|Home | About | Journals | Submit | Contact Us | Français|
To estimate per-contact probability of HIV transmission in homosexual men due to unprotected anal intercourse (UAI) in the era of highly active antiretroviral therapy (HAART).
Data were collected from a longitudinal cohort study of community-based HIV-negative homosexual men in Sydney, Australia.
A total 1427 participants were recruited from June 2001 to December 2004. They were followed up with 6-monthly detailed behavioral interviews and annual testing for HIV till June 2007. Data were used in a bootstrapping method, coupled with a statistical analysis that optimized a likelihood function for estimating the per-exposure risks of HIV transmission due to various forms of UAI.
During the study, 53 HIV seroconversion cases were identified. The estimated per-contact probability of HIV transmission for receptive UAI was 1.43% (95% CI 0.48%-2.85%) if ejaculation occurred inside the rectum occurred, and it was 0.65% (95% CI 0.15%-1.53%) if withdrawal prior to ejaculation was involved. The estimated transmission rate for insertive UAI in participants who were circumcised was 0.11% (95% CI 0.02%-0.24%), and it was 0.62% (95% CI 0.07%-1.68%) in uncircumcised men. Thus, receptive UAI with ejaculation was found to be approximately twice as risky as receptive UAI with withdrawal or insertive UAI for uncircumcised men and over 10-times as risky as insertive UAI for circumcised men.
Despite the fact that a high proportion of HIV-infected men are on antiretroviral treatment and have undetectable viral load, the per-contact probability of HIV transmission due to UAI is similar to estimates reported from developed country settings in the pre-HAART era.
Most studies of per-contact probability of sexual HIV transmission have been in heterosexual people (1-4), and few estimates have been made for sex between homosexual men (5-6). The estimation of per-contact risk in homosexual men is more complex than that of heterosexual transmission. First, sexual monogamy is more common in heterosexuals, and thus sero-discordant monogamous couples are more readily available for study (1, 7). Among homosexual men, regular relationships are frequently non-monogamous, and the HIV status of other partners is often unknown (8). Second, in contrast to heterosexual transmission, in which men always take the insertive role and women the receptive role in penetrative sex, homosexual men can take either the insertive or receptive role.
It has long been demonstrated that receptive unprotected anal intercourse (UAI) with an HIV-positive man, is the major behavioral risk factor for HIV transmission among gay and other homosexual men (9). However, the role of insertive UAI cannot be ignored (10). The phenomenon of “strategic positioning”, in which an HIV-negative man takes the insertive role while engaging in UAI with a non-seroconcordant partner in order to reduce his risk of HIV infection, has been increasingly reported (11). Yet the relative risk of insertive UAI in homosexual men has not been fully examined. Emerging evidence that circumcised men may have a lower risk of acquiring HIV during insertive anal intercourse (12), consistent with heterosexual studies demonstrating reduced risk during insertive vaginal intercourse (13-15), also suggests that the effect of circumcision on per-contact probability requires exploration.
It has been a decade since the last attempt to estimate the per-contact risk of HIV transmission in homosexual men (6), during which the landscape of HIV management has changed substantially. The majority of people with HIV in resource-rich countries now receive highly active antiretroviral therapy (HAART). In Australia, it has been estimated that about 70% of people with diagnosed HIV are currently receiving HAART, and consequently, most people with HIV have undetectable viral load (16). Despite these average decreases in viral load across populations of HIV-infected people, HIV incidence has been increasing since the late 1990s in homosexual men in most developed countries including Australia (17). There is a paucity of data on HIV transmission risk at low viral loads (18-20), and no studies have reported HIV transmission risk in the era of HAART among homosexual men with high treatment rates (18). In this paper, we calculate the per-contact risk of HIV seroconversion in a prospective cohort of initially HIV-negative homosexual men in Sydney, Australia, in an environment where most men with HIV are diagnosed, and most are receiving HAART.
The Health in Men (HIM) cohort study recruited participants from a range of community-based settings in Sydney between June 2001 and December 2004, as described elsewhere (21). Men recruited to the study met the following inclusion criteria: (1) reported having sex with other men within the previous five years, (2) lived in Sydney or participated regularly in its gay community, and (3) tested HIV-negative at baseline. They were followed to the end of June 2007. Signed informed consent was obtained from all participants. Ethics approval was granted by the Human Research Ethics Committee at the University of New South Wales.
