The spread of sexually transmitted infections (STI) in populations is not random; highly connected individuals with many short-duration partnerships (“core group”) are important for maintaining transmission.1–3 Sexual contact networks are an effective way to describe the structure of contact between sex partners in a population and predict the spread of STI.4,5 When an individual having a sexual relationship with a member of a core group has a sexual relationship with an individual in another grouping of the sexual network, the individual bridges the 2 groups and can transmit infections between otherwise isolated parts of the larger network.
Theoretical work using mathematical models has identified bridging of sexual network groupings as a predictor of the emergence and maintenance of STI transmission within a population.6 To test this theory empirically, investigators have fit statistical models to survey data sets, defining bridgers as people who have sex with commercial sex workers and noncommercial sex workers, people who have sex with injection-drug users, noninjection drug users, and bisexuals.7–18 However, people who connect any 2 demographic groupings, regardless of the STI prevalence, risk of STI acquisition, or transmission within each group, also can be defined as bridgers.19–22 Indeed, any sociodemographic criteria that characterize individuals and influence sex partner selection may be used to define hypothetical sexual network structures such as age, ethnic, socio-economic groups, and geographic distance. Characterizing these types of bridges provides important information for predicting STI transmission dynamics.
Whether an individual transmits an infection between demographic groupings requires that: (1) the individual has a sex partner in more than one group; (2) 1 partner is infected and the other is susceptible; and (3) a contact sufficient to transmit infection (“effective contact”) occurs during the infectious period. Thus, timing of the bridging relationship is important. To characterize bridging, researchers have used different temporal window lengths between the 2 partnerships including actual temporal overlap, requiring 2, 3, 6 months, 1 year, or requiring no specific time of overlap.8–22 Overlapping or concurrent partnerships, that is, partnerships that overlap in time or occur closely enough in time to facilitate STI transmissions are important factors contributing to greater population and individual risk of STI transmission,23,24 regardless of whether the relationship is a bridge.
We defined a bridger as an individual whose most recent and second most recent sex partners differed by demographic grouping (e.g., age group, socio-economic status group, racial group, spatial separation of residences, and gender [Table 1]). The implicit assumption is that partner choice is partially informed by the demographic and spatial features we consider. Work of Laumannn et al. support the assumption that demographic features may be important in influencing partner-choice.25
For example, gender bridging assumes 3 fairly independent groupings only connected via bridgers: a first group that consists of only men who choose partners that are other men, a second group that consists of men who choose women and women who choose men as sexual partners, and a third group that consists of women who choose other women as sex partners (Fig. 1). If an individual chooses a partner first from 1 group and then from another group, the individual has bridged the demographic group. It is irrelevant to us which group the bridger is from. Similarly, if we consider age bridging, we assume that women of a specific age group generally choose male partners (or female partners of this age group, but for simplicity let us just say men), and that the partner choice for men is symmetric. If 1 woman of any age has sex with men of 2 different age groups, this woman has bridged these demographic groupings. Again, we assume that her male sex partners generally choose partners from their home demographic group, and again, it is irrelevant to us for the case of defining a bridger which demographic group the woman is from.
To better understand the extent to which individuals bridge demographic groupings in their sex partnerships, and whether bridgers differ from nonbridges demographically or in factors related to STI transmission, we describe the frequency and associations of different bridging types among men and women aged 18 to 39 years participating in the Seattle Sex Survey, a random digit dialing (RDD) survey conducted in 2003 to 2004, and the associations of bridging with known STI risk factors and STI history. Through a better understanding of bridging, we may better understand how infection flows through some populations and thus how prevalence increases, remains endemic, or decreases in some populations, while remaining absent in others.
We analyzed the results of the 2003 to 2004 random digit dialing survey conducted in the Seattle area among residents age 18 to 39 years of age with fluency in the English language.26 The RDD sample included listed and unlisted numbers and was obtained from Survey Sampling, Inc. of Westport, CT. Up to 6 attempts were made to contact each number at different times of day. The survey was conducted by the Social and Economic Sciences Research Center in Pullman, WA.
