Golden, Matthew R MD, MPH*†; Stekler, Joanne MD, MPH*†; Hughes, James P PhD†; Wood, Robert W MD*†
Studies conducted in the early 1990s documented that men who have sex with men (MSM) use condoms more often with partners of discordant HIV status than with HIV-concordant partners.1-3 Subsequent reports compared the sexual mixing patterns of HIV-infected and -uninfected MSM and found that MSM disproportionately have sex with partners of the same HIV status,4,5 an observation that is consistent with findings from studies restricted to either HIV-infected6-10 or HIV-uninfected MSM.11,12 More recently, we observed that among HIV-infected MSM receiving medical care at the largest HIV clinic in Seattle, WA, 24% reported that they had decided not to have sex with another man in the last year because that potential partner was HIV negative and that 31% thought that a potential partner decided not to have sex with them because they were HIV infected.13 These studies document what is now widely referred to as serosorting.14
Serosorting is the practice of preferentially having sex with partners of concordant HIV status or of selectively using condoms with HIV-discordant partners. Reports from San Francisco, London, and Sydney suggest that this practice may be increasing,15-18 and some investigators have suggested that increased serosorting may be responsible for preventing large increases in HIV acquisition rates even as rates of bacterial sexually transmitted diseases (STDs) among MSM have risen dramatically.17 Responding to the increased attention focused on serosorting, some US health departments have developed or funded online resources to advise MSM and medical providers about the practice.19,20
The extent to which serosorting protects HIV-uninfected MSM from acquiring HIV is uncertain. Prospective studies have associated HIV acquisition with having unprotected anal intercourse (UAI) with partners that study subjects believe are HIV uninfected21 and with the number of HIV-uninfected partners a participant reports.22 These findings suggest that serosorting does not eliminate the risk of HIV acquisition. In particular, serosorting may fail if people misreport their HIV status, either because they do not know they are infected or because they knowingly mislead partners.13 Although the limits of serosorting remain ill defined, a mathematical modeling study evaluating the risk of HIV acquisition associated with serosorting suggested that because of the high risk of HIV transmission during the period after initial HIV acquisition, the practice may be more risky than having sex with persons with known, long-standing HIV infection.23 We evaluated the epidemiology and protective efficacy of serosorting among MSM tested for HIV in Seattle, WA.
This study is a secondary data analysis of information recorded in the Public Health-Seattle & King County STD Clinic electronic medical record. Clinicians and staff collected all behavioral information used in the study as part of routine clinical risk assessments, and all laboratory tests were done as part of standard clinical care. The unit of analysis throughout the study is the clinic visit, meaning that some individuals contributed more than 1 visit to the analysis.
Study Population and Data Collection
The study population was comprised of MSM evaluated in the Public Health-Seattle & King County STD Clinic between October 1, 2001, and December 31, 2007. We excluded visits from the analysis if data on a patient's sexual behavior was insufficiently complete to allow us to classify the patient as a serosorter or not a serosorter or if clinicians neglected to record information on the patient's HIV testing history. We restricted analyses that used new HIV diagnosis as an outcome to visits during which men were tested for HIV and analyses that used new bacterial STD as an outcome to visits during which men were tested for at least 1 such infection. Clinicians and HIV testing staff used a structured form to record patient histories. These histories were obtained before HIV testing, and clinicians routinely asked patients if they had sex with men, women, or both men and women in the preceding year. We defined MSM to include men who reported any male sex partners in that time frame. Clinicians and staff asked MSM whether they had anal intercourse with partners who were HIV positive, HIV negative, or of unknown HIV status and about how often patients used condoms (always, usually, sometimes, or never) with partners of each HIV status category; these questions used a 12-month time frame.
