In Table 4, we conducted a series of multivariable logistic regressions with UAI as the dependent variable and either seroconcordance or intentional seroconcordance as the independent variable, with all models based on the GEE parameters mentioned earlier. The purpose of developing these models was to understand the direction and magnitude of association between serosorting intention and UAI. To achieve less-biased estimates, we controlled for a set of participant demographic variables and partnership variables shown to be significant predictors of UAI in prior research, which may also be associated with the 2 seroconcordance outcomes.14,24–28 Sensitivity analyses assessed whether changes in how we dichotomized serosorting intent impacted the direction or significance of study findings regarding its relation to UAI. All analyses were conducted in STATA 11.2.29
Overall, 6104 men began the online survey, and 4138 began the section regarding sex partners, a response rate of 68%. Of this group, 3519 of 4138 (85%) had sex with a man in the previous 6 months. Table 1 shows demographic features of the 3519 respondents who had sex with men in the last 6 months, by reported HIV status. At the individual level, MSM unaware of their status (HIVu) were younger, poorer, and less educated than HIV− and HIV+ MSM. HIVu men were less likely than HIV− and HIV+ men to identify as homosexual (70%, 81%, and 90%, respectively; P < 0.001). Racial composition among the groups was similar, with the exception of non-Hispanic blacks being overrepresented in the HIV+ group (P < 0.001).
From 3519 respondents who reported sex with a man in the previous 6 months, we collected participant HIV status and partnership type data on 8558 partnerships in the prior 6 months (ie, each man reported, on average, data on 2–3 sexual partners). Table 2 shows characteristics of these partnerships by participant’s HIV status. Nearly half were UAI partners, and less than one-quarter were classified as main partners. Approximately half of partners were met online. A higher proportion of HIV+ men reported UAI than HIV− and HIVu men (56%, 43%, and 44%, respectively; P < 0.001). In addition, a lower proportion of HIV+ men reported main partnerships than HIV− and HIVu men (17%, 24%, and 23%, respectively; P < 0.001). A higher proportion of HIV+ men’s partners were met online than among HIV− and HIVu men (59%, 52%, and 48%, respectively; P < 0.05).
Figure 1 details classification of eligible individuals and their partnerships, by respondent serostatus, partnership seroconcordance, partnership UAI in the last 6 months, and intentionality of partnership seroconcordance. It includes data from 7950 of 8558 respondent partnerships in which all items were completed regarding HIV status, partner HIV status, UAI, and intentionality of seroconcordance. Based on CDC’s definition of serosorting (intention and seroconcordance and practice of UAI), 17% of all partnerships (16% HIV−, 1% HIV+) indicated serosorting.
Nearly half of all partnerships involved an unknown serostatus (HIVu) respondent or partner: 20% partnerships involved HIVu respondents and any partners (R:HIVu, P:any) and 28% involved known status participants and HIVu partners (R:HIV−, P:HIVu or R:HIV+, P:HIVu). For all partnerships with an HIVu member, seroconcordance or serodiscordance could not have been established at the time of the partnership, and therefore serosorting was not possible. The reason for unknown HIV status differed for HIVu respondents and HIVu partners. Most HIVu respondents (90%) reported never having received an HIV test, but nearly all respondents with HIVu partners reported never discussing their partner’s serostatus (98%), an issue further explored in a separate analysis.30
Levels of UAI were high across all subgroups in the analysis: the lowest level was over one third (37%) for R:HIV−, P:HIVu partnerships, and the highest level was over three quarters (80%) for R:HIV+, P:HIV+ partnerships. For all other partnership groupings, the range of UAI levels was remarkably narrow (43%–50%) given that this figure includes diverse partnerships such as R:HIV−, P:HIV+; R:HIV−, P:HIV−; and R:HIV+, P:HIVu partnerships.
Among HIV− respondent partnerships, known serodiscordance (R:HIV−, P:HIV+) was rare (2%). This level is less than would be expected by chance, given that 10% of the overall sample was HIV+. Although HIV− respondents reported fewer than expected serodiscordant partnerships based on random mixing alone, one third (34%) of their partnerships involved HIVu partners.
