In the United States, the proportion of HIV diagnoses attributed to heterosexual contact rose from approximately 12% of all cases reported to CDC in 1995 to 31%, or 12,860 cases out of 42,011 reported, in 2009; heterosexual contact now accounts for more newly diagnosed cases of HIV and AIDS than injection drug use [1].

CDC conducts surveillance to monitor the occurrence of HIV in the U.S. population. However, given that HIV infection may be asymptomatic for years, behaviors reported at the time of diagnosis may not reflect behaviors at the time of acquisition. Therefore, it is important to identify populations that are at increased risk for HIV infection and collect information on their behaviors to implement and evaluate effective public health actions [2].

In 2003, CDC implemented the National HIV Behavioral Surveillance System (NHBS), which conducts surveillance of behavioral risks and HIV prevalence in metropolitan statistical areas (MSAs) with high AIDS prevalence among 3 populations at high risk for acquiring HIV: men who have sex with men (MSM), injecting drug users (IDU), and heterosexuals. To develop NHBS, definitions of the target populations as well as sampling methods were required. Persons who engage in male-male sexual behavior or injection drug use substantially increase their HIV acquisition risk. For instance, in a recent survey of the general U.S. population, approximately 10% of adult men who report lifetime sexual contact with another man tested positive for HIV [3]. In contrast, prevalence of HIV among heterosexuals who do not inject drugs or engage in male-male sex in the US is very low (0.1-0.2%) [4].

For heterosexuals, defining the target population is less clear than that of MSM or IDU. Because HIV prevalence is so low in the general population, sex with an opposite sex partner is not a good marker for HIV acquisition risk. The National HIV Surveillance System attributes cases to heterosexual contact if a person reports sexual contact with a person known to have or be at high risk for HIV (e.g., a partner who injects drugs); however, information about partners’ risk behaviors is not always known, making this an inadequate definition for purposes of a behavioral survey. In other countries, behavioral surveillance focuses on specific groups of heterosexuals such as commercial sex workers and their clients [5] or those migrating from high prevalence areas [6]. Another challenge is determining scientifically sound, feasible, and replicable strategies for sampling heterosexuals at increased risk for surveillance purposes. Behavioral surveys among MSM have shown venue-based sampling (VBS) to be an effective way to reach men at risk; similarly, respondent-driven sampling (RDS) has been used successfully to reach IDU. These methods were used to recruit MSM in NHBS (NHBS-MSM) and IDU in NHBS (NHBS-IDU), respectively [7,8].

This paper describes the design, implementation, and evaluation of a pilot study to conduct community-based behavioral surveillance among heterosexuals at increased risk for HIV (NHBS-HET) in the United States, and how the findings from the pilot study were applied to subsequent NHBS-HET surveillance activities.

Of the 18,430 participants in the final NHBS-HET pilot data set who completed an interview, 14,750 (80%) met our analysis criteria. Participants were excluded from the analysis if they ever reported injection drug use (N=2,241), had male-male sex (N=414), both injected drugs and had male-male sex (N=311), or if they refused to provide this information (N=5). Participants were also excluded if they did not consent to an HIV test (N=377), did not have a negative or confirmed positive HIV test result (N=212) or reported being HIV-positive but, when tested, were HIV-negative (N=5). Additionally, we excluded persons whose survey responses were invalid (N=120). Of the participants in the sample, 207 (1.4%) had newly diagnosed HIV infection. Characteristics of participants and those with newly diagnosed HIV infection are presented in Table 1. Of the 14,750 NHBS-Pilot Study participants who met the analysis criteria, 57% were female, nearly half were less than 30 years old, and most (73%) were black (Table 1). Socioeconomic status was low. Few participants had a past-year risk behavior (traditionally) associated with HIV infection such as crack use (11%), exchange sex for money or drugs (12%) or STD diagnosis (14%).

Table 2 shows the performance of the predictor variables, including the number and proportion of participants, HIV prevalence, sensitivity and specificity. For each of the five individual risk behavior variables, HIV prevalence was between 2.1% and 3.1%. However, because relatively small proportions of participants engaged in these behaviors (8-21%), and sensitivity was low for each (range: 13-32%), we removed these individual variables from consideration. Evaluation of combined predictor variables, in which persons reporting any or more than one of these five factors would meet the criteria, resulted in improved sensitivity and specificity but did not meet our threshold for further consideration.

