Impact of a universal testing and treatment intervention on HIV incidence in Zambia and South Africa: results of the HPTN 071 (PopART) community-randomized trial

Background Universal testing and treatment (UTT) is a potential strategy to reduce HIV incidence, yet prior trial results are inconsistent. We report results from HPTN 071 (PopART), the largest HIV prevention trial to date. Methods In this community-randomized trial (2013-18), 21 communities in Zambia and South Africa were randomized to Arm A (PopART intervention, universal antiretroviral therapy [ART]), Arm B (PopART intervention, ART per local guidelines), and Arm C (standard-of-care). The PopART intervention included home-based HIV-testing delivered by community workers who supported linkage-to-care, ART adherence, and other services. The primary outcome, HIV incidence between months 12-36, was measured in a Population Cohort (PC) of ~2,000 randomly-sampled adults/community aged 18-44y. Viral suppression (VS, <400 copies HIV RNA/ml) was measured in all HIV-positive PC participants at 24m. Results The PC included 48,301 participants. Baseline HIV prevalence was similar across study arms (21%-22%). Between months 12-36, 553 incident HIV infections were observed over 39,702 person-years (py; 1.4/100py; women: 1.7/100py; men: 0.8/100py). Adjusted rate-ratios were A vs. C: 0.93 (95%CI: 0.74-1.18, p=0.51); B vs. C: 0.70 (95%CI: 0.55-0.88, p=0.006). At 24m, VS was 71.9% in Arm A; 67.5% in Arm B; and 60.2% in Arm C. ART coverage after 36m was 81% in Arm A and 80% in Arm B. Conclusions The PopART intervention with ART per local guidelines reduced HIV incidence by 30%. The lack of effect with universal ART was surprising and inconsistent with VS data. This study provides evidence that UTT can reduce HIV incidence at population level. Trial registration ClinicalTrials.gov NCT01900977


Introduction
In 2017, ~37 million people were living with HIV worldwide, with 1.8 million new infections. 1 HIV incidence is declining worldwide, but is unlikely to reach the UNAIDS target of <500,000 new infections by 2020. 2 Steep reductions in incidence are needed to curb the HIV/AIDS epidemic.
Universal testing and treatment (UTT) has been proposed as an important component of HIV combination prevention programs. 3,4 The HPTN 052 trial showed that early antiretroviral therapy (ART) initiation dramatically reduced HIV transmission among couples 5,6 , and the PARTNER study showed that viral suppression (<200 copies/ml) prevented HIV sexual transmission. 7,8 Mathematical modeling predicted that HIV incidence would fall steeply if HIV testing were delivered throughout a population and ART initiated immediately after diagnosis. [9][10][11] Early ART also confers individual health benefits. 12,13 In 2015, the World Health Organization updated its guidelines recommending immediate ART for all HIV-positive individuals 14 , and UNAIDS proposed 90-90-90 HIV testing and treatment targets (by 2020: 90% of HIV-positive individuals should know their status; 90% of those individuals should be on ART; and 90% of those on ART should be virally suppressed). 15 While there is compelling evidence supporting UTT for HIV prevention, it was not clear whether UTT could be implemented effectively at population level and impact HIV incidence.
Four community-randomized trials (CRTs) in sub-Saharan Africa addressed these questions; two (TasP and SEARCH) reported no impact of UTT on HIV incidence; a third (Ya Tsie) reported a 30% reduction in incidence, of borderline statistical significance. [16][17][18] The fourth study, HPTN 071 (PopART), is the largest HIV prevention trial ever conducted.
Here, we present the primary results of HPTN 071 (PopART); we also describe the uptake of the PopART intervention and its impact on viral suppression. The study was designed by members of the Study Team with input from the sponsor, funders and government and non-governmental partners in Zambia and South Africa, listed in the Acknowledgements. The data were collected by staff of Zambart and the Desmond Tutu TB Centre in collaboration with LSHTM and the HPTN Statistical and Data Management Center. The data were analyzed by the analytic authors identified at the beginning of the manuscript who vouch for the integrity of the analysis. All authors vouch for the integrity of the data, contributed to the preparation and review of the manuscript and agreed to its publication. The initial draft was written by the first author. The sponsor required no agreements restricting access to the data or freedom to publish the study findings.

Methods
The study design has been described previously 19 and is summarized below.

