Written informed consent was obtained from all participants for participation in Project Accept. Human experimentation guidelines of the US Department of Health and Human Services and those of the authors’ institution(s) were followed in the conduct of this research. The Project Accept study was approved by ethical review committees for the study sites (South Africa: the University of the Witwatersrand, Human Research Ethics Committee [Medical]; Tanzania: the Muhimbili University of Health and Allied Sciences [MUHAS] Senate Research and Publication Committee and the National Institute of Medical Research Ethics Committee; Zimbabwe: the Medical Research Council of Zimbabwe; Thailand: the Ministry of Public Health Ethical Review Committee of Research in Human Subjects [MOPHEC] and the Human Experimentation Committee Research Institute for Health Sciences, Chiang Mai University), and at collaborating institutions in the United States (the Committee on Human Research, University of California, San Francisco; the UCLA Office of the Human Research Protection Program (OHRPP); the Johns Hopkins University Bloomberg School of Public Health Institutional Review Board; the Medical University of South Carolina, Institutional Review Board for Human Research).

Criteria for inclusion of participants in the post-intervention assessment survey of Project Accept are described elsewhere [1]. Briefly, participants were selected from all community residents by randomly sampling households in the study community. All eligible household members aged 18–32 years were invited to participate. Each participant provided blood samples for in-country HIV testing and CD4 cell count testing. Stored plasma aliquots were used for further analyses.

HIV testing was performed at local laboratories, as previously described [7]. Two HIV rapid tests were performed in parallel. Samples were classified as HIV POS (both rapid tests reactive), HIV DISC (one rapid test reactive, one rapid test non-reactive), or HIV NEG (both rapid tests non-reactive). These terms (HIV POS, HIV DISC, HIV NEG) were used to distinguish between the initial classification of samples based solely on HIV rapid testing and the final classification of HIV infection status. A CD4 cell count was performed for all HIV POS and HIV DISC samples.

Samples were shipped to the HPTN Network Laboratory at the Johns Hopkins University School of Medicine (Baltimore, MD) for further testing. Quality assurance testing was performed to confirm the results of in-country HIV testing and to determine the HIV status of HIV DISC samples; assays used for this testing included: the Vitros EIA Human Immunodeficiency Virus Type 1 and/or 2 (HIV-1/2) Antibody Detection in Human Serum and Plasma (VITROS ECi/ECiQ Immunodiagnostic System, Ortho Diagnostics, Johnson & Johnson, Pencoed, United Kingdom), the Genetics System HIV-1 Western Blot (BioRad Laboratories, Redmond WA), and the APTIMA® HIV-1 RNA Qualitative Assay (Gen-Probe Inc., San Diego, CA). A subset of the HIV NEG samples was tested using the ARCHITECT® HIV Ag/Ab Combo assay (HIV Combo; List: 2P36; Abbott Diagnostics, Wiesbaden, Germany). The HIV Combo assay was performed only once; reactive samples were evaluated further using additional assays listed above.

Further testing to assess HIV incidence was performed using the BED-CEIA (Calypte Biomedical Corporation, Lake Oswego, OR) [5] and an avidity assay based on the Genetic Systems HIV-1/HIV-2 Plus O EIA (Bio-Rad Laboratories, Redmond, WA) [11]. Results from the BED-CEIA and avidity assays were reported as normalized optical density units (OD-n) and avidity index (%), respectively. Viral load testing was performed for a subset of study samples (Roche AMPLICOR HIV-1 MONITOR test, version 1.5, Roche Diagnostics, Indianapolis, IN). Samples were considered to be MAA positive if they had all of the following test results: BED-CEIA <1.2 OD-n, avidity index <90%, CD4 cell count >200 cells/mm³, and HIV viral load >400 copies/ml; this MAA has a mean window period of 259 days in subtype A and C epidemics [14]. HIV subtypes were determined for a subset of samples from Tanzania by phylogenetic analysis of HIV gag and gp41 sequences (GenBank accession numbers: KC589446–KC589569 [Gag] and KC589570–KC589696 [gp41]) [20].

