Introduction

Self-report of the duration of HIV infection is an important measure for HIV research. Individuals with recent infection can be treated early in their infection, which may limit the size of their viral reservoir [1] and enhance CD4+ recovery [2, 3]. Additionally, newly diagnosed individuals who are recently infected are more likely to be linked to HIV infected individuals who do not know their infection status [4]. Thus, identifying recently infected individuals and then determining their social network and testing those individuals for HIV can decrease the number of HIV infected people unaware of their infection status [4]. Furthermore, if these individuals reduce their risky behaviors as a result of knowing that they are HIV positive, the rate of HIV transmission would be reduced. For these reasons accurate self-report of newly diagnosed individuals is critical for HIV clinical trials [5] and contact tracing [6]. However, the accuracy of self-reported recent HIV infection is not known.

There are many reasons why accurate self-report of recent infection would be useful both scientifically and for public health. Individuals with recent infection have smaller viral reservoirs [1] and are being enrolled for cure protocol studies (Research In Viral Eradication of HIV Reservoirs, protocol # ISRCTN83717528), and prioritized for public health follow-up and partner notification. When using biomarkers to determine recent infection, there are multiple confounding factors that impact their ability to correctly discriminate the duration of infection at the individual level and correctly estimate population level incidence. These include race [7], gender [8], infecting subtype [9,10,11], set point viral load [12, 13], AIDS [14, 15], viral suppression [13, 14, 16], and viral breakthrough [17]. A multi assay algorithm (MAA) uses a combination of biomarkers with orthogonal properties for misclassification to help maximize the precision in distinguishing recent from non-recent infection [18]. This method has been demonstrated to accurately estimate population level incidence and to accurately identify recent infection at the individual level [19, 20].

We applied this methodology to a subset of participants in the Strategic Timing of AntiRetroviral Treatment (START) trial [21] who were: (1) diagnosed with HIV within 6 months before randomization and also reported being infected within the 6 months before randomization, (2) diagnosed with HIV within 6 months before randomization and did not know their date of infection, and (3) were diagnosed ≥ 2 years before randomization (controls). Our goal was to validate self-report of recency of HIV infection and to use the MAA in combination with self-reported date of infection and documented date of diagnosis to estimate the number of participants enrolled in START with early infection (< 6 months prior to randomization). By identifying a cohort of recently infected individuals randomized to immediate or deferred treatment who will be in long term follow up until 2021, we will be able to determine the health benefits of treatment early in infection.

Methods

START is a multinational randomized clinical trial comparing immediate versus deferred antiretroviral treatment (ART) initiation for major clinical outcomes. The study enrolled 4684 HIV-positive ART-naïve individuals from 35 countries, between April 2009 and December 2013. Participants had to have two CD4 counts > 500 cells/mm3 at least 2 weeks apart within the 60 days before enrollment and be at least 18 years of age [21]. The study results were unblinded in May 2015 due to clear and significant clinical benefit of immediate ART initiation [21].

At study entry, in addition to demographics, CD4+ count, HIV RNA level and likely mode of HIV infection, two dates were recorded: (1) the date the participant was first diagnosed with HIV as documented in the participant’s record; and (2) if known, the month and year the participant believes he/she was first infected with HIV. Plasma samples were collected at entry for future research purposes and stored centrally. These stored samples were tested with the MAA for three groups of participants. Group 1 included 167 participants who were diagnosed and reported being infected ≤ 6 months prior to randomization. Group 2 included 771 participants who were diagnosed ≤ 6 months prior to randomization and who reported the date of infection was unknown. Those diagnosed with HIV between 6 and 24 months of randomization were not studied (n = 1551), as well as those who were diagnosed within 6 months but reported being infected for > 6 months (n = 518). Group 3 consisted of 199 randomly chosen participants who were diagnosed with HIV ≥ 2 years prior to randomization (controls). All samples tested were from individuals who consented to and had stored plasma samples.

Baseline plasma samples were tested by the MAA, which was developed for cross-sectional incidence testing [19]. All samples were tested with the Limiting Antigen-Avidity assay (LAg-Avidity) [22] and those with a LAg-Avidity normalized optical density < 3.0 were further tested by the Johns Hopkins avidity modified GS HIV-1/HIV-2 Plus O EIA [23]. As all participants had a CD4+ count > 500 cells/mm3 at enrollment, those who had a LAg-Avidity normalized optical density < 2.9, had an avidity index below 85% and a baseline viral load > 400 copies/mL were classified as recently infected. This testing algorithm had been validated in samples from the U.S. where the average duration that an individual appeared recently infected was 146 days [19] and no sample from an individual infected > 4 years was classified as recent. For the algorithm used, it was previously determined that for individuals classified as recent by this testing methodology, 95% were infected < 1 year [19].

