Syphilis and male rectal gonorrhoea continued to rise even as HIV incidence levelled off among MSM. Our data support the hypothesis that MSM are increasingly selecting HIV seroconcordant partners when engaging in UAI. In the SAP data, although any UAI increased overall, UAI with partners of unknown serostatus declined among both HIV positive and HIV negative MSM. Behavioural data from ATS and the STI clinic showed an overall decrease in any UAI with known HIV positive partners during the same period. Overall prevalence of UAI with known HIV positive partners was similar between ATS (11.6%) and the STI clinic (11.0%). The decrease in UAI with HIV positive partners in recent years reflected a decline in both receptive and insertive UAI. Receptive UAI is thought to impart a higher HIV acquisition risk compared to insertive UAI, and the choice of HIV negative people as the insertive partner, referred to as strategic positioning (or “seropositioning”), may reflect a second strategy to decrease HIV risk.28,29
In aggregate, our data suggest the role of serosorting in preventing further increases in HIV transmission among MSM in San Francisco following a period of resurgent risk behaviour, STI, and HIV incidence.1,2,3,4,9,22
HIV incidence from two MSM populations seeking HIV testing remained at a relative plateau through 2004. While any UAI continued to increase overall, UAI with partners of unknown HIV serostatus has decreased since 2001. Our findings may be compared to data from Sydney where after 6 years of increases, HIV negative MSM reported decreases in UAI in 2003.30
Decreases in serodiscordant UAI suggest a recent shift in San Francisco's prevention efforts may be working. The late 1990s saw initiation of “prevention for positives” programmes targeting secondary transmission.31,32
Prevention for positives activities included media campaigns, prevention messages integrated into HIV care, and support groups.
Other factors may contribute to the increase in serosorting. The internet may facilitate partner selection according to HIV serostatus by providing a semi‐anonymous environment for disclosure.33,34,35,36,37,38,39,40
The internet has become the most common place where MSM diagnosed with STIs met their partners.41,42
Internet use was associated with a recent syphilis epidemic in San Francisco, where in 2002, 37.4% of MSM with early syphilis reported meeting partners on the internet compared to 12.2% in 2000.43
Widespread use of highly active antiretroviral therapy (HAART) may also contribute to the levelling off in HIV incidence by reducing the per contact probability of transmission.1,44,45,46,47,48,49
Mathematical models suggest a reduction in infectivity by HAART is in dynamic balance with risk behaviour.44,45,47
One model predicted if approximately 50% of HIV infected individuals receive HAART, a 20% or more increase in risky behaviour is likely to increase incidence of infection in the population.44
The model predicted population level HIV incidence would rise with HAART uptake in the San Francisco MSM community, but would subsequently fall as more HIV infected people received treatment or as risk behaviour decreased. A model to assess HIV infection among MSM in Amsterdam predicted even a 100% increase in risky behaviour with steady partners would not outweigh the protective effect of HAART if there were concurrent increases in HIV testing from 42% to 80%, and HAART coverage from 70% to 85%.48
The present study presents empirical data fulfilling these forecasts. HAART use rose rapidly from 1995 to 2001 but since has levelled off at a high coverage.1,49,50,51
We acknowledge alternative hypotheses could explain the current trends in STI, UAI, and HIV incidence. One possible explanation for the apparent discrepancy in STI and HIV incidence is that HAART has suppressed HIV transmission but not bacterial STI transmission.52
Another hypothesis is that transmission largely occurs immediately following HIV acquisition when viral loads are high and people unaware of their infection are in sexual networks of concurrent partnerships.53
In this hypothesis, serosorting by known HIV positive people would not have an important role in the epidemic. We believe these hypotheses are not mutually exclusive. HIV serosorting may account for some of the current trends in HIV and STI incidence or act in synergy with other causes.
We recognise limitations in our study. Firstly, the ecological nature of the analysis makes causal inferences uncertain. A limitation of the second generation HIV surveillance approach of using multiple data sources is that the temporal sequence of changes in trends is uncertain because of differences in the way data are collected across studies. Ideally, one would expect to observe decreases in sexual risk behaviour before declines in STI and HIV incidence. In our different datasets, however, HIV incidence peaked in 1999, UAI with a known HIV positive partner peaked in 1999 at ATS and in 1998 at the STI clinic, and UAI with unknown serostatus partners peaked in 2001. Another limitation is the data do not record biological and behavioural information from the same individuals, but rather from incompletely overlapping samples of MSM. A longitudinal cohort study of the same individuals would address this limitation; however, such studies have other potential biases and generalisation is problematic.54,55,56
None the less, evidence of a rise in STI, but not HIV, in an Amsterdam cohort of young MSM complements our findings.55
The potential biases of each data source must be considered separately. Part of our data include people seeking HIV testing.1,2
SAP data include MSM conveniently sampled at selected gay identified venues.1,2
HIV testing sites and gay bars may attract people at relatively higher risk for HIV. As with other serial cross sectional studies, we assume selection biases operating at the same locations remain relatively stable over time and therefore have less impact on interpreting trends over time than interpreting data at one point in time. A strength of our analysis is that we have several years of data collected consistently at the same locations. This is the basis for HIV sentinel surveillance as well as a principle of second generation HIV surveillance.21
Behavioural data rely on self report of sensitive information through brief public interviews. Moreover, there is great uncertainty around knowledge of partners' HIV serostatus. Another limitation is in the interpretation of statistical tests, where an unknown proportion of subjects may be repeatedly included. When we excluded subjects known to be repeat observations, the main trends and findings did not change. The issues of under‐diagnosis, under‐reporting of cases, reporting delays, and changes in screening practices pose limitations to STI surveillance data. STARHS may not provide accurate estimates of HIV incidence because of severe immunosuppression, HAART use, or laboratory error.25,26
Finally, because data were collected during the course of clinical care or prevention activities, the level of detail on risk behaviour from all sources is limited.
On the other hand, drawing inference from multiple studies using different methods and populations may mitigate some of the limitations in interpreting each study separately. This is a primary strength of the second generation HIV surveillance approach.21
The same approach was quick to detect a rise in HIV risk in San Francisco several years ago, a trend corroborated by studies of MSM elsewhere in North America, Europe, and Australia.1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17
The present data suggest a new phase of the epidemic where increased HIV serosorting may offset further expansion of HIV incidence. However, serosorting is vulnerable to imperfect knowledge of one's own and one's partners' HIV serostatus and the transmission probabilities of sexual practices.28
A serosorting strategy may be more feasible in San Francisco because of the high level of HIV testing among MSM. A recent survey found 97% of MSM in San Francisco had ever tested and 34% had tested in the past 6 months.57
Although the data indicate HIV serosorting is a possible prevention strategy, it remains a hypothesis that needs validation in a specifically designed study. Such a study would need to measure true intentions to engage in HIV serosorting as a deliberate prevention strategy and longitudinal follow up would be needed to establish causality. While the findings of our analysis may point to potential further decreases in HIV transmission, HIV incidence among MSM in San Francisco has yet to decline to the point where the epidemic will be extinguished in the near future.
- Sexually transmitted infections are increasing among men who have sex with men (MSM)
- Sexual risk behaviour—that is, unprotected anal intercourse is increasing among MSM
- HIV incidence has levelled off among MSM
- These findings may be explained by the increasing practice of “HIV serosorting”—that is, MSM are selecting sexual partners of the same HIV serostatus