Predicting HIV Incidence in the SEARCH Trial: A Mathematical Modelling Study

Abstract

IntroductionSEARCH is one of four randomized-controlled trials (RCTs) investigating the strategy of community-based treatment-as-prevention (TasP) for the reduction of HIV incidence in sub-Saharan Africa. SEARCH takes place among 32 pair-matched rural communities in three regions of East Africa and exceeded the UNAIDS 90-90-90 targets for HIV testing, linkage to care, and viral suppression in the intervention arm. We used mathematical modeling to estimate expected 3-year cumulative HIV incidence in both arms of the trial, using different assumptions about two main sources of uncertainty: scale-up of antiretroviral therapy (ART) in the control arm, and the degree of mixing between SEARCH residents and non-residents.\n\nMethodsWe used the HIV modelling software EMOD-HIV to configure and calibrate a new model of the SEARCH communities. The 32 trial communities were clustered into six nodes (three for the control arm and three for the intervention arm) using k-means clustering based on community HIV prevalence, male circumcision rates, mobility, and geographic region. The model was parameterized using data on demographics, HIV prevalence, male circumcision rates, and viral suppression data collected at trial baseline in 2013, and calibrated to nodespecific and age-specific HIV prevalence, ART coverage, and population size. Using data on ART scale-up in subsequent follow-up years in the trial, we varied linkage to ART in the control arm and the degree of external mixing between SEARCH residents and non-residents.\n\nResultsIf no external mixing and no additional control arm ART linkage occurred, we estimate the trial would report a relative risk (RR) of 0.60 (95% CI 0.54-0.67, p<0.001), with all simulations showing a significant result. However, if SEARCH residents mixed equivalently with non-residents and ART linkage in the control arm also increased such that the control arm also exceeded the 73% viral suppression target, the RR is estimated to be 0.96 (95% CI 0.87-1.06, p=0.458) and 72% of simulations are non-significant. Given our \"best guess\" assumptions about external mixing and year 3 data on ART linkage in the control arm, the RR is estimated to be 0.90 (95% CI 0.81-1.00, p=0.05), with 49% non-significant simulations.\n\nConclusionThe SEARCH trial is predicted to show a 4-40% reduction in cumulative 3-year incidence, but between 18-72% of simulations were non-significant if either or both ART linkage in the control arm and external mixing are substantial. Despite achieving the 90-90-90 targets, our \"best guess\" is that the SEARCH trial has an equal probability of reporting a non-significant reduction in HIV incidence as it does a significant reduction.