Abstract
INTRODUCTION: HIV Prevention Trials Network (HPTN) 071 (PopART) was a cluster-randomized trial to evaluate universal testing and treatment (UTT) strategies for HIV prevention. HPTN071 compared three arms: (A) combination prevention with UTT; (B) combination prevention with universal testing and antiretroviral therapy initiation according to local guidelines; and (C) standard of care (SOC). Interventions were implemented in entire randomized communities, with impacts on HIV incidence measured in "population cohorts," that is the HPTN071 sample. Unexpectedly, a significantly lower incidence was not observed in arm A relative to SOC. Importantly, rates of participation in the HPTN071 sample differed among population subgroups, for example men were underrepresented. METHODS: To correct for underrepresented subgroups, PopART intervention effects are estimated in a population of interest, adults aged 18-44 in trial provinces, characterized with two nationally representative HIV-focused surveys. The HPTN071 sample is weighted to match the population of interest by demographics and HIV risk factors. Risk of HIV acquisition is compared across arms, both in the trial population (unweighted) and the population of interest (weighted). Both (1) the risk of HIV acquisition between 1 and 3 years and (2) the risk of HIV acquisition by 3 years are compared. RESULTS: In the trial population, estimated risk in arm A is, counterintuitively, slightly higher than SOC (Year 1-3 Risk Difference [RD]: 0.10%; 95% CI: -1.15%, 1.25%). After weighting, risk in arm A is lower than SOC in the population of interest (RD: -0.34%; 95% CI: -2.04%, 0.96%). Weighting also strengthened the estimated effect in arm B relative to SOC (unweighted RD: -0.66%, 95% CI: -1.88%, 0.46%; weighted RD: -1.18%, 95% CI: -2.85%, 0.15%). Weighted year 3 risk difference estimates indicated even stronger possible intervention effects: A versus SOC -0.83% (95% CI: -2.94%, 0.99%), B versus SOC -1.86% (95% CI: -3.80%, -0.09%). CONCLUSIONS: PopART interventions are estimated to be more protective in the population of interest than observed in the HPTN071 sample. These results partially explain the unexpected finding in arm A, providing further support for UTT strategies for HIV prevention. This analysis also highlights the importance of considering heterogeneous treatment effects among population subgroups when measuring the overall efficacy of HIV interventions.