All eligible men willing to participate were interviewed annually face-to-face, with six-monthly telephone interview between these visits. At baseline, participants reported whether they had been circumcised, and self-report was almost perfectly correlated with examination findings by the study nurse in a subset of participants (22). At each interview, detailed quantitative data on the number of episodes of insertive and receptive UAI in the last six months were collected for regular and for casual partners, by HIV status of these partners (negative, positive or unknown), and, for receptive UAI, by whether or not ejaculation occurred. Episodes of protected anal intercourse involving condom failures, including condom breakage and slippage, were included as episodes of UAI of each relevant mode, and were not separately recorded. In very few instances (less than 10 occasions during the study) participants reported that they “don't know” or ‘refused’ to indicate the number of episodes of UAI types; in such circumstances the number of episodes was recorded as zero.
Methods of ascertainment of HIV seroconversion have been described elsewhere (23). Briefly, incident HIV infections were identified through annual HIV testing at follow-up visits (n=31) and by matching against the national HIV registry to identify infections in people who tested outside the study (n=22).
Among HIV seroconverters for whom we had data on HIV seroconversion symptoms (n=17), the date of HIV infection was estimated according to the following decision process: if a Western Blot was complete (n=8) then the date was chosen as the earlier of the midpoint between the last HIV-negative test and first HIV-positive test or two weeks prior to the onset of symptom; if a Western Blot was incomplete (n=9) then the date was chosen as the latest of the midpoint between the last HIV-negative test and first HIV-positive test or two weeks prior to the onset of symptom. Among HIV seroconverters for whom we had no data on HIV seroconversion symptoms (n=36), the midpoint between periodic HIV tests was used to estimate the date of HIV infection.
Our analysis included all episodes of UAI reported to take place between the first follow-up interview and the end of study for those who remained HIV-negative, and to the estimated date of HIV seroconversion for those who became HIV-infected during the study. All episodes of UAI reported at baseline were excluded from the per-contact risk calculation.
In 13 participants whose HIV seroconversions were identified through matching with Australia's national HIV registry, the estimated date of HIV infection was later than their last interview due to loss to follow-up. In these individuals, there were no behavioral data available at the time of estimated infection. Information obtained from the last interview was carried forward for per-contact risk calculation in seven cases in whom the estimated date of infection was less than 12 months after the last interview. Those whose estimated date of infection was more than 12 months after the last interview were excluded (n=6).
Statistical analyses were performed using STATA 10.0 (STATA Corporation, College Station, TX). Total numbers of episodes of UAI by sexual position (insertive, receptive with withdrawal and receptive with ejaculation) were tabulated according to partners’ HIV status. Proportions of HIV seroconverters and non-seroconverters who engaged in UAI by sexual positioning and partners’ HIV status were also compared using a Chi-square test.
A bootstrapping technique was performed to obtain a simulation-based probability distribution for estimates of the per-contact probability of HIV transmission for insertive (with or without circumcision) or receptive (with or without ejaculation) UAI. Ten thousand simulations were executed with Matlab (Mathworks, MA); for each simulation ‘N’ individuals were randomly sampled (with replacement) from the pool of ‘N’ people. The algorithm determined the optimal transmission probabilities that maximized the likelihood function:
where yi = 1 if seroconversion took place and yi = 0 if man i remained uninfected, and
is the probability that man i remains uninfected after , , acts of insertive, receptive with ejaculation, and receptive with withdrawal/no ejaculation respectively, and βI, βIC, βR, βRW are the probabilities of HIV transmission per unprotected insertive (uncircumcised), insertive (circumcised), receptive with ejaculation, and receptive with withdrawal act of UAI respectively. The symbol ci represents each man's circumcision status (ci = 1 for circumcised and ci = 0 for uncircumcised). The number of UAI exposures with HIV-infected partners was determined by the sum of the number of UAI exposures reported with HIV-positive partners, the number of UAI events with partners of unknown status multiplied by the assumed HIV prevalence in the population, and the number with partners that were assumed to be negative multiplied by the assumed HIV prevalence in the population who have not been diagnosed with HIV.
The bootstrapping algorithm maximized the log-likelihood function using a random walk minimization to estimate the transmission risk parameters under a number of conditions: including men who (i) only reported having UAI with HIV-positive partners, (ii) only reported UAI with HIV-positive partners or partners of unknown HIV status, (iii) reported any UAI, or (iv) reported insertive or receptive UAI. For simulations in which UAI acts with men of unknown status or men presumed to be HIV-negative are included, a variety of assumptions were made about the HIV prevalence in the pool of such partners: HIV prevalence of 5%, 10%, or 15% in partners of unknown serostatus and HIV prevalence of 0.5%, 1%, 1.5%, or 2% in partners presumed to be HIV-negative. The reported estimates were based on Sydney studies which have estimated the HIV prevalence in partners of unknown HIV status at 10% and of reported HIV-negative partners of 0.5% (23-24).
