Factors associated with sexual risk behaviors with non-steady partners and lack of recent HIV testing among German men who have sex with men in steady relationships: results from a cross-sectional internet survey

Analyses were conducted on data from the 2013 SMA study, an online, cross-sectional study of MSM in Germany funded by the Bundeszentrale für gesundheitliche Aufklärung (BZgA; “Federal Center for Health Education”) [10]. Participants were recruited from various dating sites aimed at gay men and other MSM. After completing the survey, participants could choose to accept a voucher for a free HIV test, redeemable at select testing sites in several larger cities across Germany. Overall, 16,734 men responded to the survey.

The online survey protocol was evaluated and approved by the ethical review board of the Charité University Clinic in Berlin (EA1/266/13). Suggestions by the data protection office of the federal state of Berlin to improve data protection for survey participants were implemented.

We limited our analyses to those participants who reported being in a steady partnership with at least one man. Specifically, participants could indicate being in a monogamous relationship (where no sex with other partners was permitted), being in an open relationship (where sex with other partners was permitted, with or without certain conditions), or having not discussed this with their steady partner. This information was only collected from participants who reported that their HIV-status and the HIV-status of their steady partner were both either negative or unknown. Because the survey asked participants about behaviors within the past 12 months, we also excluded any individuals who did not report being in their current relationship for at least one year. This was done to ensure that any sexual risk or testing behaviors reported did indeed occur while the participant was in a steady relationship, and not before relationship formation. Finally, we considered only those individuals who reported having at least one non-steady anal sex partner in the past 12 months. Although participants adhering to an agreement that does not permit outside partners or only engaging in non-penetrative sexual activities with outside partners are technically also following a negotiated safety regime [8], we were particularly interested in determining those factors associated with safety (condom use and/or testing) among men who engaged in anal sex with non-steady partners. This left us with 1731 participants.

We then developed a binary outcome variable. Individuals in the high-risk outcome group reported UAI with at least one non-steady partner, as well as no HIV test, in the past 12 months. These individuals were compared to a reference group consisting of all remaining individuals in the study sample described above. Although many individuals in this reference group do not fulfill both negotiated safety conditions, this approach allowed us to assess those factors specifically associated with the highest level of risk for HIV transmission within steady partnerships.

Potential predictor variables for our outcome measure included a variety of factors found to be significantly associated with UAI or with lack of HIV testing in previous studies, or else considered to be biologically plausible causal factors for either of these behaviors. We considered: sociodemographic characteristics (age [11], income [11, 12], education [13], size of city of residence, residence in a state in the former Western or Eastern part of Germany, immigration background, sexual orientation); relationship characteristics (type and length of relationship [2, 14, 15], knowledge of partner’s HIV status [15], frequency of UAI with steady partner [13]); psychological and social variables (symptoms or diagnosis of depression or anxiety [15], any professional help for psychological symptoms, internalized homophobia [15, 16], experience of violence due to sexual orientation, “outness” [12], social support [11], HIV-related stigma [17]); use of establishments (such as bars, clubs, or saunas) that cater toward gay men [13]; proportion of sexual partners met on the Internet [18]; knowledge and attitudes toward HIV [19], condom use [14, 15], and antiretroviral therapy (ART) [12, 20]; satisfaction with sex life in general; and substance use [12, 13, 15]. All variables were self-reported. More specifically, “outness” was assessed by asking participants to report the proportion of their family, heterosexual friends, and colleagues who knew their sexual orientation, as well as whether or not their primary doctor was aware that the participant was attracted to men. Participants’ level of social support was assessed by asking participants a) with how many people they felt able to talk about serious personal problems, b) how much interest friends and acquaintances generally show in their lives, and c) how easy it is for them to receive help from family and friends when needed. Responses were reported on a 5-point scale and combined. Internalized homophobia was assessed using an 8-item scale, and HIV-related stigma was assessed using two separate 6-item scales, on which participants could report varying levels of agreement or disagreement. Attitudes toward ART were assessed using a similar methodology. Finally, participant attitudes toward condom use were obtained by asking participants the extent to which they felt condoms disrupted sex, as well as the extent to which they believed partners would make assumptions concerning their HIV status should they insist on condom use. Variables are described in more detail in the Additional file 1.

We began our analyses by calculating descriptive statistics for our reduced dataset. We then identified candidate variables for our multivariate model by conducting bivariate logistic regression analyses.

For the multivariate regression model, only those variables that were significantly associated (p < 0.05) with the outcome risk behavior in bivariate analyses were considered. Before beginning model selection procedures, any variables with very high rates of missing values were removed from consideration. Relationship type was included in the model from the beginning in order to control for this variable throughout the model-fitting process. Variables were iteratively considered for inclusion or exclusion manually, guided by the add1 and drop1 functions in the R package “stats” [21]. Once all relevant main effects were included, interactions between variables were considered. This process was continued until model fit could not be significantly improved by adding or dropping any variables or interactions. Finally, we observed results for predictors with multiple levels, and combined levels if the direction and magnitude of their association with the outcome variable was similar. The final multivariate model was assessed for fit by observing Pearson and deviance residuals as well as cooks distances. Multicollinearity was explored using variance inflation factors (VIF).

To gain additional insight into novel prevention strategies that may be accepted by and effective among the outcome group, variables found to be significant in bivariate analyses but not judged to be relevant predictors for the main multivariate model were added manually.

We concluded our analyses by assessing the reasons reported by participants for never testing for HIV, or for not testing in the past five years. Similar reasons were combined as described in the Additional file 1. We also observed reasons reported for not accepting the free HIV test voucher at the end of the survey.