HIV and STD Testing
Clinicians recommended HIV testing to all patients who reported that they have not previously tested HIV positive. We tested patients who agreed to a blood draw using a second-generation HIV EIA (Vironostika HIV-1 Microelisa System; bioMerieux, Durham, NC, or rLAV Genetic System; Bio-Rad Laboratories, Hercules, CA). We offered OraQuick rapid HIV antibody tests (Orasure technologies, Bethlehem, PA). to MSM patients at high risk for HIV infection; these tests were performed on finger-stick blood until 2007, and thereafter on saliva. In addition, since 2003, we performed pooled HIV RNA testing on all MSM who tested for HIV and who agreed to a blood draw.24,25
Clinicians routinely obtained rectal specimens for gonorrhea and chlamydial culture from MSM who acknowledged having receptive anal intercourse in the preceding year. We obtained urethral specimens (swab or urine) from men if they had signs or symptoms of urethritis or if they reported exposure to a partner with gonorrhea or chlamydia. We tested urethral specimens by culture or APTIMA Combo 2 (Gen-probe, San Diego, CA). Our laboratory performed gonorrhea cultures on modified Thayer-Martin media and chlamydial cultures on McCoy cell culture.26
To assess the relative contribution of different sexual behaviors [eg, unprotected receptive anal intercourse (URAI) with an HIV-positive partner] to the risk of acquiring HIV, we first used generalized estimating equations to define the odds ratio (OR) for testing HIV positive associated with each behavior. We used these results to create a hierarchy of risk behaviors and divided the population into groups, with each visit classified based on the single highest risk behavior reported during the visit. We created a multivariate generalized estimating equations model by initially including all sexual risk behaviors and then removing variables with the highest P values stepwise, retaining only significant variables (P < 0.05) in the final model.
To assess how effective serosorting might be in preventing HIV or bacterial STD, we classified men's visits into 4 groups: (1) men who reported UAI with partners of the same HIV status only (serosorters); (2) men who reported UAI with partners of discordant HIV status or partners of unknown HIV status (nonconcordant UAI); (3) men who reported only protected anal sex; and (4) men who denied having anal sex. We assessed the risk of testing positive for HIV based on this classification. Because clinicians and staff in our clinic collected sexual history information for a 1-year time frame, we initially restricted this analysis to persons who had tested negative for HIV in the preceding year.
In analyzing trends in serosorting, we defined men as serosorters if they had UAI with men of the same HIV status only; we restricted this analysis to each man's first visit to the clinic in each calendar year. Because we had data for <3 months in 2001, we combined 2001 data with 2002 data. For analyses using bacterial STD as an outcome, we defined a bacterial STD to include rectal or urethral infections with Chlamydia trachomatis or Neisseria gonorrhoeae or early syphilis (primary, secondary, or early latent syphilis). Analyses evaluating trends in sexual risk behavior used generalized estimating equations with calendar year as an independent variable. We performed all analyses using the SAS system (SAS, Cary, NC). The University of Washington Division of Human Subjects approved all study procedures.
Between October 1, 2001, and December 31, 2007, clinicians and staff in the STD clinic evaluated 8314 MSM during 15,416 new problem visits. Charts included complete sexual behavioral data, including HIV testing history, for 12,449 (81%) clinical encounters, and these visits comprised the study population. Analyses using new HIV as an outcome, which were restricted to visits during which men were tested for HIV, included 8122 (75%) of the 10,851 visits by men without a prior HIV diagnosis. Analyses using bacterial STD as an outcome, which were restricted to visits during which men were tested for at least 1 bacterial STD, included 11,483 (92%) of 12,449 visits.
Table 1 presents data comparing clinic visits by men with and without a prior HIV diagnosis. Older men and African American men comprised a larger proportion of visits, and Asian men comprised a smaller proportion of visits, among MSM with a prior HIV diagnosis than among men without such a diagnosis. Comparing visits by men with and without a prior HIV diagnosis, previously diagnosed MSM sought care more often for symptoms or because of contact to STD and reported higher numbers of sex partners. Sexual repertoire and mixing patterns also significantly differed based on HIV status. In an analysis restricted to each patient's first visit to the clinic during the study period, symptom as a reason for visit was not significantly more common among persons with a prior HIV diagnosis, but all other associations observed in the analysis comparing visits by men with and without a prior HIV diagnosis remained statistically significant with the same direction of association (data not shown).