Among HIV+ respondent partnerships, fewer were known serodiscordant (R:HIV+, P:HIV−) (29%) than would be expected based on the prevalence of HIV− men (69%) in our sample. Similarly, more HIV+ respondent partnerships were serconcordant (30%) than would be expected based on random partnering based on the 10% HIV prevalence in our sample.
Table 3 shows proportions of partnerships that were seroconcordant and proportions of these seroconcordant partnerships classified as intentionally seroconcordant. Overall, the proportion of seroconcordant partnerships was 64% for HIV− respondents and 30% for HIV+ respondents; within seroconcordant partnerships, intention to choose a seroconcordant partner was 80% for HIV− respondents and 48% for HIV+ respondents. The intersection of seroconcordance and stated intention shows that 51% of HIV− partnerships and 15% of HIV+ partnerships were intentionally seroconcordant.
Demographic and behavioral factors associated with being in a seroconcordant partnership were largely similar for HIV− and HIV+ respondents, except for age. HIV+ respondents aged 40 years or older were significantly more likely to be in seroconcordant partnerships than their 18- to 24-year counterparts [OR: 3.3; 95% confidence interval (CI): 1.5 to 7.0], but age had no significant association with seroconcordance for HIV− partnerships (OR: 1.0; 95% CI: 0.8 to 1.2). Relative to white respondent partnerships, point estimates indicated decreases in odds of seroconcordance for HIV− black (OR: 0.72; 95% CI: 0.58 to 0.89) and HIV+ black (OR: 0.72; 95% CI: 0.44 to 1.2) respondent partnerships. One-time casual partners were less likely to be seroconcordant than main partners for HIV− (OR: 0.43; 95% CI: 0.38 to 0.49) and HIV+ (OR: 0.67; 95% CI: 0.46 to 0.98) respondents. Online partnerships were more likely than offline partnerships to be seroconcordant for HIV− (OR: 2.1; 95% CI: 1.8 to 2.4) and HIV+ (OR: 1.7; 95% CI: 1.2 to 2.5) respondents.
Among HIV− seroconcordant partnerships, intention to be in such a partnership was associated with being aged 40 years or older (OR: 0.56; 95% CI: 0.42 to 0.74) and bisexuality (OR: 1.7; 95% CI: 1.2 to 2.4). Among HIV+ seroconcordant partnerships, intention was associated with participant’s black race (OR: 2.5; 95% CI: 1.1 to 5.8), Hispanic ethnicity (OR: 9.6; 95% CI: 2.0 to 47), and meeting a partner online (OR: 1.9; 95% CI: 1.2 to 2.9). Thus, seroconcordance and intentional seroconcordance shared few demographic associations.
Models of associations between UAI and the independent variable, any seroconcordance or intended seroconcordance, were stratified by participant serostatus (Table 4). For HIV− men, a small but significant association was found between seroconcordance and UAI (OR: 1.2; 95% CI: 1.0 to 1.3; P = 0.03). However, UAI was not associated with intended seroconcordance (OR: 1.1; 95% CI: 0.99 to 1.3). For HIV+ men, seroconcordance was strongly associated with UAI (OR: 3.2; 95% CI: 2.2 to 4.6), and intended seroconcordance was similarly associated (OR: 1.9; 95% CI: 1.3 to 2.9). Across all 4 models, main partnership was strongly associated with UAI. For HIV− men, the association (OR: 5.8; 95% CI: 5.0 to 6.8) was substantially stronger than the association between seroconcordance and UAI described earlier. Indicating a potential dose–response relationship, casual repeated partnerships, compared with 1-time partnerships, were also associated with UAI across the models. The only other significant control variable across models of UAI was ethnicity, with “other” ethnicity members less likely to report UAI than white, non-Hispanics. Sensitivity analyses indicated that using different cutpoints on the Likert scale to dichotomize the serosorting intention variable did not alter the direction or significance of study findings regarding serosorting and its relation to UAI.
We found high levels of stated intention to be in a seroconcordant relationship among HIV− seroconcordant partnerships (80%) and moderate levels (48%) among HIV+ seroconcordant partnerships. For HIV− men, nearly two thirds of all their partnerships were seroconcordant, and in most of that, seroconcordance was intended. High rates of stated intention indicate that seroconcordance was not solely because of social or sexual network structures but, instead, may be the product of a deliberate harm-reduction strategy. Supporting this hypothesis was the rarity of known serodiscordant relationships (2%) among HIV− respondents relative to HIV prevalence in our sample of 10%. Some HIV− men might avoid the cognitive dissonance of known serodiscordant partnerships, instead taking on undefined risk by not ascertaining their partner’s serostatus (34% of HIV− men’s partnerships were with partners of unknown serostatus).
