Study setting and population
A detailed description of the study population and methods is available elsewhere; a brief overview is provided here [1, 19]. This study is a secondary analysis of a randomized controlled trial (RCT) that evaluated the impact of conditional economic incentives on staying free of new curable STIs among MSWs in Mexico City. We conducted this secondary analysis because there is a dearth of information about prevalence and incidence of STIs among MSWs. The RCT was not powered to analyze the effects on incident STI/HIV by study arm. Thus, we report here the incidence of STI/HIV and its determinants for the entire sample. This study took place from January 2012 to May 2014. Participants were recruited by trained research staff from community sites where MSWs were known to congregate in Mexico City, as determined in previous studies [19, 20]. Participants were also recruited through referral to the research team from within the Condesa HIV Testing Clinic. Participants were tested and treated for STIs, as indicated, at Clínica Condesa. Treatment was provided free of charge, including antiretroviral treatment for those identified as HIV-positive. All participants provided informed consent. All procedures were approved by Institutional Review Boards at Brown University in Providence, United States of America, and the National Institute of Public Health in Cuernavaca, Mexico.
The study population consisted of a convenience sample of 227 cisgender men, ages 18–40, who attended a clinic appointment at Clinic Condesa, and who either self-identified as MSWs (n = 152) or who did not self-identify as a MSWs, but who declared that they were a man who had sex with a male partner in exchange for money in the past six months and who had at least 10 male sexual partners within the last month (n = 75). Note that these criteria for inclusion in the study allow for participants that have sex with women in addition to men (hence, some study variables include vaginal intercourse), but who nevertheless meet the study’s definition of male sex worker. These criteria were determined based a previous study involving observations and in-depth interviews with sex workers in Mexico City [20]. Transgender women were excluded from the present study because Clínica Condesa has a separate program for them.
At the baseline visit, participants filled out a survey with questions regarding sociodemographic characteristics and health behaviors. At baseline (0 months) and follow-up visits one (6 months) and two (12 months), participants filled out the survey again and were tested and treated (as indicated) for syphilis, chlamydia, gonorrhea, and HIV.
Data collection and measures
Data collection was done in partnership with the Mexican National Institute of Public Health (INSP) and the Consortium for HIV/AIDS and Tuberculosis Research (CISIDAT). Participants were administered the survey using laptop computers with audio computer assisted interviewing (A-CASI) questionnaires. All variables were assessed for missingness, range, and distribution. Blood and urine samples were obtained from the participants using bio-safety protocols. Samples were analyzed by trained laboratory personnel.
The main outcome of interest was new, confirmed cases of STIs and HIV. Urine specimens were tested for gonorrhea and chlamydia at the INSP Laboratory (PCR Cobas-Amplicor; Roche, Basel, Switzerland); and blood specimens served to measure the presence of HIV, hepatitis B, hepatitis C and syphilis antibodies at the Condesa Clinic Laboratory [Abbott HIV-1 and HIV-2, Ag/Ab Combo, anti-HBc, anti-HCV and syphilis TP quimioluminiscence immunoassay (Abbott Laboratories, North Chicago, IL, USA)] running in Architect i2000 (Abbott, North Chicago, IL, USA); HIV-positive samples were confirmed with HIV-1 and HIV-2 CombFirm (Orgenics, Alere, Israel). Anti-HBc+ was tested with Determine HBsAg and syphilis TP+ (Abbott) with tittered VDRL test (titre ≥ 1꞉8 was used as the cut-off for active infection). At the baseline survey, two subgroups were defined for the markers of syphilis and hepatitis B: antibody positivity was regarded as a lifetime marker of past or present infection, whereas treponemic antibody positivity together with VDRL demonstrated active syphilis, and anti-HBc plus HBsAg positivity indicated current hepatitis B virus infection.
Based on findings from prior literature and known associations between specific sexual risk behaviors and incident STIs, we created a conceptual framework of likely predictors, the majority of which were time-varying, and included: age, education, drug use, condom use, and frequency and types of sexual [7, 21]. The demographic variable age was continuous and the other demographic variable, highest educational attainment, was categorical. Four separate variables were included to describe sexual activity: had vaginal, anal, or oral sex with clients last week; had vaginal, anal, or oral sex with non-paying partners in the past week; had insertive anal sex with any of 3 most recent clients; and had receptive anal sex with any of 3 most recent clients. The first two variables describing sexual activity were continuous variables and the latter two were binary variables. Consistent condom use during sex in the past month and drug use with any of three most recent clients were similarly coded as binary variables. See Appendix A for further details on each of these variables.
Lastly, since this study is a secondary analysis of a RCT that evaluated the impact of conditional economic incentives on staying free of new curable STIs, a variable for randomization to the four study arms of the original RCT was included in our model. This controls for the effect of incentives and conditionalities. Detailed descriptions of the main covariates and outcomes variables in this study are provided in Appendix A. Income and wealth were not included in the model because nonresponse was high for these variables.
Statistical analyses
Incidence rates were estimated using the person-time method (i.e., by dividing the total number of new HIV/STI infections observed during the study period by the total number of person years at risk). We calculated 95% confidence limits using a bias-corrected and accelerated bootstrap method with 1000 replicates [22, 23]. We chose this method because it yields appropriate confidence intervals even with relatively small sample sizes. Participants lost to follow-up stopped accruing person years at their last known study visit. To calculate person-time for seroconversions that happened during follow-up intervals, we took the midpoint of the follow-up interval as an estimate of the time at which seroconversion occurred. During the course of the study, 43 participants were lost to follow up at the 6-month visit and an additional 25 participants were lost to follow up at the 12-month visit. A detailed analysis of loss to follow up for this cohort was conducted in a previous study [24]. Participants with prevalent HIV infection at baseline were included in the analyses for incident STIs, but were excluded for analyses estimating HIV incidence. Participants with prevalent STIs at baseline were excluded for the STIs for which they tested positive, but were still included for calculations of incident STIs for which they tested negative at baseline. Since HIV is an incurable STI, someone diagnosed with HIV at six months would test positive again for HIV at 12 months. Thus, once diagnosed with HIV, individuals were excluded for analyses estimating HIV incidence but were included in the analyses for other incident STIs.
We estimated marginal models using generalized estimating equations (GEE) with a logit link and binomial variance to examine unadjusted and multivariable-adjusted time-varying predictors of incident STIs [25]. The GEE model provides marginal estimates, for which the estimate is averaged over all values of the covariates, which could be correlated. All models used an unstructured correlation structure. In the main analysis, we used a composite STIs outcome, and then we excluded HIV prevalent cases in a secondary analysis to examine combined incident STIs/HIV. The results of our present GEE model are conditional on returning to the Clinic for follow-up. We used quasi-likelihood independence model criterion (QIC) to select the best working correlation structure and the best subset of covariates as diagnostic measures of model fit [26]. Data were analyzed using STATA 13.1 (StataCorp LP, College Station, Texas, USA) and SAS 9.4 (SAS Institute Inc., Cary, North Carolina, USA).