Baseline characteristics of Groups 1–3 were summarized with simple summary statistics. The following baseline characteristics were compared: age, gender, race, geographic location, likely mode of HIV infection, baseline CD4+ and CD8+ cell count, and baseline HIV RNA level. These baseline characteristics were chosen to describe the groups because we hypothesized that those with recent infection would differ from those who were not recently infected on many of the characteristics. Specifically, those with recent infection were hypothesized to be younger, men who have sex with men, and enrolled by sites in high-income countries where they were more likely be tested. Those with early infection would also be more likely to have elevated HIV RNA levels. Proportions were compared using Chi squared tests and medians compared using Wilcoxon rank sum tests. Logistic regression analysis which considered multiple factors was used for three comparisons performed in the following order: (1) within group 1, participants classified as recently infected by the MAA versus those who were not; (2) participants classified as recently infected by the MAA in Group 2 versus all participants in Group 1; and (3) all participants in Group 1 plus participants in Group 2 who were classified as recently infected by the MAA versus participants in Group 2 who were not classified as recently infected by the MAA. Odds ratios were used to compare groups. The first comparison was made in order to determine whether those who self-reported recent infection differed according to the MAA result. The 2nd comparison was made to determine whether those classified as recently infected by the MAA who did not know the date of their infection differed from those who self-reported recent infection. The 3rd comparison was made to describe the differences in baseline characteristics between those we classified as recently infected by self-report or MAA with those who were not classified as recently infected. Statistical analyses were performed using SAS, version 9.3 (SAS Institute, Cary, NC). A p value < 0.05 was considered significant. All p values cited are two-sided and not adjusted for multiple comparisons.

Results

Sample Description and MAA Results

Among the 180 participants who reported being infected within 6 months of randomization, 167 (93%) consented to storage of specimens for future HIV research and had a baseline specimen collected. Of those diagnosed with HIV in the 6 months prior to randomization with unknown date of infection, 93% (771/830) had baseline samples for analysis. 1605 participants reported a diagnosis of HIV two or more years before randomization, 1539 (96%) consented to store specimens, and 199 of those participants (13%) were randomly chosen for this study as controls. Characteristics of the three groups studied are shown in Supplementary Table 1, overall and according to MAA results.

A significantly higher percentage of individuals who reported being infected ≤ 6 months were recently infected by the MAA compared to controls, participants who were diagnosed with HIV for at least 2 years before enrolling in START (65% [109/167] vs. 2.5% [5/199]; χ2 (1, N = 366) = 166.8, p < 0.001) (Table 1). Additionally, the group who reported being infected ≤ 6 months had lower LAg-Avidity values than the group diagnosed 2 or more years before enrollment (median normalized optical density 1.86 [IQR 0.99, 3.01] vs. 4.53 [3.84, 4.99]; χ2(1, N = 366) = 162.5, p < 0.001).

Table 1 Results from the MAA for the three HIV infection groups
Full size table

Among the 771 participants who were diagnosed with HIV within 6 months prior to enrollment and reported the date of infection as unknown, 206 (26.7%) were recently infected by the MAA. The median LAg-Avidity results for those 206 participants was slightly higher than for the 109 participants in Group 1 who were recently infected by the MAA (OD-n 1.62 [1.05, 2.16] versus 1.46 [0.70, 1.94]; χ2(1, N = 314) = 4.5, p = 0.034).

Comparison of Baseline Characteristics of Participants Recently Infected Versus not Recently Infected

Among participants in Group 1 (diagnosed with HIV within the past 6 months and HIV infected within the past 6 months) the only baseline characteristic that differed between those classified as recently infected by the MAA and those who were not was HIV RNA level (26,323 copies/mL vs. 19,114 copies/mL; χ2(1, N = 167) = 2.6, p = 0.11 in univariate analysis and χ2(1, N = 165) = 4.8, p = 0.028 in multivariate analysis) (Supplementary Tables 1 and 2).

Those in Group 2 (diagnosed with HIV within the past 6 months and with a date of HIV infection unknown) classified as recently infected by the MAA were similar to participants in Group 1 with two exceptions, the percentage of participants who were MSM (69.4 vs. 82.6%; χ2(1, N = 373) = 7.6, p = 0.006 in univariate analysis and χ2(1, N = 373) = 4.3, p = 0.039 in multivariate analysis) and CD8+ count (945 vs. 1,086 cells/mm3; χ2(1, N = 373) = 7.3, p = 0.007 in univariate analysis and χ2(1, N = 371) = 10.5, p = 0.001 in multivariate analysis, Table 2 and Supplementary Table 1).