About one third of HIV+ men’s partnerships were seroconcordant, with just under half of these reported as intentionally concordant. Seroconcordant HIV+ partnerships made up a higher proportion of HIV+ men’s partnerships than would be expected by chance, but the absolute level was low (30%). Similar to this study, other research has found seroconcordance among both HIV− and HIV+ men to occur at higher than expected levels.16
Associations between UAI and different measures of serosorting did not support our hypothesis that men who described their seroconcordance as intentional would be more likely to practice UAI. Our findings regarding the limitations of seroconcordance intent in predicting UAI behavior, however, echo a meta-analysis that concluded that, across multiple domains, health behavior intentions accounted for only 22% of the variance in health behaviors.31
Measurement of the “intent to serosort” construct has been fraught with difficulties relating to time order, improperly aggregated behavior, and the degree to which preimposed strategies are formulated and subsequently influence sexual decisions (in addition to other, more common behavioral measurement limitations such as recall bias). Our measures attempted to mitigate problems relating to time order by assessing participant’s knowledge of partner’s HIV status before first sex. Furthermore, our measures sought to mitigate the need for participants to aggregate behavioral strategy across different partnerships, because, in actuality, men may use different strategies in different situations. Yet, our measures were limited to the extent that participants either (1) did not approach sexual partnerships with a conscious, preformulated HIV prevention plan or (2) did not consistently implement within partnerships such HIV-prevention plans. A separate study found that over half of HIV− MSM practicing UAI outside of their main partnership reported that they had no specific HIV risk-reduction strategy32; this may indicate that, for a substantial proportion of individuals, conscious and preformulated plans may not accurately describe their HIV risk-reduction behaviors.
Several possible scenarios could explain why measurement of intention did not lead to stronger associations with UAI. First, although seroconcordance was associated with UAI, intention might have no impact on this relationship. If this were the case, using a stated intention measure would effectively reduce sensitivity of the “any seroconcordance” measure and would misclassify respondents. Another possibility is that our measure of intent is not adequate to detect a true relationship between seroconcordance intentionality and UAI. The most probable scenario is that some combination of the above factors influenced our findings regarding intent and UAI. In addition, it is possible that, despite our large number of respondents and partnerships, we did not have sufficient power to detect differences between the magnitude of the associations between UAI and the 2 definitions of seroconcordance. Despite uncertainty regarding intent, this study yields clear implications for future research. The CDC’s definition of serosorting requiring intent may reduce sensitivity unnecessarily. Upcoming studies should always include a behavioral measure of serosorting, based on seroconcordance and UAI, and not rely solely on identity measures of intentional serosorting.
Regardless of intentionality, seroconcordance was strongly associated with UAI for HIV+ men and weakly associated with UAI for HIV− men, findings similar to those from other studies of serosorting.10,33 Except the small (2%) set of HIV+ seroconcordant partnerships, there was remarkably little variation in levels of UAI across all other partnerships grouped by seroconcordance in the study: 37%–50%. This translated to very different partnership types having similar levels of UAI. This could be not only seen as indicating continued and high overall UAI but also interpreted as a lack of substantial increase in UAI among HIV− men in seroconcordant relationships (45%) relative to HIV− men in serounknown (37%) or serodiscordant (48%) relationships. This finding, combined with failure of intention to improve the association between seroconcordance and UAI, calls into question the plausibility of incorporating serosorting behaviors among HIV− men into HIV-prevention interventions.
Of all factors in our models of UAI, main partnership had the strongest correlation, indicating the importance of relationship commitment as a factor in sexual practices. Data from the NHBS and other surveys of MSM also show main partnership to be highly associated with UAI, a contributor to the estimate that a substantial proportion of incident HIV transmissions occur within main partnerships.24,25 The implication of small effects of seroconcordance and large effects of main partnership on UAI among HIV− men indicates prevention programs are likely to benefit from increased focus on factors that develop during relationships, such as trust and commitment.