Study population
HPTN 071 (PopART) was conducted between 2013-2018, in 21 urban/peri-urban communities in Zambia and Western Cape Province, South Africa (total population ~1 million; average ~50,000/community). Each community was the catchment population of a government clinic. Communities were arranged in seven triplets matched on geographical location and estimated HIV prevalence. Communities in each triplet were randomly allocated to three study arms in simultaneous public ceremonies. Restricted randomization was used to ensure balance across study arms on population size, baseline ART coverage and HIV prevalence. 19 The three study arms are shown in Figure S1. Arm A communities received the PopART intervention (see below) with universal ART. Arm B communities received the PopART intervention with ART provided according to local guidelines. Arm C communities did not receive the PopART intervention, but received standard-of-care at government clinics, including HIV testing and ART offered according to local guidelines. The PopART intervention, delivered to Arm A and B communities only, was a combination prevention package ( Figure S2). Specially trained community health workers (Community HIV-care Providers, CHiPs) delivered services at annual household visits (see supplementary text). CHiPs worked in pairs, each pair responsible for ~500 households.

Intervention
Data collected by CHiPs were used primarily to support service delivery but also to evaluate intervention coverage.
At each visit, CHiPs offered HIV counseling and rapid testing, and provided support for linkage to care and ART adherence for HIV-positive clients. They referred uncircumcised HIV-negative men for voluntary medical male circumcision and HIV-positive pregnant women for antenatal care including prevention of mother-to-child HIV transmission. CHiPs also screened clients for symptoms of tuberculosis and sexually transmitted infections, with referral for diagnosis and treatment, and promoted and provided condoms.
In all 21 communities, HIV care and ART were provided at local government clinics. In Arm A, these clinics offered ART irrespective of CD4 count throughout the trial, with written consent for those initiating ART outside of local guidelines until universal ART became standard. In Arms B and C, the clinics provided ART initially at a CD4 threshold of 350 cells/ml, which increased to 500 cells/ml in 2014. Universal ART was offered from April 2016 (Zambia) and October 2016 (S Africa) ( Figure S3).

Outcome evaluation
The effect of the intervention on population-level HIV incidence was measured in a Because the original enrollment target (2,500 adults/community) was not reached in PC0, additional participants were enrolled at 12 months (PC12N) and in arms A and C only at 24 months (PC24N), excluding households sampled previously.
At each visit, PC participants were interviewed by a field research assistant (separate from the CHiPs) using a structured questionnaire that included collection of demographic, socioeconomic and behavioral data, as well as data related to HIV prevention, diagnosis and treatment; data were collected electronically. Following the interview, blood was collected by a research nurse, who also offered HIV rapid testing to all PC participants.
The pre-defined primary study outcome was HIV incidence between PC12 and PC36, comparing Arm A and Arm B to Arm C. This approach provided one year to fully establish the study intervention before measuring study outcomes. Other outcomes reported here include viral suppression (VS, < 400 copies HIV RNA/ml) and the estimated coverage of HIV testing and ART based on CHiPs data from Arms A and B.

Statistical considerations
Sample size calculations were informed by initial projections of intervention effect based on mathematical modeling 19,20 which suggested that HIV incidence might be reduced by up to 60% in Arm A and 25% in Arm B, compared with Arm C. Assuming HIV incidence in Arm C of 1.0 to 1.5 per 100 person-years (py), a between-community coefficient of variation (k) within matched triplets of 0.15-0.20, 2,500 PC participants/community with 85% HIVnegative at baseline, and 25% loss to follow-up over three years, study power would exceed 75% or 85% for effects of 35% or 40%, respectively.
Analysis methods are described in detail in the Statistical Analysis Plan, completed before data unblinding. 21 Briefly, HIV incidence was measured in PC participants who were HIV-  Table S1). For the primary outcome, statistical inference used a two-stage approach recommended for CRTs with <15 clusters/arm. 22,23 At the first stage, Poisson regression on data from all three study arms was used to compute E, the expected number of events (incident HIV infections) in each community, after adjusting for age, sex and baseline HIV prevalence, assuming a null intervention effect. At the second stage, two-way analysis of variance was carried out on log(O/E) (log ratio-residuals), where O was the observed number of events in each community, with matched triplet and study arm as factors. The test statistic is the estimated difference in means of log(O/E) between study arms, with two-sided p-values and 95% confidence intervals (CI) computed using the t-distribution. The corresponding rate ratios and 95%CI for the comparison of Arms A and C, and Arms B and C, were calculated with exponentiation. Similar methods were used for the analysis of viral suppression, except that logistic regression was used at the first stage without adjustment for HIV prevalence. The robustness of the above analyses was assessed using a permutation test based on the restricted randomization scheme.
Because the analysis plan did not include a method for controlling type I error when are reported with point estimates and 95% confidence intervals (which have not been adjusted for multiplicity and should not be used to infer treatment effects).
In Arms A and B, CHiPs data were used to estimate the proportion of HIV-positive community members who knew their HIV status and were on ART, using methods and assumptions described in supplementary text.