Among visits by men without a prior HIV diagnosis who tested for HIV during their visit (Table 1), men who reported HIV testing within the last 12 months reported higher numbers of sex partners and riskier sexual repertoires during a larger proportion of their clinic visits than men who had not recently HIV tested. Recently, HIV-tested men were also diagnosed with a bacterial STD during a larger proportion of their visits than men who had not been recently HIV tested. However, men who had not been HIV tested in the preceding 12 months tested HIV positive during a larger proportion of visits than men who had tested in the preceding year.
Figure 1 presents trends in the proportion of visits (n = 10,218) by MSM clinic patients who reported having no anal intercourse, only protected anal intercourse, serosorting, and nonconcordant UAI; results are presented separately for men with and without a prior HIV diagnosis. Among men who had previously tested HIV positive, the proportion of visits during which men reported serosorting significantly increased over the study period (P = 0.03), whereas the proportion of visits during which men reported having nonconcordant UAI significantly decreased. Among men without a prior HIV diagnosis, the proportion of visits during which men reported serosorting also increased over the study period and the proportion of visits during which men reported nonconcordant UAI decreased, but these changes were small and were not statistically significant. The proportions of visits during which men reported not having anal intercourse or having no UAI did not significantly change.
Table 2 presents data on the association of different sexual risk behaviors with newly testing HIV positive among visits by men who reported testing HIV negative in the preceding 12 months. A total of 117 (2.3%) of the 5149 tests performed on these men were HIV positive. The highest risk of testing HIV positive was associated with having URAI with an HIV-positive partner, and receptive anal intercourse was consistently more risky than insertive anal intercourse. On multivariate analysis, testing HIV positive was significantly associated with URAI with HIV-positive partners and URAI with HIV-negative partners. Number of sex partners in the preceding 12 months was not significantly associated with HIV in this multivariate model, nor did inclusion of number of sex partners in the last 12 months alter the significant association between testing HIV positive and URAI with HIV-positive or HIV-negative partners (data not shown). When this analysis was repeated in a population that included all MSM with complete behavioral data (ie, not only men who reported testing HIV negative in the preceding 12 months), testing HIV positive was also significantly associated with URAI with partners of unknown HIV status (data not shown).
Classifying men based on their highest risk sexual behavior, the largest proportion of men with newly diagnosed HIV (32%) reported URAI with an HIV-negative man as their highest risk behavior. In contrast, although men who reported URAI with an HIV-positive partner had a much higher risk of testing HIV positive (8.7% vs. 3.0%), only 13% of men with newly diagnosed HIV acknowledged that behavior.
The risk of testing positive for HIV was highest when men reported engaging in nonconcordant UAI, intermediate among serosorters and lowest among men who reported no UAI (Table 3). In an analysis restricted to visits by men who tested negative for HIV in the preceding 12 months, the risk of testing HIV positive was lower among serosorters than among men who engaged in nonconcordant UAI, though this difference was not significant [OR 0.73, 95% confidence interval (CI) 0.48 to 1.1]. This difference was significant when data included all visits by men tested for HIV during the study period (OR 0.54, 95% CI 0.40 to 0.74). Serosorting and nonconcordant UAI were associated with a comparable risk for bacterial STD (Table 4); this risk was significantly higher than the risk observed during visits when men denied having UAI or any anal sex and was independent of number of sex partners or the HIV status the patient reported at time of clinic visit.
Among visits by men who reported testing HIV negative in the preceding 12 months, both the risk associated with serosorting and the proportion of newly diagnosed cases that serosorted increased over time. From 2001 to 2003, 3 (0.85%) of 351 serosorters newly tested positive for HIV, whereas from 2004 to 2005, 15 (3.8%) of 396 serosorters tested HIV positive and from 2006 to 2007, 22 (2.8%) of 779 serosorters tested HIV positive (OR 2.1, 95% CI 1.0 to 4.5, 2004-2007 vs. 2001-2003). In contrast, the risk associated with nonconcordant UAI remained stable across the 3 time periods at 4.2% (13/313), 3.2% (13/412), and 3.5% (23/661), respectively (OR 0.80, 95% CI 0.42 to 1.5, 2004-2007 vs. 2001-2003). Likewise, the risk associated with anal intercourse only with condoms remained stable at 1.2% (5/399), 1.9% (9/482), and 1.5% (14/946) (OR 0.1.3, 95% CI 0.49 to 3.4, 2004-2007 vs. 2001-2003). Between 2001 and 2003, 3 (14%) of 21 MSM diagnosed with HIV reported serosorting at their testing visit, whereas from 2004 to 2005 and from 2006 to 2007, 15 (40%) of 37 and 22 (37%) of 59 new diagnoses, respectively, occurred in men who reported serosorting (OR 3.8, 95% CI 1.0 to 13.7, 2004-2007 vs. 2001-2003).