We contacted 31,617 (85.4%) of the 37,000 telephone numbers in the initial sampling frame. Of these, 28,946 (91.6%) did not meet eligibility requirements for reasons such as not a Seattle residential number, no one over 18, deceased, disconnected number, fax/data line, temporarily out of service, always busy, no answer, telephone answering machine, and telecommunication barriers. Of the remaining 2671 numbers, 1477 (55.3%) refused to participate, leaving 1194 (44.7%) individuals who agreed to participate and completed an interview. Using the 2008 guidelines of the American Association for Public Opinion Research, the adjusted response rate, assuming 42.4% of unknown eligible cases were eligible, was 24.1%; the unadjusted response rate was 14.8%. Of the 1194 participants, 1013 (84.8%) reported ever engaging in vaginal, oral, or anal intercourse with at least 2 partners, and were included in this analysis.
The survey instrument included questions on sexual history, partner and partnership characteristics of the respondent’s 5 most recent partnerships, STI history, and demographics. Specifically, subjects were asked if they had ever had a physician diagnosis of any of the following STIs: HPV, chlamydia, hepatitis B, hepatitis C, genital herpes, gonorrhea, syphilis, pelvic inflammatory disease, nongonococcal urethritis, and HIV/AIDS. The survey was pretested on a sample of the study population and revised before initiating data collection. The telephone survey required approximately 20 minutes to conduct and was administered using computer-assisted telephone interviewing software, which standardized the interview and minimized data entry errors. Respondents were given the option to select the gender of the available interviewers.
A respondent was considered to have a monogamous relationship with their most recent sex partner if they answered “0” to the question “once you began sexual activity with (this partner), with how many other people did you engage in sexual activity?” If they responded with 1 or more partners, the participant was considered to have had a nonmonogamous relationship with their most recent sex partner.
We examined bridges that occur for age, education, race, spatiality, and gender, as defined in Table 1 and illustrated in Figure 1. We considered a bridge to have occurred independent of the duration of any temporal gap between sex partnerships. In a previous study of the same dataset we observed that most gaps between consecutive partnerships were sufficiently short to allow transmission of most STI (i.e., were briefer than the average period of infectivity of most STI). If network-specific data were missing for either of a participant’s 2 most recent partnerships, we conservatively assumed no bridging occurred. Age, education, race, and spatial bridging all had between 9% and 20% missing data. None of the gender bridging data were missing. For age bridging, we used differences in year of birth.
We restricted our analyses to the 1013 (83%) respondents who had ever engaged in oral, vaginal, or anal sex, and had described their 2 most recent sex partnerships. Descriptive analyses were performed to assess the prevalence of each bridging type, and if any associations existed at the univariate level between a bridging type and socio-demographic and behavioral variables, as well as known STI risk factors We examined the distributions of (i) subject age, (ii) shortest courting period, (iii) age of first sex and (iv) lifetime number of partners for each bridging type compared to respective nonbridges; we used Mann-Whitney-Wilcoxon 2 sample tests to test for statistically significant differences in these distributions. The population distributions for each bridging type and these variables were constructed to give visual representations. We constructed statistical models using bridging status to predict past STI diagnosis, and other STI risk factors. We conducted all above analyses for both genders separately and together to ascertain modification by gender of participants.
In reality an individual is not just a bridger of 1 type or another, but of all types that define a sexual network simultaneously. In light of this, we examined bridge-duos, to test for positive or negative correlations with 1 bridge type and all others individually. We also analyzed the distribution of full bridge classifications—that is, considering age, education, race, spatial, and gender bridging all at once. We refer to these as multibridge classifications. To test if each multibridge type occurred more or less frequently than expected by chance alone, we used the marginal distribution of each individual bridge type (Table 1) to calculate expected frequencies. These were used with the observed frequencies of each full classification to calculate the χ2 statistic.