Our study had a number of important limitations. Our respondents comprised a sample of Internet-using men who opted into a banner-ad–based survey, making our findings subject to selection bias. Study findings, therefore, are not representative of all US MSM, or all Internet-using MSM. The cross-sectional nature of our data also limited our assessments of relationships to associations. Because we asked participants to report retrospectively on their intent, data are subject to recall bias; it is not clear what impact such bias might have on our results. The median interval from relationship initiation to interview was less than 6 months, however, which somewhat mitigates this concern. In additionally, we relied on self-reported data for all measures, which could result in social desirability bias. We excluded HIVu participants only from analyses directly related to serosorting, because serosorting is not a meaningful strategy without knowledge of one’s own serostatus. Exclusion of this group is nontrivial, because HIVu respondents were younger, had lower income, less educated, and less likely to identify as homosexual than their known serostatus peers. A similar limitation is that all partnerships men avoided, such as an HIV− man avoiding an HIV+ partner, were by definition not captured in our measures of sexual behavior. Despite these limitations, this study’s methodology also has some advantages. First, by conducting an online survey, we were able to collect data from across the United States. Second, our measurement of intent and seroconcordance sought to resolve a long-standing question regarding the definition of serosorting and the HIV prevention significance of intentional versus unintentional seroconcordance.
According to our findings, understanding whether seroconcordance is intentional is not necessary to understand the relationship between seroconcordance and UAI. We found a small but significant association between HIV− partnership seroconcordance and higher levels of UAI. For men living with HIV infection, having an HIV+ partner was strongly associated with UAI, despite the reality that these men less often expressed that their serosorting was intentional than did their HIV− counterparts. These results indicate that intention to be in a seroconcordant partnership may be less determinant of UAI than seroconcordance itself.
3. Eaton LA, Kalichman SC, O’Connell DA, et al.. A strategy for selecting sexual partners believed to pose little/no risks for HIV: serosorting
and its implications for HIV transmission. AIDS Care. 2009;21:1279–1288.
4. Lattimore S, Thornton A, Delpech V, et al.. Changing patterns of sexual risk behavior among London gay men: 1998-2008. Sex Transm Dis. 2011;38:221–229.
5. Bruce D, Harper GW, Suleta K. Sexual risk behavior and risk reduction beliefs among HIV-positive young men who have sex with men. AIDS Behav. 2013;17:1515–1523.
6. Golden MR, Stekler J, Hughes JP, et al.. HIV serosorting
in men who have sex with men: is it safe?. J Acquir Immune Defic Syndr. 2008;49:212–218.
7. Eaton LA, Kalichman SC, Cain DN, et al.. Serosorting
sexual partners and risk for HIV among men who have sex with men. Am J Prev Med. 2007;33:479–485.
8. Velter A, Bouyssou-Michel A, Arnaud A, et al.. Do men who have sex with men use serosorting
with casual partners in France? Results of a nationwide survey (ANRS-EN17-Presse Gay 2004). Euro Surveill. 2009;14(47).
9. Jin F, Crawford J, Prestage GP, et al.. Unprotected anal intercourse, risk reduction behaviours, and subsequent HIV infection in a cohort of homosexual men. AIDS. 2009;23:243–252.
10. Zablotska IB, Imrie J, Prestage G, et al.. Gay men’s current practice of HIV seroconcordant unprotected anal intercourse: serosorting
or seroguessing?. AIDS Care. 2009;21:501–510.
11. Marks G, Millett GA, Bingham T, et al.. Prevalence and protective value of serosorting
and strategic positioning among Black and Latino men who have sex with men. Sex Transm Dis. 2010;37:325–327.
12. Prevalence and awareness of HIV. Infection among men who have sex with men––21 cities, United States, 2008. MMWR. Morb Mort Wkly Rep. 2010;59:1201–1207.
13. van den Boom W, Stolte I, Sandfort T, et al.. Serosorting
and sexual risk behaviour according to different casual partnership types among MSM
: the study of one-night stands and sex buddies. AIDS Care. 2012;24:167–173.
14. Dubois-Arber F, Jeannin A, Lociciro S, et al.. Risk reduction practices in men who have sex with men in Switzerland: serosorting
, strategic positioning, and withdrawal before ejaculation. Arch Sex Behav. 2012;41:1263–1272.