Using sexual history data collected during more than 10,000 visits by MSM to our STD clinic over more than a 6-year period, we evaluated the prevalence of serosorting, trends in its occurrence, and the protective efficacy of the practice. We found that our MSM patients acknowledged a behavior consistent with serosorting during more than a quarter of all clinical encounters; that the practice was increasing, particularly among HIV-infected MSM; that serosorting was associated with a lower risk of testing HIV positive than nonconcordant UAI, but a higher risk than universal condom use; that serosorting was associated with no decrease in the risk for bacterial STD; and that the protective efficacy of serosorting may be declining.
Our findings related to the protective efficacy of serosorting are consistent with previously published and presented findings. Two cohort studies of MSM have found that UAI with partners believed to be HIV negative was an independent risk factor for acquiring HIV.21,22,27 Similarly, a case-control study conducted in Seattle found that MSM who recently acquired HIV were more likely than MSM who tested HIV negative to report having UAI with a causal partner they believed to be HIV uninfected.28 All 3 of these studies suggest that HIV-uninfected men who serosort remain at risk for HIV. The Seattle case-control study did not observe an association between testing HIV negative and participants' report that they had decided not to have sex with partners because those partners were HIV positive, a direct measure of men's decision to serosort that we have used in a previous study.13 This finding casts some doubt on the protective efficacy of serosorting when it is adopted by HIV-uninfected men as a conscious effort to diminish HIV risk. Overall, our findings support the conclusion that serosorting offers partial protection from HIV; it is better than having nonconcordant UAI, but not as good as universal condom use. However, that conclusion should be tempered by the realization that our approach to classifying serosorters was based on whether men had nonconcordant UAI and not whether they made a purposeful decision to serosort. Also, from both a personal and a public health perspective, the limits of serosorting seem to be profound. The practice appeared to reduce the risk of acquiring HIV by <50% compared with having sex with partners of unknown or discordant HIV status, and 32% of all MSM with newly diagnosed HIV reported that UAI with a partner they believed to be HIV negative was their most risky sexual behavior. This finding suggests that the population-attributable risk associated with serosorting may be high.
We also observed that the proportion of HIV-infected men who could be classified as serosorters is increasing, though this trend seemed to primarily affect HIV-infected MSM. This finding is consistent with observations from London,15,29 where researchers have observed an increase in serosorting among HIV-positive men, but not HIV-negative men. In contrast, public health officials in San Francisco and Sydney have reported findings they believe to be consistent with a trend toward increased serosorting among HIV-uninfected MSM.16,17 We cannot say the extent to which differences observed in these studies reflect true variations in behavior versus differences in how the studies recruited their populations or defined serosorting. Serosorting among HIV-infected men may place them at risk for HIV superinfection30 and, as shown by our study, does not protect them from bacterial STD. On the other hand, serosorting among HIV-infected MSM does not pose the same risk for inadvertent HIV transmission as serosorting among HIV-uninfected men. Insofar as the trends we observed also included a decline in nonconcordant UAI among HIV-infected MSM, one could speculate that the rise in serosorting among HIV-infected MSM decreased HIV transmission and increased the protective efficacy of serosorting for HIV-uninfected MSM. Unfortunately, we found that the risk of acquiring HIV among HIV-uninfected serosorters may be increasing.
The fact that serosorting seems to have become more risky for HIV-uninfected MSM is surprising and concerning. In a separate study, we recently found that the period between HIV tests among MSM testing HIV positive in STD clinics in Seattle and 3 other US cities declined between 2001 and 2006.31 Like the decline in nonconcordant UAI we observed in this study, one would think that such a change would make serosorting safer for HIV-uninfected MSM; the period during which men are unknowingly infected should decrease and, with it, the risk of inadvertently transmitting HIV to partner should be reduced. These disparate trends merit additional study and highlight the complexity of trying to understand HIV transmission trends based on limited information about behavioral risks. Although we believe our findings merit confirmation, from a practical perspective, the observation that the protective efficacy of serosorting may be declining should prompt additional caution related to serosorting.