Table 2 Comparison of participants in Group 2 confirmed as recently infected by MAA (unknown HIV infection date, n = 206) with those in Group 1 (self-reported HIV infection ≤ 6 months, n = 167); all participants diagnosed with HIV ≤ 6 months before randomization
Full size table

Table 3 (and Supplementary Table 3) compares participants in Group 1 together with those in Group 2 who were classified as recently infected by the MAA (recently infected group) to participants in Group 2 who were not classified as recently infected by the MAA. There were a number of differences between these two groups. In the multivariate analysis, participants recently infected by self-report and/or MAA were younger, more likely to be Asian, less likely to black, less likely to be heterosexual, more likely to be enrolled at sites in the U.S., Europe or Australia, and have higher HIV RNA levels.

Table 3 Comparison of participants in Group 1 (self-reported HIV infection ≤ 6 months, n = 167) combined with those in Group 2 confirmed as recently infected by the MAA (unknown HIV infection date, n = 206) versus those in Group 2 not confirmed as recently infected (n = 565); all participants diagnosed with HIV ≤ 6 months before randomization
Full size table

Discussion

In START we demonstrated that nearly two-thirds of participants who were diagnosed with HIV within 6 months before enrollment and reported HIV infection within 6 months before enrollment had MAA findings which confirmed their recent infection, whereas, only 2.5% of participants who were diagnosed with HIV more than 2 years before enrollment (controls) had positive assay results. Those reporting recent infection and confirmed by the assay did not differ in terms of baseline characteristics from those who were not. This observation taken together with the limitations of the MAA assay used [19], indicate that the classification of participants as recently infected by self-report in START (Group 1) is likely accurate.

There are several limitations to this study. The gold standard used to identify/confirm recent infection was the combination of biomarkers as part of a MAA. The MAA used has a window period of 146 days for HIV-1 clade B infected individuals [19, 26]. This means that the average duration an individual will appear as recently infected by this MAA is 4.8 months (146 days, 95% CI 122 to 170 days) after seroconversion. In START, the median time from diagnosis to enrollment was 2.4 months for those who reported being infected ≤ 6 months. MAA was based on a plasma sample taken at the time of randomization. Had a plasma sample been available at the time of HIV diagnosis to study the agreement of the MAA and self-reported recent infection, agreement would likely have been higher. Furthermore, of the five individuals who were known to be infected more than 2 years, but were classified as recently infected by the MAA, three were subjects where the time between diagnosis and blood draw was < 4 years. Additionally, four of these five individuals were from countries where HIV-1 non-B clades circulate [24]. Since the incidence assays used have different properties depending on the infecting clade [10, 11, 22], it is not surprising that the majority of the misclassified long term infected samples come from HIV-1 non-B endemic areas.

Among participants with an unknown infection date who were diagnosed with HIV in the 6 months prior to enrollment, 206 (27%) were recently infected according to the MAA. These 206 participants had similar baseline characteristics to those who self-reported being recently infected. Thus, taken together with those self-reporting recent infection, 373 participants in START were infected within 6 months before enrollment. Compared to the 565 participants diagnosed with HIV in the 6 months prior to enrollment who were not recently infected, the 373 participants who were recently infected were younger, more likely to be Asian, less likely to black, less likely to be heterosexual, more likely to be enrolled at sites in the U.S., Europe or Australia, and have higher HIV RNA levels.

There are several explanations for why particular factors were associated with recent infection in the START trial population. The entry requirements for the trial required high CD4 counts which would select for individuals who were early in their infection or with a less pathogenic strain of HIV. Therefore the finding that higher viral load was associated with recent infection is not surprising. The majority of heterosexual participants in the trial came from Africa, where frequent testing is not the norm. For example a recent study in Zambia, 40% of men had never been tested for HIV in their lifetime [25]. The increased odds of Asians being recently infected is not surprising as most of those subjects came from Thailand which has a long history of HIV prevention and 75% of MSM reporting getting tested for HIV in the past year [25].

Although we do not know of any published studies comparing self-report of recency of HIV infection to biomarkers, a few studies have compared self-report of HIV status to biomarkers of HIV infection. For example, Fisher et al. compared enzyme-linked immunosorbent assay test results to self-report of HIV status, finding that the specificity of self-report was above 99% at baseline and follow-up while the sensitivity was 32% at baseline and 61% at follow-up [26]. Additionally, Ross et al. compared Western blot results to self-reported HIV infection and found that the sensitivity of self-report was 93% and the specificity was above 99% [27]. However in our study there were no pre-determined reasons for misinforming the study staff about their date of HIV infection. Our results showed a similarly high level of self-report specificity compared to these two studies while our sensitivity was lower than that observed by Ross et al. and slightly higher than the sensitivity seen by Fisher et al. at follow-up [26, 27].

In summary, in the START trial 373 participants with recent HIV infection were enrolled. To our knowledge, this subgroup represents the largest study of immediate versus deferred ART among individuals recently infected with HIV.