15. McFarland W, Chen YH, Raymond HF, et al.. HIV seroadaptation among individuals, within sexual dyads, and by sexual episodes, men who have sex with men, San Francisco, 2008. AIDS Care. 2011;23:261–268.
16. McFarland W, Chen YH, Nguyen B, et al.. Behavior, intention or chance? A longitudinal study of HIV seroadaptive behaviors, abstinence and condom use. AIDS Behav. 2013;17:1352–1361.
17. Prestage G, Brown G, Down IA, et al.. “It’s hard to know what is a risky or not a risky decision”: gay men’s beliefs about risk during sex. AIDS Behav. 2012.
18. Chen YH, Vallabhaneni S, Raymond HF, et al.. Predictors of serosorting
and intention to serosort among men who have sex with men, San Francisco. AIDS Educ Prev. 2012;24:564–573.
19. Philip SS, Yu X, Donnell D, et al.. Serosorting
is associated with a decreased risk of HIV seroconversion in the EXPLORE Study Cohort. PLoS One. 2010;5(9).
20. Golden MR, Dombrowski JC, Kerani RP, et al.. Failure of serosorting
to protect African American men who have sex with men from HIV infection. Sex Transm Dis. 2012;39:659–664.
21. Khosropour CK, Sullivan PS. Mobile Phone-based Data Collection to Enhance Retention of Racial/ethnic Minorities in a Longitudinal Internet-based HIV Behavioral Risk Study of MSM
in the United States. Paper presented at: 6th IAS Conference on HIV Pathogenesis, Treatment, and Prevention; July 17–20, 2011; Rome, Italy.
22. Rosenberg ES, Khosropour CK, Sullivan PS. Heterogeneous racial differences in disclosure of HIV status by serostatus and partnership sexual risk among US MSM
. Paper presented at: The 6th IAS Conference on HIV Pathogenesis, Treatment, and Prevention; July 17–20, Rome, Italy.
23. Gallagher KM, Sullivan PS, Lansky A, et al.. Behavioral surveillance among people at risk for HIV infection in the U.S.: the National HIV Behavioral Surveillance System. Public Health Rep. 2007;122(suppl 1):32–38.
24. Goodreau SM, Carnegie NB, Vittinghoff E, et al.. What drives the US and Peruvian HIV epidemics in men who have sex with men (MSM
)?. PLoS One. 2012;7:e50522.
25. Sullivan PS, Salazar L, Buchbinder S, et al.. Estimating the proportion of HIV transmissions from main sex partners among men who have sex with men in five US cities. AIDS. 2009;23:1153–1162.
26. Beyrer C. HIV epidemiology update and transmission factors: risks and risk contexts––16th International AIDS Conference epidemiology plenary. Clin Infect Dis. 2007;44:981–987.
27. Grov C, DeBusk JA, Bimbi DS, et al.. Barebacking, the Internet, and harm reduction: an intercept survey with gay and bisexual men in Los Angeles and New York City. AIDS Behav. 2007;11:527–536.
28. Crepaz N, Marks G, Liau A, et al.. Prevalence of unprotected anal intercourse among HIV-diagnosed MSM
in the United States: a meta-analysis. AIDS. 2009;23:1617–1629.
29. Stata Statistical Software: Release 11 [computer program]. College Station, TX: StataCorp LP; 2011.
30. Winter AK, Sullivan PS, Khosropour CM, et al.. Discussion of HIV status by serostatus and partnership sexual risk among internet-using MSM
in the United States. J Acquir Immune Defic Syndr. 2012;60:525–529.
31. Armitage CJ, Conner M. Efficacy of the theory of planned behaviour: a meta-analytic review. Br J Soc Psychol. 2001;40(Pt 4):471–499.
32. Mitchell JW. HIV-Negative and HIV-discordant gay male couples’ use of HIV risk-reduction strategies: differences by partner type and couples’ HIV-status. AIDS Behav. May 2013;17:1557–1569.
33. McConnell JJ, Bragg L, Shiboski S, et al.. Sexual seroadaptation: lessons for prevention and sex research from a cohort of HIV-positive men who have sex with men. PLoS One. 2010;5:e8831.
Keywords:© 2013 by Lippincott Williams & Wilkins
HIV prevention; MSM; serosorting; measurement