Our study population was comprised of clinical evaluations of MSM who sought care at a single STD clinic in Seattle, WA. The self-selected nature of this sample, the geographical focus of the population, and the need to exclude some visits from the analysis because of missing data all may limit the generalizability of our findings. Also, we did not have data on patients' number of sex partners of different HIV status or number of sex acts with partners of each HIV status, information that would have permitted us to conduct a more detailed analysis. We did not routinely collect data on how patients knew a partner's HIV status, and some infections we attribute to serosorting failure may reflect patient's assumptions about partners' HIV status rather than inaccurate explicit discussions between men before sex. As in other studies of sexual behavior, our data are based on self-report, which may be inaccurate. Finally, as in most previous studies, we defined serosorting based on patients' reports of their partners' HIV status and their condom use. In many instances, decisions to have sex or use condoms with partners were probably not based on an active decision to serosort; people form sexual partnerships based on myriad factors (age, race, socioeconomic status, social connections, emotional bonds, etc.), many of which may affect the likelihood that their partner is HIV infected or wants to use condoms. These factors could affect whether we classified a person as a serosorter and might be independent of whether that person made a conscious decision to serosort. As a result, our analysis may not accurately represent the true prevention efficacy of serosorting as an active prevention strategy.
In summary, we found that serosorting is common and that it seems to be increasing among HIV-infected MSM. It offers some protection against HIV, but a large proportion of persons with newly diagnosed HIV report UAI with partners they believe to be HIV uninfected as their highest risk sexual behavior. We also observed what may be a decline in the protective efficacy of serosorting over time. Whether trying to increase serosorting is a good idea depends on what behavior it replaces. To the extent that men adopted it as an alternative to more consistent condom use, it is undesirable. Insofar as it replaces nonconcordant UAI, it should be encouraged. Clinicians and counselors should be able to discuss serosorting with their patients and clients, making clear that it is not an optimal strategy for avoiding HIV but that it may be a good step toward safety for some people. Unfortunately, the complexity of serosorting makes it difficult to craft simple, clear, and accurate public health messages about the practice. We believe that widespread efforts to promote serosorting, particularly among HIV-uninfected persons, should be undertaken with caution and, if adopted, should be accompanied by rigorous evaluations to assure that they do not increase the population's HIV risk. Beyond this, our findings highlight the need for HIV-uninfected men who serosort to test for HIV frequently using the most sensitive available tests (ie, HIV RNA or HIV antigen tests) and to honestly discuss their HIV status and testing history with sex partners.
1. Dawson JM, Fitzpatrick RM, Reeves G, et al. Awareness of sexual partners' HIV status as an influence upon high-risk sexual behaviour among gay men. AIDS
2. Hoff CC, Stall R, Paul J, et al. Differences in sexual behavior among HIV discordant and concordant gay men in primary relationships. J Acquir Immune Defic Syndr Hum Retrovirol
3. Marks G, Crepaz N. HIV-positive men's sexual practices in the context of self-disclosure of HIV status. J Acquir Immune Defic Syndr
4. Golden MR, Brewer DD, Kurth A, et al. Importance of sex partner HIV status in HIV risk assessment among men who have sex with men. J Acquir Immune Defic Syndr
5. Van der Bij AK, Kolader ME, de Vries HJ, et al. Condom use rather than serosorting explains. J Acquir Immune Defic Syndr
6. Cox J, Beauchemin J, Allard R. HIV status of sexual partners is more important than antiretroviral treatment related perceptions for risk taking by HIV positive MSM in Montreal, Canada. Sex Transm Infect
7. Parsons JT, Schrimshaw EW, Wolitski RJ, et al. Sexual harm reduction practices of HIV-seropositive gay and bisexual men: serosorting, strategic positioning, and withdrawal before ejaculation. AIDS
. 2005;19(Suppl 1):S13-S25.
