Background: Men who have sex with men (MSM) remain at great risk for HIV infection. Program planners and policy makers need descriptions of interventions and quantitative estimates of intervention effects to make informed decisions concerning prevention funding and research. The number of intervention strategies for MSM that have been examined with strong research designs has increased substantially in the past few years.
Objectives: 1. To locate and describe outcome studies evaluating the effects of behavioral HIV prevention interventions for MSM.2. To summarize the effectiveness of these interventions in reducing unprotected anal sex.3. To identify study characteristics associated with effectiveness.4. To identify gaps and indicate future research, policy, and practice needs.
Search strategy: We searched electronic databases, current journals, manuscripts submitted by researchers, bibliographies of relevant articles, conference proceedings, and other reviews for published and unpublished reports from 1988 through December 2007. We also asked researchers working in HIV prevention about new and ongoing studies.
Selection criteria: Studies were considered in scope if they examined the effects of behavioral interventions aimed at reducing risk for HIV or STD transmission among MSM. We reviewed studies in scope for criteria of outcome relevance (measurement of at least one of a list of behavioral or biologic outcomes, e.g., unprotected sex or incidence of HIV infections) and methodologic rigor (randomized controlled trials or certain strong quasi-experimental designs with comparison groups).
Data collection and analysis: We used fixed and random effects models to summarize rate ratios (RR) comparing intervention and control groups with respect to count outcomes (number of occasions of or partners for unprotected anal sex), and corresponding prevalence ratios (PR) for dichotomous outcomes (any unprotected anal sex vs. none). We used published formulas to convert effect sizes and their variances for count and dichotomous outcomes where necessary. We accounted for intraclass correlation (ICC) in community-level studies and adjusted for baseline conditions in all studies. We present separate results by intervention format (small group, individual, or community-level) and by type of intervention delivered to the comparison group (minimal or no HIV prevention in the comparison condition versus standard or other HIV prevention in the comparison condition). We examine rate ratios stratified according to characteristics of participants, design, implementation, and intervention content. For small group and individual-level interventions we used a stepwise selection process to identify a multivariable model of predictors of reduction in occasions of or partners for unprotected anal sex. We used funnel plots to examine publication bias, and Q (a chi-squared statistic with degrees of freedom = number of interventions minus 1) to test for heterogeneity.
Main results: We found 44 studies evaluating 58 interventions with 18,585 participants. Formats included 26 small group interventions, 21 individual-level interventions, and 11 community-level interventions. Sixteen of the 58 interventions focused on HIV-positives. The 40 interventions that were measured against minimal to no HIV prevention intervention reduced occasions of or partners for unprotected anal sex by 27% (95% confidence interval [CI] = 15% to 37%). The other 18 interventions reduced unprotected anal sex by 17% beyond changes observed in standard or other interventions (CI = 5% to 27%). Intervention effects were statistically homogeneous, and no independent variable was statistically significantly associated with intervention effects at alpha=.05. However, a multivariable model selected by backward stepwise elimination identified four study characteristics associated with reduction in occasions of or partners for unprotected anal sex among small group and individual-level interventions at alpha=.10. The most favorable reductions in episodes of or partners for unprotected anal sex (33% to 35% decreases) were observed among studies with count outcomes, those with shorter intervention spans (<=1 month), those with better retention in the intervention condition than in the comparison condition, and those with minimal to no HIV prevention intervention delivered to the comparison condition. Because there were only 11 community-level studies we did not search for a multivariable model for community-level interventions. In stratified analyses including only one variable at a time, the greatest reductions (40% to 54% decreases) in number of episodes of or partners for unprotected anal sex among community-level interventions were observed among studies where groups were assigned randomly rather than by convenience, studies with shorter recall periods and longer follow-up, studies with more than 25% non-gay identifying MSM, studies in which at least 90% of participants were white, and studies in which the intervention addressed development of personal skills.
Authors' conclusions: Behavioral interventions reduce self-reported unprotected anal sex among MSM. These results indicate that HIV prevention for this population can work and should be supported. Results of previous studies provide a benchmark for expectations in new studies. Meta-analysis can inform future design and implementation in terms of sample size, target populations, settings, goals for process measures, and intervention content. When effects differ by design variables, which are deliberately selected and planned, awareness of these characteristics may be beneficial to future designs. Researchers designing future small group and individual-level studies should keep in mind that to date, effects of the greatest magnitude have been observed in studies that used count outcomes and a shorter intervention span (up to 1 month). Among small group and individual-level studies, effects were also greatest when the comparison condition included minimal to no HIV prevention content. Nevertheless, statistically significant favorable effects were also seen when the comparison condition included standard or other HIV prevention content. Researchers choosing the latter option for new studies should plan for larger sample sizes based on the smaller expected net intervention effect noted above. When effects differ by implementation variables, which become evident as the study is conducted but are not usually selected or planned, caution may be advised so that future studies can reduce bias. Because intervention effects were somewhat stronger (though not statistically significantly so) in studies with a greater attrition in the comparison condition, differential retention may be a threat to validity. Extra effort should be given to retaining participants in comparison conditions. Among community-level interventions, intervention effects were strongest among studies with random assignment of groups or communities. Therefore the inclusion of studies where assignment of groups or communities was by convenience did not exaggerate the summary effect. The greater effectiveness of interventions including more than 25% non-gay identifying MSM suggests that when they can be reached, these men may be more responsive than gay-identified men to risk reduction efforts. Non-gay identified MSM may have had less exposure to previous prevention messages, so their initial exposure may have a greater impact. The greater effectiveness of interventions that include efforts to promote personal skills such as keeping condoms available and behavioral self-management indicates that such content merits strong consideration in development and delivery of new interventions for MSM. And the finding that interventions were most effective for majority white populations underscores the critical need for effective interventions for MSM of African and Latino descent. Further research measuring the incidence of HIV and other STDs is needed. Because most studies were conducted among mostly white men in the US and Europe, more evaluations of interventions are needed for African American and Hispanic MSM as well as MSM in the developing world. More research is also needed to further clarify which behavioral strategies (e.g., reducing unprotected anal sex, having oral sex instead of anal sex, reducing number of partners, avoiding serodiscordant partners, strategic positioning, or reducing anal sex even with condom use) are most effective in reducing transmission among MSM, the messages most effective in promoting these behaviors, and the methods and settings in which these messages can be most effectively delivered.