We conducted a series of sensitivity analyses. First, to examine how sensitive our inferences were to how we handled missing data, we compared our results in which we assumed individuals with missing bridging data were nonbridgers, to results where we assume those with missing data (i) were bridgers and (ii) were excluded all together. Second, to examine how sensitive our inferences were to the temporal window used to define the partnerships being bridged by a bridger, we compared our results in which we did not use a temporal window, to results where the bridging relationships had to occur within the past year. All analyses were performed using SAS version 9.27
Characteristics of the overall study population were described previously.26 In the subset included here, the 1013 participants who reported ever engaging in vaginal, oral, or anal intercourse with at least 2 partners, the median age was 31 (27, 35 for 25th, 75th percentiles, respectively) and 56% were women. Most participants (83%) reported exclusive lifetime heterosexual history, but 4% reported exclusive homosexual history and 14% reported sexual activity with both sexes. About one-fifth (20%) of participants reported a past STI diagnosis. The median age of sexual initiation was 17 years of age (15, 19 for 25th, 75th percentiles, respectively); the median lifetime number of partners was 9 (5, 16 for 25th, 75th percentiles, respectively). The median shortest courting period for the entire study population was 1 day (0, 21 days for 25th, 75th percentiles, respectively). Twenty percent reported they were not monogamous during their most recent partnership.
Prevalence of Bridging Behaviors
Of the 1013 participants who described their 2 most recent sexual partnerships, 763 (75%), were classified as a bridger of some type. Education bridges were the most common (46%), followed by spatial (34%), age (29%), race (24%), and gender bridges (3%) (Table 1). When we limited the analysis to subjects with more than one partner during the previous 2 years, the percentages and rank ordering were almost the same (education: 49%; age: 40%; spatial: 36%; race: 31%; gender: 1%). When further limited to subjects with more than one partner during the previous year, the ranks stay consistent, but age bridging is more common (education: 47%; age: 46%; spatial: 37%; race: 35%; gender: 1%). Unless otherwise specified, we present results with the first set of data to maximize our sample size and power.
We present the prevalence of categorical demographic variables and STI risk factors by bridging type in Table 2. Age bridging was negatively associated with exclusive heterosexual history (P = 0.02), positively associated with lifetime history of STI diagnosis (P <0.01), and positively associated with ever having exchanged sex for money (P <0.01). Race bridging was positively associated with male sex (P <0.01), positively associated with both having less than a bachelor’s degree and having more education than a bachelor’s degree, but negatively associated with only having a bachelor’s degree (P <0.01). Race bridgers were also less likely to be currently religious (P <0.01). Spatial bridging was positively associated with ever exchanging sex for money (P = 0.04). Gender bridging was positively associated with female sex (P = 0.03). Most other categorical features appeared to be distributed similarly among bridgers and nonbridgers.
Table 3 presents the distributions of continuous variables and each bridging type. We test for differences in the bridger versus the nonbridger distributions using the Mann-Whitney-Wilcoxon test. Age bridging was associated with older subject age, shorter courting periods, and more sexual partners over the lifetime. Race bridging was associated with younger age. Figure 2 shows the distribution of age bridges versus nonbridges stratified by both sex and self-reported monogamy with most recent partner demonstrating an association of age bridging with shorter courting periods, earlier age of sexual initiation, and more lifetime number of partners; and that these associations were modified by male gender and self-reported nonmonogamy.
Predicting STI History and Related Risk Factors by Bridging Behaviors: Multivariate Analysis.