Four predictor variables assessed sexual network risk. For each of these variables or combinations of variables, the HIV prevalence exceeded the overall sample prevalence of 1.4%. Two of these variables met sensitivity and specificity criteria for further evaluation: 1) reporting not knowing the HIV status of one’s most recent partner or that one’s most recent partner was an IDU, had been incarcerated, used crack or, for women, was MSM; and 2) sexual concurrency.

We also present data on several social/structural factors including living in an HRA: low income (participant’s household income did not exceed Department of Health and Human Services [HHS] poverty guidelines); and limited education (high school graduate or less). Each of these performed well with respect to HIV prevalence, sensitivity, and specificity. A combined variable including anyone with low income or limited education (which we termed low socioeconomic status [SES]) performed even better, with a sensitivity of 96% and specificity of 12%.

Practicality was assessed for variables meeting criteria for sensitivity >70% and specificity <40%. Use of sexual network measures as eligibility criteria would limit our ability to monitor partner HIV status and other characteristics over time. The use of these variables is also limited by the concern that a large proportion of people did not know the risk behaviors or HIV status of their partners. Use of questions based on characteristics of the most recent partner also might underestimate risk for persons with more than one partner. As a result of these limitations, sexual network risks were eliminated from consideration as potential ways to identify the target population eligibility criteria.

All social/structural factors presented met the threshold for further evaluation. Of these, living in an HRA and low SES performed best and were evaluated for practicality. Use of residency in an HRA was less optimal for several reasons. For instance, some participants had difficulty identifying their census tract of residence on a map; because NHBS is an anonymous survey, staff do not collect address or other detailed information that would make identifying census tract of residency easier. Furthermore, when performance of HRA residency as a predictor variable was assessed by region, it was determined that it performed poorly in the Midwest and the West because a smaller proportion of participants lived in HRAs (data not shown), which would have been problematic for a national surveillance system. The HRA criterion also performed worse in project sites that used RDS than those that used VBS; participants with newly diagnosed HIV infection recruited through the RDS method were less likely to live in an HRA than those recruited through the VBS method.

In contrast, the use of low SES as a method for defining the target population posed fewer concerns with respect to practicality. Measurement of self-reported income and education is less complicated than residency in an HRA. Moreover, it had excellent sensitivity in cities within all U.S. regions and territories and among all heterosexual sub-populations (data not shown).

The results of these analyses were presented at an NHBS principal investigators’ meeting in October 2009 and representatives from each NHBS project site were asked to weigh the evidence and make a recommendation about 1) the final definition of the target population and its operationalization through specific eligibility criteria, and 2) the sampling method that should be used. RDS was the recommended sampling method for NHBS-HET, endorsed by project areas that had used VBS as well as those that used RDS.

The final NHBS-HET definition incorporated the low SES criteria described above. Participants for NHBS-HET would be eligible if they met the general NHBS eligibility criteria (live in the MSA, not previously participated in current cycle, able to complete the interview in English or Spanish), are male or female, have had vaginal or anal sex with an opposite-sex partner in the previous 12 months, and are aged 18-60 years. The upper age limit was increased to 60 years based on the high prevalence of new diagnosis among those aged 40-50 years in the pilot study. The SES criteria were incorporated into the RDS method by offering only NHBS-HET participants who are low SES as measured by income and education (described above) and have not injected drugs in the previous 12 months an opportunity to recruit others into the study. By using low SES as recruitment criteria, but not as eligibility criteria, we sought to limit the potential inconvenience to participants traveling to a field site only to be found ineligible for the study, and the potential negative impact the study might suffer as a result. However, since only persons meeting the recruitment criteria may recruit others, the resulting sample would be comprised primarily of the low SES target population. The HRA concept is applied only for initial participants, who must be residents of an HRA; in addition, field sites must be located within HRAs. These HRA criteria make sure that the study starts in, and recruitment is maintained within, lower SES communities. This combination of eligibility and recruitment criteria, in conjunction with the RDS method, were selected to ensure that future cycles of NHBS-HET attain a sample of heterosexuals at increased risk for HIV infection.