Ethical considerations
PC participants provided written informed consent before enrollment. Community members visited by CHiPs provided verbal consent for participation in the intervention and data collection. In Arm A, clinic patients provided written informed consent when ART was initiated outside of prevailing local guidelines (2013-2016).
Ethical approval for the study was granted by ethics committees at the London School of Hygiene and Tropical Medicine, University of Zambia, and Stellenbosch University.

Enrollment and follow-up
The CONSORT diagram ( Figure 1) shows the enrollment and follow-up of PC participants; 38,474 adults were enrolled at baseline (PC0), with 5,014 and 4,813 additional enrollments at PC12N and PC24N, respectively (total enrolled: 48,301).
At PC12, and again at PC24, 13% of PC participants were terminated from the study, most because of confirmed permanent relocation out of the study community (Table S2), and were censored from further observation; ~75% of remaining participants completed each visit. The final survey (PC36) reached 72% of eligible participants. Retention was similar across study arms at PC36 (73%, 73% and 71% in Arms A, B and C, respectively). More women (71%) than men (29%) were enrolled in PC0, with 40% of participants aged <24 years (Table 1). Socio-demographic and behavioral characteristics were similar across study arms. Approximately 17% of men reported having undergone medical circumcision.

Impact of the intervention on HIV incidence
Estimated effects of the intervention on HIV incidence are shown in Table 2 and Figure 2.
Between PC12 and PC36 (primary outcome), 553 incident HIV infections were observed during 39,702py follow-up (1.4/100py; women: 1.7/100py; men: 0.8/100py). Incidence in Arm C (geometric mean across communities) was 1.6/100py overall ( Table 2). Incidence in Arm A was 1.5/100py; the adjusted rate ratio (AdjRR) compared with Arm C was 0.93 (95%CI: 0.74-1.18, p=0.51). Incidence in Arm B was 1.1/100py; the AdjRR compared with Arm C was 0.70 (95%CI: 0.55-0.88, p=0.006). HIV incidence was lower in Arm B vs. Arm C in all seven matched triplets, while incidence was lower in Arm A vs. Arm C in only four triplets ( Figure   2). A permutation test based on the restricted randomization scheme showed even stronger evidence of an effect in Arm B vs. Arm C (p=0.001), but not in Arm A vs. Arm C (p=0.48).
The findings were essentially similar when the analysis was restricted to PC participants enrolled at PC0 (Table S6).

Impact of the intervention on viral suppression
Proportions of HIV-positive PC24 participants with VS were 71.9% in Arm A, 67.5% in Arm B and 60.2% in Arm C ( Table 2). The adjusted VS prevalence ratios were 1.16 (95%CI: 0.99-1.36, p=0.07) for Arm A vs. Arm C and 1.08 (95%CI: 0.92-1.27, p=0.30) for Arm B vs. Arm C. In Arms A and B, VS at PC24 was higher in women than in men, and considerably higher in those aged ≥25 years than those aged 18-24 years ( Table 2 and Table S7). VS in Arms A and B increased steeply from ~55% at PC0 to ~75% at PC36 (Table S8). VS in participants who self-reported ART use was consistently high in Arms A and B (86-91%, Table S9).

Coverage of the intervention
Based on CHiPs data, the estimated proportions of all HIV-positive adults who were on ART at the end of the study were 81% in Arm A and 80% in Arm B (Table S10). Figure 3 shows estimated ART coverage by age and sex, indicating similar coverage in Arms A and B, lower coverage in men than women, and lower coverage among younger compared with older individuals. ART coverage was also similar in Arms A and B in most triplets ( Figure S6).