To explore the effects of bridging on known STI risk factors, we conducted a series of multivariate analyses. Linear models were fit to lifetime number of partners, history of STI diagnosis, shortest courting period, and age of sexual initiation using simple linear regression, logistic regression, and Cox proportional hazards models. Because of a strong age modifying effect, we stratified the sample into young (<25) and old (>24) age groups (Table 4). As the distribution of lifetime number of partners was highly skewed, we log transformed this variable before fitting the model. In each model we explicitly adjusted for the participant’s gender, whether they had exclusive same-sex partners, and whether they were monogamous with their most recent sex partner, in addition to bridge type. Besides these adjustments and stratification by age group, we did not adjust for any other of the participant’s demographics. For each model, a positive parameter value indicates a riskier outcome, that is, a higher lifetime number of partners, a positive STI diagnosis, shorter lifetime courting period, and earlier age of sexual initiation. We ran all regression models presented both in their full form, as presented, as well as their simple forms using only 1 predictor at a time. In each case, those factors significant in the full model were also significant in the simple model, with associations in the same direction.
In those aged 18 to 24, no bridging type was significantly associated with STI risk factors after adjustment (Table 4). By contrast, in those aged 25 to 39, age bridging was significantly associated lifetime number of sexual partners, shortest courting period, and history of STI diagnosis (Table 4).
Individuals engaging in 1 set of bridging partnerships often engaged in another (data not shown) Age-education, age-race, race-education, and education-spatial were all positively associated with one another (P <0.01). Gender bridging was not significantly associated with any other type. No duo was negatively associated, even at a nonsignificant level.
Because gender bridging was not associated individually with any other bridging type, and because of its small sample size, we ignored it when creating multibridge classifications. There were more exclusive nonbridgers than expected in our sample, perhaps because we assumed no bridge when data were missing (P <0.01). There were also fewer age-education-race–only bridgers than expected (P <0.01). However, there were more age-only (P = 0.02), education-only (P <0.01), and race-only (P = 0.02) bridgers than expected.
When we varied how we handled missing data related to bridging status, our results were largely consistent. Whether assuming those with missing information were bridgers, or simply excluding those with missing bridging information, if there were discrepancies they were usually of the variety in which a previously nonsignificant result became significant while the trend remained the same.
Sensitivity analyses limiting the data to bridges occurring in the past year were somewhat uninformative as the population size dropped from 1013 to 199. The proportions of bridgers for each dimension were 46%, 47%, 35%, 37%, and 1% for age, education, race, spatial, and gender bridges, respectively. Because there were only 2 gender bridgers, we did no further sensitivity analysis with them. However, the trends were similar to those observed in Table 4.
Among the 1013, 18- to 39-year-old sexually active Seattle residents reporting 2 or more sex partners who participated in a random digit dialing survey in 2003 to 2004, most (74%) bridged a demographic grouping during their 2 most recent sexual partnerships. Only age bridging was associated with nonmonogamy. Age bridging also was associated with increased lifetime history of STI, but only among those aged 25 to 39 after adjustment for other bridging, monogamy, gender, and same-sex partnerships. These data provide some insight into the connectivity of the larger sexual network. Differences in connectivity patterns between mixing classes affect how infection prevalence increases, remains endemic, or decreases in some mixing classes, while remaining absent in others.
The prevalence of respondents reporting age and spatial bridging we observed in Seattle differed from earlier reports in other populations. Among 4490 adolescents with 2 or more partners participating in the National Longitudinal Study of Adolescent Health,20 there was more age bridging than in our study (69% vs. 29%). The difference in age bridging may reflect our more strict definition of an age bridger: we required a 5-year age difference, rather than only 2 years in the participant’s 2 most recent partners. The discrepancy might also reflect the different underlying age distributions of each study; ours included 18- to 38-year-olds compared with 13- to 19-year-olds. The proportion reporting age bridging in our study (29%) resembled the 21% to 26% reported by young men aged 14 to 24 from Denver and Baltimore, with chlamydial infection, in which bridging was defined as a 2-year difference in age of a subject’s partners22. A study in Sweden used contact tracing to perform a clustered sampling approach, and classified 8% of the 851 individuals sampled as spatial bridges,21 much lower than our 34%. Similarly, in a study conducted in King County, WA among participants in a randomized trial of enhanced partner notification and expedited partner treatment for gonococcal or chlamydial infection,19 spatial bridging was much less common (5%) than our study (34%). Both the Sweden and King County study used contact tracing to classify spatial bridging, which may tend to overrepresent partners easily reached because of residential proximity to index cases. In a North Carolina study consisting of recent HIV cases, the percent of gender bridges was 15%, much more common than the 3% observed in our random sample.15 We are unaware of other studies that have examined race or education bridgers.
