communities for the period 2010–30 For ease of comparison the top row (A, B) shows scenarios 1, 2, and 4, and the bottom row (C, D) shows scenarios 1, 3, and 4. Solid lines show the median of the distribution of the arithmetic mean of HIV incidence per 100 person-years across all intervention communities and shaded areas show 95% credible intervals of mean HIV incidence. Median and 95% credible intervals are across model output from 1000 parameter sets that have been randomly drawn from each community.