8. Halkitis PN, Green KA, Remien RH, et al. Seroconcordant sexual partnerings of HIV-seropositive men who have sex with men. AIDS
. 2005;19(Suppl 1):S77-S86.
9. Wolf K, Young J, Rickenbach M, et al. Prevalence of unsafe sexual behavior among HIV-infected individuals: the Swiss HIV Cohort Study. J Acquir Immune Defic Syndr
10. Poppen PJ, Reisen CA, Zea MC, et al. Serostatus disclosure, seroconcordance, partner relationship, and unprotected anal intercourse among HIV-positive Latino men who have sex with men. AIDS Educ Prev
11. Koblin BA, Chesney MA, Husnik MJ, et al. High-risk behaviors among men who have sex with men in 6 US cities: baseline data from the EXPLORE Study. Am J Public Health
12. Crawford JM, Kippax SC, Mao L, et al. Number of risk acts by relationship status and partner serostatus: findings from the HIM cohort of homosexually active men in Sydney, Australia. AIDS Behav
13. Golden MR, Wood RW, Buskin SE, et al. Ongoing risk behavior among persons with HIV in medical care. AIDS Behav
14. Suarez T, Miller J. Negotiating risks in context: a perspective on unprotected anal intercourse and barebacking among men who have sex with men-where do we go from here? Arch Sex Behav
15. Elford J, Bolding G, Sherr L, et al. High-risk sexual behaviour among London gay men: no longer increasing. AIDS
16. Mao L, Crawford JM, Hospers HJ, et al. “Serosorting” in casual anal sex of HIV-negative gay men is noteworthy and is increasing in Sydney, Australia. AIDS
17. Truong HM, Kellogg T, Klausner JD, et al. Increases in sexually transmitted infections and sexual risk behaviour without a concurrent increase in HIV incidence among men who have sex with men in San Francisco: a suggestion of HIV serosorting? Sex Transm Infect
18. Elford J, Ibrahim F, Bukutu C, et al. Sexual behaviour of people living with HIV in London: implications for HIV transmission. AIDS
. 2007;21(Suppl 1):S63-S70.
21. Koblin BA, Husnik MJ, Colfax G, et al. Risk factors for HIV infection among men who have sex with men. AIDS
22. Buchbinder SP, Vittinghoff E, Heagerty PJ, et al. Sexual risk, nitrite inhalant use, and lack of circumcision associated with HIV seroconversion in men who have sex with men in the United States. J Acquir Immune Defic Syndr
23. Butler DM, Smith DM. Serosorting can potentially increase HIV transmissions. AIDS
24. Stekler J, Swenson PD, Wood RW, et al. Targeted screening for primary HIV infection through pooled HIV-RNA testing in men who have sex with men. AIDS
25. Stekler J, Maenza J, Stevens CE, et al. Screening for acute HIV infection: lessons learned. Clin Infect Dis
26. Stamm WE, Tam M, Koester M, et al. Detection of Chlamydia trachomatis
inclusions in Mccoy cell cultures with fluorescein-conjugated monoclonal antibodies. J Clin Microbiol
27. Philip S, D Donnell D, Yu X, et al. Serosorting, but not seropositioning, is associated with decreased risk of HIV seroconversion in the EXPLORE study cohort. Paper presented at: 15th Conference on Retroviruses and Opportunistic Infections; February 4, 2008; Boston, MA.
28. Thiede H, Jenkins RA, Carey JW, et al. Determinants of recent HIV infection among Seattle-area men who have sex with men. Am J Public Health
. 2008 (epub ahead of print).
29. Elford J, Bolding G, Sherr L, et al. No evidence of an increase in serosorting with casual partners among HIV-negative gay men in London, 1998-2005. AIDS
30. Piantadosi A, Chohan B, Chohan V, et al. Chronic HIV-1 infection frequently fails to protect against superinfection. PLoS Pathog
31. Helms DJ, Weinstock HS, Mahle K, et al. MSM attending STD clinics HIV testing more frequently: implications for HIV prevention and surveillance. Paper presented at: 2008 National STD Prevention Conference; March 13, 2008; Chicago, IL.
© 2008 Lippincott Williams & Wilkins, Inc.