The NHBS-HET pilot study resulted in a technique to identify heterosexuals at risk for HIV infection and a method to sample them. Our findings were based on nearly 15,000 interviews conducted among persons in MSAs with high AIDS prevalence across the United States. Following an investigation of the performance of predictor variables for heterosexuals with newly diagnosed HIV infection, we determined that the definition of a heterosexual at increased risk would focus on persons with low SES. Eligibility for future NHBS-HET cycles would be limited to a man or woman living in an NHBS MSA who has had vaginal or anal sex with an opposite-sex partner in the previous 12 months; the opportunity to recruit other participants would be offered only to participants with low socioeconomic status (measured by high poverty or low education) and no injecting-drug use in the previous 12 months. Together, these eligibility and recruitment criteria comprise our definition of heterosexuals at increased risk of HIV.

The pilot study analysis was an important step in the effort to better conceptualize heterosexual risk of HIV. Earlier research on the topic defined heterosexuals at risk solely on the basis of high risk sexual behavior [11], such as sex with multiple partners [12]. Results from the NHBS pilot study shift this focus from risky behaviors to at-risk communities by emphasizing the role of social factors – specifically, low SES - in which individuals exist and behaviors occur. Other contemporaneous research highlights the importance of patterns of sexual partnership formation, as well as the social conditions that are associated directly or indirectly with the likelihood of having an infected partner [13].

Our focus on social factors also mirrors other research conducted in the last decade, which has revealed that poverty [13, 14] and concomitant social disparities such as racial inequality [15, 16] are underlying determinants of HIV infection risk [17] even in people who do not have high-risk behaviors [18]. Differences in sexual behaviors have also been shown to not account for increased risk for STDs among blacks [19, 20], and that STD disparities between blacks and whites are greater among individuals in low-risk groups than those in high risk groups [21]. In addition, a previous analysis of NHBS-HET pilot study data demonstrated that race was not significant in a multivariate model which included SES, suggesting that low SES, not black or Hispanic race/ethnicity, is more appropriate to define a high-risk heterosexual population in the U.S. [22]. Nonetheless, the link between low SES and HIV infection among heterosexuals remains unclear and must be investigated further. A recent study of U.S. women – the Women’s HIV SeroIncidence Study (ISIS) – utilized some aspects of the NHBS-HET pilot study design to enroll and retain women at risk for HIV infection in a longitudinal HIV incidence study. However, unlike NHBS, ISIS eligibility was based on behavioral risk factors for women such as sex with a partner who is HIV positive, recent injection drug use, and residency in an HRA [23]. Results from ISIS may help clarify the causal link between poverty, risk behaviors, and HIV incidence among heterosexual women in the US.

There are several limitations to our analysis. First, the NHBS-HET pilot study was a convenience sample in which data from two different methods were combined. Data are not weighted to account for biases in venue attendance (VBS) or differences in network size and composition (RDS); therefore, the results are not generalizable to all heterosexuals at risk in the NHBS MSAs or in the United States. Second, NHBS-HET recruitment targeted residents of areas with high rates of poverty and HIV diagnoses, which may have led to an overestimation of HIV prevalence and also may have limited the applicability of our analysis to communities outside of these highly selected areas. Third, positive HIV status may have been underreported during this interviewer-administered survey, thereby inflating estimates of participants with unrecognized infection. The NHBS questionnaire is interviewer-administered which may increase reporting bias especially for sensitive topics such as sex and drug use.

The second cycle of NHBS-HET was conducted from June through December 2010 using the new definition of heterosexual at risk with the eligibility and recruitment criteria derived from the pilot study described in this paper. CDC is currently analyzing results from the 2010 data collection cycle. Future cycles of NHBS-HET, with its focus on people with low SES and the communities they live in, will help tailor and implement prevention programs for those who are most at risk in the effort to stem the tide of heterosexually acquired HIV infection in the United States.