Our prevalence estimates reflect the probability of taking sexual partners from 2 demographic groupings given some underlying partner acquisition rate of taking any new partner. This is without respect to time, and thus should not be interpreted as a rate, without assuming some underlying partner acquisition rate. The estimates are reported only for those eligible to bridge (the 1013 reporting 2 or more partners, conservatively assuming that those not reporting complete information on a partner characteristic were not bridgers) and do not represent the prevalence of various bridging types in the overall population. Further, we assume that the characteristics we considered (age, education, race, spatiality, and gender) influence partner choice and thus influence sexual network structure.
The reader should take into account the limitations of our data and analysis in interpreting our results. First, for a person to be classified as a bridger, the participant had to report information on the bridging characteristic for their last 2 partners. Not surprisingly, for some variables the amount of missing data were large (20% for education bridgers). In these cases, we conservatively assumed no bridge occurred, probably underestimating the prevalence of bridging. If people who were not reporting data were more likely to engage in riskier (or less risky) sexual behaviors, we might also underestimate the association of bridging behaviors with STI risk factors. As examined in our sensitivity analysis, most associations presented in our analysis remained significant and all remained in the same direction when we changed how we handled those with missing bridging data.
Second, bridging occurs over time; in the bulk of our analysis, we present bridging of subjects’ 2 most recent partnerships without taking into account the timing of the bridging. It is plausible that a person we label as a bridger, bridged 2 demographic strata in partnerships that were many years apart. While it would still be possible to spread viral STIs, this window would be insufficient to transmit bacterial infections. As described in our sensitivity analysis, the proportions of each bridging type and their relative rank order remained largely consistent independent of which temporal window was used—either no required temporal window, or only partnerships in the past year, and the underlying trends were similar (although the small sample size greatly increased the variability). However, as demonstrated previously using these data, on average, the gap length between partnerships covers the infectious periods of most STIs of interest.21 A final limitation is that participants were identified using random digit dialing scheme. In addition to self-selection that may occur based on survey content, it is possible that some participants were never contacted as they used caller ID or other technologies to screen calls. At the time of conducting the survey, 2003 to 2004, these technologies were less prevalent than they are currently. Further, estimates from the current survey are very similar to those of a previous random digit dialing survey conducted in the same population of 1995, which included many of the same questionnaire batteries.26
Our analysis highlights that sexual networks are diverse and complex, with multiple potential paths for infection flow. While it is relatively easy to demonstrate that bridgers are important for disease spread through populations, bridgers may not necessarily experience an increased STI risk. Rather, bridging may be thought of as a population level factor, and thus population level units are necessary to consider its impact on STI risk and transmission.
1.Anderson RM, May RM. Epidemiological parameters of HIV transmission. Nature 1988; 333:514–519.
2.Anderson RM, Medley GF, May RM, et al. A preliminary study of the transmission dynamics of the human immunodeficiency virus (HIV), the causative agent of AIDS. IMA J Math Appl Med Biol 1986; 3:229–263.
3.Hethcote HW, Yorke JA. Gonorrhea Transmission Dynamics and Control
. Lecture Notes in Biomathematics. Vol. 56. Berlin, Germany: Springer-Verlag; 1984.
4.Gupta S, Anderson RM, May RM. Networks of sexual contacts: implications for the pattern of spread of HIV. AIDS 1989; 3:807–817.
5.Koopman J, Simon C, Jacquez J, et al. Sexual partner selectiveness effects on homosexual HIV transmission dynamics. J Acquir Immun Defic Syndr 1988; 1:486–504.