Selected plasma samples were tested for the presence of ARV drugs [21]. Samples were analyzed using a qualitative, high-resolution accurate mass spectrometric method developed to detect 15 ARV drugs, including protease inhibitors (PIs: amprenavir, atazanavir, darunavir, indinavir, lopinavir, nelfinavir, ritonavir, saquinavir, and tipranavir), non-nucleotide reverse transcriptase inhibitors (NNRTIs: efavirenz, nevirapine), and nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs: emtricitabine, lamivudine, tenofovir). Detailed methods are described in the legend for Tables S1 and S2 in File S1.

The Laboratory Data Management System (LDMS, Frontier Science & Technology Research Foundation, Inc.) was used to track specimen collection, testing, storage, and shipping. Data from study sites and in-country laboratories were submitted to the Project Accept Statistical Center. After those data were merged and reviewed, they were submitted to the HPTN Statistical and Data Management Center (SDMC). Data from the HPTN Network Laboratory were submitted directly to the HPTN SDMC.

In Project Accept, 46,693 (94.4%) of 49,461 eligible participants provided blood samples for analysis (Table 1). Samples were initially classified as HIV POS, HIV DISC, or HIV NEG based on results from two HIV rapid tests (see Methods). CD4 cell count testing was performed for HIV POS and HIV DISC samples. Overall, 7,724 samples were HIV POS, 274 were HIV DISC, and 38,695 were HIV NEG (Table 1). The initial prevalence estimate based on in-country testing was 16.5% (1.02% in Thailand, 5.99% to 30.8% at the four African sites, Table 1). All HIV POS and HIV DISC samples and a subset of HIV NEG samples were shipped to the HPTN Network Laboratory for further analysis. The number of incident infections identified in Thailand was too low to contribute to the endpoint assessment (3 incident infections in 14 matched communities). Therefore, the endpoint assessment was limited to the four African sites.

For quality assurance, the first two batches of HIV NEG samples received at the HPTN Network Laboratory from each study site were tested using the automated, fourth generation HIV screening assay (HIV Combo, see Methods; Figure 1). Non-reactive test results were obtained for 9,631 (98.9%) of the 9,741 HIV NEG samples tested, confirming that the participants were HIV-uninfected; the remaining 110 samples had reactive test results (Table 2). A high proportion of the reactive samples (59% of the 110 samples) were from Soweto (2.24% of the HIV NEG samples from Soweto were reactive, compared to 0.35–0.79% at the other three African study sites). The high frequency of reactive test results from this site prompted an investigation that revealed that a technologist working at site did not change pipette tips while preparing plasma aliquots (Figure 2). Based on these findings, HIV Combo testing was performed for all HIV NEG samples from Soweto that were processed during the period when the technologist was employed in the laboratory. The following approach was used to exclude samples that may have been contaminated: (1) the technologist was on duty that day and at least one HIV NEG sample had a reactive HIV Combo test, or (2) the technologist was not on duty that day (or it could not be determined whether the technologist was on duty), and at least two HIV NEG samples had reactive HIV Combo tests. Note that all samples meeting these criteria were excluded, regardless of their original designation (HIV POS, HIV DISC, or HIV NEG). Overall, this process resulted in excluding 1,710 HIV NEG samples and 292 samples that were either HIV POS or HIV DISC. After excluding for contamination, 58 (0.62%) of 9,314 HIV NEG samples were HIV Combo reactive, including 0.43% of the HIV NEG samples from Soweto. Those samples were analyzed further using the testing algorithm shown in Figure 1; 44 were classified as HIV negative, seven were classified as acute infections (HIV RNA positive, HIV antibody negative), one was classified as early HIV infection (HIV RNA positive, EIA positive, Western blot indeterminate), and six were classified as established HIV infections (Western blot positive). Note that samples originally classified as HIV NEG were analyzed for quality assurance only; results from these samples were not used in the endpoint assessment.

