The survey protocol and consent procedures were approved by the Uganda Virus Research Institute's institutional review board, the Uganda National Council of Science and Technology, and the CDC. The survey was conducted anonymously; informed consent was obtained verbally; no personal identifiers were collected.
This survey was cross-sectional and used RDS. The survey is part of the “Crane Survey”, a joint activity by Makerere University, the Ministry of Health (MoH), and the US Centers for Disease Control and Prevention (CDC). Prior to the survey, formative research was carried out over several weeks through approximately a dozen key informant interviews using semi-structured interview guides or open discussions to inform sampling design, seed identification, social connectedness among MSM in Kampala, number and location of survey offices, protection of privacy, compensation, data and biological measures, language, and other procedures. Key findings included that RDS was recommended and preferred over time-location sampling, that a single survey office would suffice, that anonymity was paramount, and that biomarker results needed to be returned and addressed. We conducted the survey between May 2008 and April 2009.
The survey was carried out in Kampala, Uganda's capital and largest city with approximately 1.8 million residents.
Inclusion criteria were male sex, age ≥18 years, residence in greater Kampala, and self-reported anal sex with another man in the preceding 3 months. Exclusion criteria included coupon receipt from a stranger and language barriers.
RDS methodology is well described elsewhere 
and represents an advanced version of chain referral sampling. We initiated sampling with eight seeds purposively selected by age, HIV status, and geographic location in Kampala; all were socially well networked. After seeds began their recruiting efforts, candidate recruits presented the coupons they had received to the single survey office in Old Kampala, near the city center. Sampling was interrupted twice; first as a result of the arrest of three local lesbian, gay, bisexual, and transgender (LGBT) activists at the PEPFAR Implementer's Meeting in Kampala in early June 2008 
. Sampling remained at a low level until it rebounded in August, but was again affected in September 2008 by arrests of alleged homosexuals 
. A total of six additional seeds were added during the survey. However, following the second round of arrests, sampling rates remained low until we stopped sampling in March 2009, close to the survey's scheduled end.
We issued 1,706 coupons and redeemed 455 of these. The number of coupons issued per recruit ranged from two at the survey's beginning to six after the arrests and slump in sampling. A total of 300 eligible recruits (n
286) and seeds (n
14) participated in the survey; the remainder were ineligible, mostly because they did not report anal sex with men in the last 3 months. The longest recruitment wave was 11; equilibrium for HIV serostatus was reached after wave number 2, and that for nationality, age, and condom use at last sex was reached after wave number 3, 4, and 4, respectively, the computed
design effect (using actual survey data) for HIV infection was 2.3.
Survey office procedures
Candidate participants were screened for eligibility face-to-face by survey staff. Following a briefing about specific interview terms, such as the definition of sexual intercourse, frequency of sex, partner types, or commercial sex and a short computer-based tutorial about audio-computer-assisted self-interviewing (ACASI), enrolled participants underwent a standardized interview using Questionnaire Design Studio (QDS v2.5) software (NOVA, Bethesda, Maryland, USA). A small number of recruits preferred a computer-assisted personal interview with trained staff using the same QDS instrument. Following the interview and pre-test counseling, recruits provided venous blood and urine samples, and had rectal swabs collected. At the end of the first visit, recruits received instructions and coupons for peer recruitment.
Recruits were scheduled to return to the survey office two weeks later. Survey staff post-test counseled recruits for all biomarkers measured, provided treatment for non-viral sexually transmitted infections (STI) according to World Health Organization and MOH guidelines, and referred HIV-positive recruits to health care providers with whom we had made arrangements prior to survey start. Recruits who failed to return to the survey office could not be reached by survey staff as no personal identifiers were collected. Recruits who returned to the survey office when not all biomarker results were yet available were asked (and compensated for) to return for an additional visit.
At both visits to the survey office, we compensated recruits for their time and transport costs (US$3.00), and, at the return visit, recruitment efforts (US$1.00 per successfully recruited eligible peer). At the time of the survey, US $3.00 could purchase three kilograms of sugar.