Discussion
This study provides evidence that UTT can reduce HIV incidence at population level. In Arm B, HIV incidence was reduced by 30% compared to the standard-of-care control arm; surprisingly, there was no evidence of such an effect in Arm A. While these urban communities had high mobility, analysis of available data do not suggest any appreciable differences in migration across study arms. Further analyses of qualitative and quantitative data from the study communities, and data from an ongoing phylogenetic study, may shed light on the unexpected Arm A result.
Strengths of the study included the large sample size, enrollment of a randomly-sampled cohort to measure HIV incidence and VS at community level, delivery of ART through routine services at government clinics, the availability of extensive process data used to improve and refine delivery of the intervention, and strong community engagement. While our study communities were not chosen to be representative of Zambia or South Africa as a whole, conduct of the study in large urban communities with high rates of mobility should increase the generalizability of the findings to other urban areas of Southern Africa with generalized HIV epidemics.
A limitation of the study was the relatively small number of randomized communities (seven/arm). The difference in observed effects in Arm A vs C and B vs C may thus be a chance finding, given the similar levels of ART coverage and VS in Arms A and B, and the similar nature of the Arm A and B interventions during most of the primary analysis period.
To evaluate the overall effect of the PopART intervention vs standard of care, we therefore conducted a post-hoc analysis combining Arms A and B and found an estimated rate ratio of 0.81 (95%CI:0.66-0.99) compared with Arm C, consistent with a 20% reduction in incidence.
Another limitation is that data on uptake of interventions among HIV-positive participants in the PC may be subject to a Hawthorne effect, because participants had regular contact with research staff offering HIV testing and providing referral to care. We would expect the Hawthorne effect to be greatest in Arm C, where PC participants did not have access to CHiP services for testing and referral. Thus, for uptake estimates we rely mainly on intervention data, which were only available from Arm A and B communities. The results of HPTN 071 (PopART) are consistent with programmatic and survey data [24][25][26][27] , and should be considered alongside those of the other three trials that measured the effect of UTT on HIV incidence in Africa, all of which were smaller and undertaken in largely rural communities. The TasP trial 16 found no effect on HIV incidence, which may have reflected the similar HIV testing services provided in the intervention and control arms, with low levels of ART coverage in both arms. The SEARCH trial 17 also found no effect on HIV incidence, which may have reflected intensive baseline community-based HIV testing in both intervention and control arms. The Ya Tsie trial 18 observed a 30% reduction in incidence, which was of borderline statistical significance given the relatively small numbers of HIV seroconversion events.
Our finding of a 20-30% reduction in HIV incidence at population level was measured against a background of decreasing incidence in Arm C, possibly attributable to gradually increasing coverage of ART in the general population. This indicates that combination prevention including UTT can make a substantial contribution to HIV epidemic control.
Importantly, the effects seen in our study, Ya Tsie study and others 28 were achieved by delivering intensive household-based HIV-testing services; this may have played a more important role than changes in ART guidelines. The universal "test" component of a "testand-treat" strategy is vital, as is continued attention to primary HIV prevention interventions. 29,30 Results from planned cost-effectiveness and modeling studies will provide information on the value-for-money and long-term impact of such strategies which will help to inform policy and practice. ART coverage data from HPTN 071 (PopART), like data from other studies, draws special attention to the challenges in achieving ART coverage targets in young people, men, and communities with high mobility. 31       The denominator is the number of HIV-positive participants. 3 HIV status not determined occurred when a participant did not consent to specimen collection, no sample was available or when lab testing did not result in a determination of infection status. 4 HSV-2 status not determined occurred when a participant did not consent to specimen collection, or no sample was available.  3 P-value compared to t-distribution with 12 degrees of freedom. 4 Adjusted for age, sex, baseline HIV prevalence 5 P-value compared to t-distribution with 11 degrees of freedom, . 6 Overall prevalence is geometric mean of individual community proportions with viral      The table shows the HIV incidence rate between PC12 and PC36 and proportion of HIVpositive participants with viral suppression at PC24 for each triplet and overall, and for men and women, in each study arm. The table also shows the unadjusted and adjusted rate ratios for incidence and viral suppression overall, and for men and women. Viral suppression was defined as HIV viral load <400 copies/mL.