A total of 272 HIV DISC samples remained after sample exclusion (see above, Table 2). These samples were analyzed using the testing algorithm shown in Figure 1; 265 (97.4%) of the samples were confirmed to be from individuals who did not have HIV infection. The remaining seven samples included one acute infection and six established infections (Table 2).

Quality assurance testing of HIV POS samples was restricted to samples with avidity assay results that suggested that HIV antibodies were absent or present at very low levels. Those samples were further analyzed using the testing algorithm for HIV DISC samples (Figure 1); 52 of those samples (0.70% of 7,406 HIV POS samples) were from HIV-uninfected individuals. This was consistent with the rate of false positive in-country test results observed in the Project Accept pilot study (0.63%) [7]. The remaining samples included three acute infections, three early infections, and 7,348 established HIV infections (Table 2).

We determined the frequency of subtype D infection at the Project Accept site in Tanzania by testing 113 samples from that site (at least 10 samples from each community, see Methods). Overall, 9.7% of the samples were subtype D (with 44.3% subtype A, 22.1% subtype C, and 23.9% intersubtype recombinant). The highest prevalence of subtype D in a single community was 30% (3/10), while three communities had no subtype D HIV among the samples tested. Overall, 534 (7.26%) of the 7,354 established infections in Project Accept were from Tanzania (Table 2). Based on a subtype D prevalence of 9.7%, we estimated that only 0.7% of the established infections in Project Accept would be subtype D. Even if some of the non-recent subtype D infections were misclassified as MAA positive, it would be unlikely to influence overall HIV incidence estimated in the trial. There was no significant difference in the performance of the BED-CEIA and avidity assay with subtypes A and C in our validation studies [15].

Results from in-country and quality assurance testing identified 7,361 confirmed HIV infections in Project Accept, including four acute infections, three early infections, and 7,354 established infections (Table 3). The 7,354 samples from participants with established infections were analyzed using the MAA (see Methods, Figure 3). For this analysis, all 7,354 samples were tested with the BED-CEIA and the avidity assay. HIV viral load testing was only performed for samples that had a BED-CEIA result <1.2 OD-n, an avidity index <90%, and a CD4 cell count >200 cells/mm³. This evaluation identified 467 MAA positive samples (Table 3).

In resource-limited settings, individuals with recent HIV infection are often not aware of their HIV status; even if they are, they are not likely to initiate ARV treatment until later in the course of their disease. For this reason, ARV treatment has been used as a surrogate marker of long-standing HIV infection in some settings [22]. As a final step in our incidence assessment, samples classified as MAA positive were tested for the presence of ARV drugs; this testing was performed using a high-throughput, qualitative ARV drug screen that includes ARV drugs available in the study countries at the time that the trial was performed (Table 4) [21]. ARV test results were obtained for 461 (98.7%) of the 467 MAA positive samples; 29 (6.7%) of those samples had at least one ARV drug detected and were not included in the primary endpoint assessment, leaving a total of 439 infections classified as incident (Table 3). For comparison, we also tested all 88 samples that were classified as MAA negative solely on the basis of viral suppression (i.e., samples with BED-CEIA <1.2 OD-n, avidity index <90%, CD4 cell count >200 cells/mm³, and HIV viral load <400 copies/ml); 57 (64.8%) of the 88 samples in this group had at least one ARV drug detected (Table 4). Detailed results of ARV drug testing are provided in Tables S1 and S2 in File S1.

Based on the analysis described above, the overall HIV prevalence in Project Accept was 16.5%, with a range from 5.93% in Tanzania to 30.8% in Vulindlela, South Africa (Table 3). These prevalence rates are very similar to the initial prevalence rates determined solely by in-county HIV rapid testing (Table 1). In the HIV incidence assessment, the following infections were classified as incident: acute infections (N = 4), early infections (N = 3) and infections that were classified as MAA positive that either had no ARV drugs detected or had no ARV test result (N = 438); overall, 455 incident infections were identified.

The overall annual HIV incidence in the Project Accept was 1.60%, with a range from 0.78% in Tanzania to 3.90% in Vulindlela, South Africa (Table 3).