Our study supports the approach to recruit, screen, and enroll high risk couples outlined by McMahon and colleagues (2003)  and extends these protocols to the study of FSWs and their intimate, non-commercial partners in a resource poor setting. Based on this experience, we offer several suggestions with applicability to other studies of high risk couples in diverse social and cultural contexts. Main points of reflection center on collaborative decision making and considering the local context in protocol development; strategies to mitigate participant risk, particularly regarding possible IPV; and suggestions for assessing the methodological rigor of the screening process of couple-based studies.
Our adaption of McMahon et al's protocol  is innovative in several ways. First, while self-reported relationship status is typically used in couple-based studies and may be sufficient in many settings, prior experience working with drug-involved FSWs who live and work along the Mexico-U.S. border suggested that they were street savvy and that verifying partner status would provide additional oversight to the screening process. Indeed, in Tijuana, a member of a "couple" excluded by the CVS angrily complained to the Field Coordinator that the questions differed from what another enrolled FSWs told them to expect. We generated a large enough pool of questions to create multiple versions of the CVS and thwart potential participants posing as a couple. Moreover, our questions were developed with multiple perspectives from team members to assure cultural relevance and sensitivity to the social context of this border population, including issues related to family, informal economies, violence, local drug markets, and law enforcement activity.
Collaborative efforts are vital in protocol construction when working in resource poor and international settings. Drawing on formative work in the region, eliciting input from the entire research team and field staff, and field testing the individual measures and flow of the entire process proved vital to the final protocol development. Eliciting a wide range of input ensured the inclusion of multiple viewpoints and different academic perspectives, thus strengthening the final protocol. Fieldworkers, particularly those with extensive experience in other research projects with similar populations, know the local cultural and socioeconomic contexts and can make valuable contributions to protocol and instrument design, including assessing the appropriateness of questions. Local input is imperative in adapting measures for relevance within the linguistic, socio-cultural, and economic contexts of unique geographic regions and study populations [61–63]. The protocols and instruments for our study benefitted greatly from this collaborative process.
Field testing the screening instruments and overall enrollment process was also critical to the successful implementation of our study. Based on experience field testing the CVS with known couples, we modified the original protocol and decided to use the same interviewer to administer the questions to each partner sequentially to try to get a better sense of whether or not the couple was legitimate. We suggest that in certain contexts a single interviewer, regardless of gender matching with the interviewee, who can probe for consistency of responses may improve the more quantitative approach developed by McMahon et al. (2003) . If a single interviewer is well trained in a structured approach and gains some experience with the population, project purpose, and screening questions, then the validity and reliability of the screening instruments may be improved.
In addition, the CVS screening questions required both closed and open-ended answers. The majority of the questions were closed-ended with a list of options based on formative fieldwork. Eliciting open-ended answers, however, meant that interviewers administering the CVS could probe each partner for specific details to try to better assess the veracity of the information provided. Open-ended questions can also generate emergent data for other analyses, such as assessing the locations where participants purchase drugs in a constantly changing local drug market. Open and close-ended questions are equally easy to assess for concordance in the field. Close-ended questions render ad hoc analyses much easier, but including a few open-ended questions can personalize the interview and add another layer of reflection to the screening process that later can be quantified for analyses.
Within a nexus of sex work, drug use, and HIV/STI risk in places like the Mexico-U.S. border region, everyday violence can become normalized and internalized [64, 65]. As such, research protocols must be sensitive to screen for cases of potential violence. Among FSWs, it is important to develop specific screening questions that clearly define cases of IPV which should be excluded because participation in a study might place them at further risk. Based on the Revised Conflict Tactics Scales [58–60] and input from the field teams, our screening tools clearly assessed the timeframe, frequency, type, and severity of violence experienced by participants within their intimate relationships, and generated automatic disqualification from the study based on key responses. Staff were also trained to screen out other cases of IPV on a case-by-case basis if need be, but this situation did not occur at either site.
We also followed McMahon et al's (2003)  suggestion to recruit through women first in order to provide her with greater decision making power, screen out cases of extreme IPV, and reduce female participants' risk. This approach, however, may bias the sample in favor of "less risky" couples. A recent study found that among FSWs enrolled in an HIV intervention who reported having steady partners, those who reported IPV were significantly more likely to report that their partners engaged in known HIV risk factors such as injection drug use and having had sex with another partner while in their current relationship than were FSWs who did not report IPV . It is worth clarifying, however, that we did not screen out couples reporting any IPV, but only those reporting extreme violence that could be life threatening. Overall, 4.2% of women who took the primary screener and 0.8% of couples who were administered the CVS were excluded due to extreme IPV. Thus, while our sample may underestimate risk in this population, the relatively small numbers of participants excluded based on this criteria suggest that the impact on our findings will be minimal.
Regardless, researchers have a responsibility to carefully assess the prevalence of IPV in the population to make an informed decision that protects the safety of the participants. Staff and participant safety protocols for emergent cases of violence and a list of local referrals for psychological counseling and other forms of assistance should also be developed. While our protocols were originally designed with the female participant's safety in mind, it is worth noting that during the CVS stage, two couples (one at each site) were screened out due to male partner concerns about IPV, but no additional couples were screened out due to female concerns about IPV. These results suggest the utility of individually screening the female partner for IPV prior to male partner involvement in the process, and serve as a reminder that males are not immune from experiencing partner violence within the context of an intimate relationship.
As advocated by Witte and colleagues (2004) , recruitment protocols should be carefully defined and codified in a manual of procedures. Although we created such a manual and conducted extensive training to standardize procedures across sites, we also recommend that the research team view recruitment and screening as a process of multiple components, which allows for a certain level of flexibility in the field. It was important to strictly adhere to the study inclusion and exclusion criteria, which were checked at each phase of screening. It was equally important, however, to enroll couples based on the totality of evidence, including the concordance of couples' answers about each other in the CVS and the local field staff's knowledge and observations of the couples whenever possible. Instead of opting for hard cutoffs in matching partners' answers with each other on the CVS, this process took into account proximity of answers. Overall, nine couples (3.8%) were disqualified based on the CVS. In seven instances, couples were not automatically disqualified by the computer programming of the CVS, but rather by field staff who determined that their answers were too discordant and their interpersonal interactions too awkward or distant to indicate that they were a real couple. The astute judgments of well trained and culturally attuned staff who have insight into the local social context are an invaluable part of the recruitment process.
Finally, while this protocol appeared to have excluded those who did not qualify for the study, more sophisticated analyses are needed to determine the effectiveness of these screening tools . Researchers could test the effectiveness of individual questions in the CVS by determining which ones show higher concordance of responses by partners. Another option would be to calculate similarity coefficients to assess how "close" couples measure in their responses . Coefficients could be used to test for differences between couples who were included versus excluded from the study, and assess differences between couples whose score signaled high similarity versus those whose scores were not as close. Such analyses would lend methodological rigor to the recruitment processes of couple-based studies and help other researchers adapt their own protocols to include measures grounded in empirical evidence. Comparisons of couples' CVS scores to their responses to other quantitative and qualitative instruments used in the study could also help contribute evidence regarding the validity of screening instruments and procedures. Couples should also be observed prospectively to assess correlations between initial CVS scores, relationship stability, and dissolution.