The HIM study enrolled 1,427 men from June 2001 to December 2004. The median age at enrolment was 35 years (range 18-75 years). The vast majority (95.2%) of participants self identified as gay or homosexual. Nearly two-thirds of men (65.7%) reported being circumcised at baseline.
A total of 1,381 men had at least one follow-up interview by the end of the study in June 2007, and 53 seroconverted with HIV. The overall follow-up time was 5,160 person-years, with a median of 3.9 years per participant. The estimation of per-contact risk was based on 1,136 men, including 46 HIV seroconverters, who reported at least one episode of UAI during the study.
Over time, these 1,136 men reported a total of 228,056 episodes of UAI (Table 1). There were slightly more episodes of insertive UAI than receptive (56.1% vs 43.9%). The majority (87.0%) of episodes of UAI, regardless of sexual positioning, were with partners reported to be HIV-negative. Very few participants (n=93, 8.2%) reported receptive UAI with HIV-positive partners, and the majority of episodes (76.8%) in this situation involved the HIV-positive partner withdrawing prior to ejaculation.
HIV seroconverters were significantly more likely to report insertive UAI with HIV-positive partners and receptive UAI with withdrawal with partners that were HIV-positive or of unknown HIV status (Table 2). An unexpected finding, based on small numbers of men, was that HIV seroconverters reported significantly fewer episodes of receptive UAI with ejaculation with HIV-positive partners than non-seroconverters. However, this result was skewed by six men who did not seroconvert despite reporting a total of 502 episodes of this behavior.
Estimates under various assumptions of HIV prevalence in partners who were reported to be HIV-negative or of unknown HIV status are shown in Figure 1. Similar transmission risk estimates were obtained across different assumptions (Figure 1). In the scenario that HIV prevalence was 10% in partners of unknown HIV status and 0.5% in partners thought to be HIV-negative, the estimated per-contact probability of HIV transmission for insertive UAI in participants who were circumcised was 0.11% (95% CI 0.02%-0.24%) (Table 3), and it was 0.62% (95% CI 0.07%-1.68%) in those who were uncircumcised. For receptive UAI, the per-contact probability was 1.43% (95% CI 0.48%-2.85%) if ejaculation inside the rectum occurred, and it was 0.65% (95% CI 0.15%-1.53%) if withdrawal occurred prior to ejaculation. Thus, receptive UAI with ejaculation was approximately twice as risky as receptive UAI with withdrawal or insertive UAI for uncircumcised men and over 10-times as risky as insertive UAI for circumcised men. Regardless of circumcision status, the pooled data estimates of the per-contact probability for insertive UAI was 0.16% (95% CI 0.05%-0.31%), for receptive UAI with ejaculation was 1.47% (95% CI 0.51%-2.93%), and for receptive UAI with withdrawal was 0.74% (95% CI 0.18%-1.68%).
In contrast to HIV transmission risk in heterosexuals (2, 4, 7, 25-26), data on HIV transmission in homosexual men are limited (5-6). There have been no publications estimating per-contact probability of HIV transmission between homosexual men in the era of HAART. The participants recruited in the current study came from a setting with high coverage of HAART. Despite this, our estimates of HIV transmission probabilities were found to be similar to those reported from developed settings prior to HAART. For receptive UAI, we estimated the per-contact risk to be 1.43% if ejaculation occurred and 0.65% if withdrawal occurred without ejaculation. We estimated the per-contact risk for insertive UAI to be 0.11% in men who were circumcised and 0.62% in uncircumcised men. Due to differences in sampling and mathematical methods in different studies, it is difficult to directly compare results between studies. Nevertheless, our estimate of the per-contact risk of receptive UAI is very similar to that from a cohort of homosexual men recruited in the US in the early 1990s (of 0.82%, which did not differentiate whether or not withdrawal was involved) (6).
Our estimate of the per-contact risk for insertive UAI in uncircumcised men was similar to that for receptive UAI with withdrawal, but was 80% lower in those who were circumcised. In comparison, among heterosexual men per-contact transmission risk was reduced by 50-60% in three randomized controlled trials of circumcision in African settings (13-15). Our estimate of transmission risk for insertive UAI is approximately twice previous estimates (6).
