Study design
We conducted a secondary analysis of HIV surveillance data collected from April 2008 to September 2009 in three cities – Lima, Iquitos, and Pucallpa. Sentinel surveillance surveys have been regularly administered to high-risk MSM and TW in Peru since 1996 to assess emerging trends in the HIV and STI epidemics [2].
Participant selection
Participants were recruited through “snowball” sampling techniques as well as by trained peer outreach workers who represented diverse MSM sub-cultures. Recruitment strategies included the use of posters, distribution of flyers, and informational meetings at previously-mapped venues in participating cities. Potential participants were referred by outreach workers to sentinel study sites where they were enrolled in the study. Participants were provided with condoms and reimbursement for the cost of transportation.
Enrollment was limited to anatomically born males at least 18 years old who reported anal sex with a male or male-to-female transgender partner in the past 6 months, who had not been tested for HIV in the previous 12 months, and who had never had a positive HIV test result, and who resided in one of the participating cities. Enrollment was also limited to subjects who reported at least one of the following high-risk sexual behaviors: insertive or receptive anal intercourse without a condom during the last sexual episode; insertive or receptive anal intercourse with more than 5 partners in the last 6 months; receipt of money, drugs, gifts, or accommodation in exchange for sex in the last 6 months; STI diagnosis in the last 6 months; or an HIV positive partner in the last 6 months.
Persons were excluded if they were HIV positive by self-report, previously enrolled in an HIV vaccine or clinical trial, or had a mental or psychiatric condition that prevented them from providing informed consent. Of the 899 participants who were enrolled in the surveillance study, our analysis excluded those who did not report a male sexual partner within the preceding 3 months. The final number of participants included in our analysis was 735.
Data collection
All participants completed an anonymous survey with the use of CASI (Computer Assisted Self-Interviewing). The first part of the survey included sections on basic demographics, knowledge of HIV/STIs, history of STIs (past 6 months), access and frequency of use of condoms and lubricants, sexual role and identity, exchange sex activity (past 6 months), number of male and female partners (past 3 months), and alcohol and drug use (past month). The remaining sections asked an identical set of partnership and event-level questions about the last sexual encounter with each of the participant’s three most recent partners in the past 3 months.
The surveillance protocol received ethics approval for human subjects research from the Asociación Civil Impacta Salud y Educación (IMPACTA) Institutional Bioethics Committee and the U.S. Naval Medical Research Center Detachment (NMRCD, now Naval Medical Research Unit-6) Institutional Review Board (IRB). Secondary analysis of deidentified data was considered exempt from review by the Office of the Human Research Protection Program of the University of California, Los Angeles.
Analysis
The primary outcome for our analysis was whether the participant knew their partner’s serostatus. If participants responded that their partner was HIV-positive or HIV-negative, they were categorized as having knowledge of their partner’s serostatus. If they reported that the partner had not been tested, or if the partner’s test result was not known, they were categorized as not having knowledge of their partner’s serostatus.
We developed a conceptual framework for our multivariate model prior to conducting the analysis. An extensive review of the literature and available data from the survey database guided our choice of predictor variables for our conceptual framework. As such, the final multivariate model included some variables which were not significantly associated with the primary outcome in bivariate analyses. Variables at both the participant level (age, education, sexual identity, number of male partners) and the partnership level (type and length of partnership, participant use of alcohol during sexual intercourse with the partner) were included, but many others, such as participating city, could not be included in the final multivariate model because of potential instability of the model. The inadequate sample size of TW precluded a separate analysis for this group.
Since the dependent variable - knowledge of partner HIV serostatus – was a partnership-level variable, and each participant reported data on up to three partnerships, data aggregation would violate the assumption of total independence between units under analysis (those units being partnerships, not participants). Therefore, bivariate and multivariate logistic regression analyses were conducted by clustering results to account for correlation of data. Units (partnerships) were clustered by using a variable that uniquely identified each participant. Crude and adjusted odds ratios (OR) are presented with 95% confidence intervals. We considered values statistically significant if the 95% confidence interval did not include 1.
We also analyzed the association between knowledge of a sex partner’s HIV serostatus and unprotected anal intercourse (UAI) with that partner as well as the association between the partner’s serostatus itself and the likelihood of unprotected anal intercourse . Sexual behavior at the last sexual encounter with each partner was dichotomized into unprotected anal intercourse (UAI) and non-UAI, which included anal intercourse with a condom as well as any non-penetrative forms of sex. Multivariate analysis using logistic regression clustered by participant adjusted for the effects of the same predictor variables. All data analyses were conducted using Stata 11.2 (College Station, TX).