The main interview's key domains included demographics, life-time sexual characteristics, sexual behaviors in the last three months, sexual violence, and STDs. Condom use was examined both quantitatively (number of protected sex acts divided by all sex acts in last three months), as well as qualitatively (condom use ever or never, condom use at last sex). Lifetime exposure to homophobic abuse was measured through the question “Did you ever suffer any violence or abuse because you have sex with other men?”. Respondents who affirmed such abuse, where probed about the type of abuse, including moral abuse (isolation or exclusion), verbal abuse (threats or insults), mistreatment, or having been subjected to physical or sexual violence. Respondents were also probed about blackmail (“Have you ever been blackmailed by someone because you have sex with other men?”) and rape (“Were you ever forced to have sex against your will?”). Other data measures included alcohol use, drug use (including injection drug use), as well as buying sex (defined as paying for sex with money, goods or services) or selling sex (defined as giving sex in exchange for money, goods or services).
Laboratory testing was performed off-site at the STD Reference Laboratory based at Mulago Hospital, Kampala; a small number of tests were also performed at the CDC laboratory in Entebbe, Uganda. Blood specimens were stored at 2 to 8°C at the survey office and transported twice daily to the laboratory. Testing for antibodies against HIV was performed through a parallel testing algorithm using Vironostika® HIV Uniform II plus O2 (bioMeriéux, Marcy l'Etoile, France) and Murex® HIV Ag/Ab Combination (Abbott Laboratories, Abbott Park, Illinois, U.S.A.); discordant results were resolved through the use of HIV 1/2 STAT-PAK rapid test (Inverness Medical, Princeton, New Jersey, U.S.A.). All recruits enrolled into the survey accepted HIV testing. Plasma was also tested for Treponema pallidum (TP) infection, using the Anti-syphilis IgG ELISA (Biotec Laboratories, Suffolk, UK) for screening and, if reactive, the Rapid Plasma Reagin Syfacard-R Test (Murex Biotech, Dartford, UK) to detect current TP infection. Urine specimens and rectal swabs were tested for the presence of Neisseria gonnorhea (NG) and Chlamydia trachomatis (CT) DNA (Cobas Amplicor or Amplicor PCR, Roche Diagnostics, Branchburg, New Jersey, U.S.A.).
Data management and analysis
For sample size calculations, we assumed an HIV prevalence of 14%, 95% confidence intervals (CI) of 10.4%–18.3% (approximately twice that of urban men in general 
), and a design effect of 2 
. Aiming for an effective sample size of 300, we adjusted the target sample size to 600.
Survey events (enrollments, recruiter-recruitee links, coupons numbers issued, unique codes, etc.) were tracked with an in-house developed software. Interview data were checked for errors and inconsistencies, and cleaned after importation from QDS into Statistical Analysis Software - SAS v9.2 (SAS Institute, Gary, North Carolina).
Of the 300 eligible respondents, 5 provided little interview data and were excluded from this analysis. Our principal outcome of interest was HIV infection; predictor variables included demographics, sexual orientation and experience, sexual behavior in the 3 months preceding the interview, HIV testing (history), sexual violence, abuse, or blackmail, as well as laboratory markers of sexually transmitted infections and STD symptoms. Condom use over the last three months includes both male and female sex partners (unless stated otherwise) and was computed as a continuous variable with the denominator being the number of sex acts and the numerator being the number of sex acts protected by condoms. We examined condom use as a categorical variable, indicating whether in the preceding three months condoms were used for less than 33% of all sex acts, for 33%–66% of sex acts, or for more than 66% of sex acts.
We present weighted data except for continuous data; univariate analyses were conducted in RDSAT version 6.0.1. (www.respondentdrivensampling.org
). Individual HIV sampling weights were generated in RDSAT and exported to STATA. Using HIV weights imported from RDSAT, logistic regression was conducted for bivariate and multivariate analysis in STATA 10.0 (Stata Corporation, College Station, Texas). For multivariate analysis we employed backward elimination using predictor variables associated with HIV infection at a level of P
≤0.2 in bivariate analysis. The final weighted model displays all predictor variables significantly associated with HIV infection at a level of P
Human subjects considerations
Using scanners and Griaule software (Griaule Biometrics, San Jose, CA, USA), recruits' fingerprints were imaged (but not stored) to generate unique alphanumeric codes and thus facilitate linking recruits' return visits to their initial visits and laboratory results, to detect duplicate recruits (recruits attempting to enroll multiple times), and to detect recruits presenting coupons that had already been issued to other recruits. In addition to this survey, we sampled four other groups concurrently using the same survey office and hence masked the group identity of any survey respondent visiting in the survey office.