Our finding that the per-contact probability of HIV transmission is similar to that in the pre-HAART era was unexpected given the close correlation between HIV viral load and its infectiousness in heterosexual and vertical transmission (27). In Australia, homosexual men have very high rates of recent HIV testing (28), about 70% of HIV-positive men are receiving HAART, and 75% of those on treatment have undetectable viral load (16). Thus it is surprising that our estimates of HIV transmission risk were similar to those in an era when few HIV-positive men would have had undetectable viral load.
There are some potential explanations for this unexpected finding. First, primary HIV infection, which is associated with higher viral load and thus higher infectiousness (29-30), may have a larger role in the dynamics of HIV transmission than expected. In addition, individuals with primary HIV infection are usually unaware of their HIV status. It is likely that some of the partners not identified as HIV-positive could have had primary HIV infection. Second, the proportion of undiagnosed HIV infections or prevalence in the population could be higher than we expected (31). We assumed that the prevalence of HIV among sexual partners thought to be HIV-negative and among those with unknown HIV status were 0.5% and 10%, respectively. However, we conducted a sensitivity analysis and found our estimates to be consistent across broad assumptions. Third, it may be possible that HIV transmission by anal intercourse is not as closely related to viral load as it is in vaginal transmission (27). There is a paucity of data on HIV transmission risk at low viral loads, and there are almost no data on transmission and viral load in homosexual men (18-19). Fourth, the prevalence of other sexually transmissible infections (STIs) in Sydney, as in many parts of the developed world, was higher during the timeframe of this study than the levels during the pre-HAART era. The presence of other STIs may increase the risk of HIV transmission (32).
Our samples were recruited from a large variety of community-based sources, and the only behavioral criterion was that participants needed to report having sex with another man in the last five years. Compared with the previous mentioned US study (6), which required participants to report risky behavior, our estimate could be more representative of gay community-attached men in general. Being one of the largest cohort studies examining incident HIV infection in homosexual men, only 46 HIV seroconverters who reported at least one episode of UAI were included in the analyses. Almost a quarter of a million episodes of UAI were reported by study participants, although only around 10,000 of these were with partners who were reported to be HIV-positive. Due to limited power, co-variates such as STIs and recreational drug use could not be included in the current estimations.
As with other observational studies relying on participants’ self-report, recall bias could influence the accuracy of the results. The study implemented six-monthly telephone interviews between annual face-to-face visits to minimize the possible inaccuracy of self-reported sexual behavior due to the long interview interval. The use of face-to-face interviews might have also reduced social desirability bias arising from the studies that collect sensible sexual behaviors.
Despite a more than ten-year gap from the last estimation of HIV transmission risk in homosexual men and the substantially improved treatment availability, the per-contact risk of HIV transmission with an HIV-positive partner does not seem to have reduced. While these updated estimates are valuable in determining the risk of HIV transmission, caution should be exercised before interpreting the results at the level of individual men. There is considerable heterogeneity between individuals including various biological and genetic factors associated with HIV infectiousness and susceptibility. This is emphasized by the occurrence of twelve seroconversion cases in the cohort of this study as a result of fewer than ten episodes of UAI per person and six cases that did not seroconvert despite extremely large numbers of receptive UAI episodes with HIV-positive partners. However, our estimates are useful for understanding the average magnitude of transmission risk due to different types of sexual exposures among homosexual men in the era of HAART.
Funding: The National Centre in HIV Epidemiology and Clinical Research and the National Centre in HIV Social Research are funded by the Australian Government Department of Health and Ageing. The Health in Men Cohort study was funded by the National Institutes of Health, a component of the U.S. Department of Health and Human Services (NIH/NIAID/DAIDS: HVDDT Award N01-AI-05395), the National Health and Medical Research Council in Australia (Project grant # 400944), the Australian Government Department of Health and Ageing (Canberra), and the New South Wales Health Department (Sydney).
FJ is supported by the Post-doctoral Training Fellowship (#571402), and DPW is supported by a Career Development Award (#568705) from the National Health and Medical Research Council.
The authors thank the HIM study participants, the dedicated HIM study team and the participating doctors and clinics.
Fengyi Jin performed the statistical analyses and drafted the manuscript; James Jansson carried out mathematical calculations; David P Wilson took overall responsibility for the project, developed the calculation algorithm and assisted in the analyses and drafting of the manuscript; Matthew Law, Garrett P Prestage, Iryna Zablotska, John C G Imrie, Susan C Kippax, John M Kaldor and Andrew E Grulich assisted in formulating the analyses and drafting the manuscript.
Conflict of interest: None
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.