6.Doherty IA, Shiboski S, Ellen JM, et al. Sexual bridging socially and over time: a simulation model exploring the relative effects of mixing and concurrency on viral sexually transmitted infection transmission. Sex Transm Dis 2006; 33:368–373.
7.Kahn JG, Gurvey J, Pollack LM, et al. How many HIV infections cross the bisexual bridge? An estimate from the United States. AIDS 1997; 11:1031–1037.
8.Havanon N, Bennett A, Knodel J. Sexual networking in provincial Thailand. Stud Fam Plann 1993; 24:1–17.
9.Gorbach PM, Sopheab H, Phalla T, et al. Sexual bridging by Cambodian men: potential importance for general population spread of STD and HIV epidemics. Sex Transm Dis 2000; 27:320–326.
10.Morris M, Podhisita C, Wawer MJ, et al. Bridge populations in the spread of HIV/AIDS in Thailand. AIDS 1996; 10:1265–1271.
11.Choi KH, Gibson DR, Han L, et al. High levels of unprotected sex with men and women among men who have sex with men: a potential bridge of HIV transmission in Beijing, China. AIDS Educ Prev 2004; 16:19–30.
12.Hor LB, Detels R, Heng S, et al. The role of sex worker clients in transmission of HIV in Cambodia. Int J STD AIDS 2005; 16:170–174.
13.Liu H, Grusky O, Li X, et al. Drug users: a potentially important bridge population in the transmission of sexually transmitted diseases, including AIDS, in China. Sex Transm Dis 2006; 33:111–117.
14.Aral SO, Hughes JP, Stoner B, et al. Sexual mixing patterns in the spread of gonococcal and chlamydial infections. Am J Public Health 1999; 89:825–833.
15.Hightow LB, Leone PA, Macdonald PD, et al. Men who have sex with men and women: a unique risk group for HIV transmission on North Carolina College campuses. Sex Transm Dis 2006; 33:585–593.
16.Morse EV, Simon PM, Osofsky HJ, et al. The male street prostitute: a vector for transmission of HIV infection into the heterosexual world. Soc Sci Med 1991; 32:535–539.
17.Izazola-Licea JA, Gortmaker SL, de Gruttola V, et al. Sexual behavior patterns and HIV risks in bisexual men compared to exclusively heterosexual and homosexual men. Salud Publica Mex 2003; 45(suppl 5):S662–S671.
18.Prabhu R, Owen CL, Folger K, et al. The bisexual bridge revisited: sexual risk behavior among men who have sex with men and women, San Francisco, 1998–2003. AIDS 2004; 18:1604–1606.
19.Kerani RP, Golden MR, Whittington WL, et al. Spatial bridges for the importation of gonorrhea and chlamydial infection. Sex Transm Dis 2003; 30:742–749.
20.Ford K, Sohn W, Lepkowski J. American adolescents: sexual mixing patterns, bridge partners, and concurrency. Sex Transm Dis 2002; 29:13–19.
21.Nordvik MK, Liljeros F, Osterlund A, et al. Spatial Bridges and the Spread of Chlamydia: The Case of a County in Sweden. Sex Transm Dis 2007; 34:47–53.
22.Jennings JM, Luo RF, Lloyd LV, et al. Age-bridging among young, urban, heterosexual males with asymptomatic Chlamydia trachomatis. Sex Transm Infect 2007; 83:136–141.
23.Morris M, Kretzschmar M. Concurrent partnerships and the spread of HIV. AIDS 1997; 11:641–648.
24.Watts CH, May RM. The influence of concurrent partnerships on the dynamics of HIV/AIDS. Math Biosci 1992; 108:89–104.
25.Laumann, EO, Michael RT, Gagnon JH, et al. The Social Organization of Sexuality
. Chicago, IL: University of Chicago Press; 1994.
26.Aral SO, Patel DA, Holmes KK, et al. Temporal trends in sexual behaviors and sexually transmitted disease history among 18 to 39 year old Seattle, Washington, residents: results of random digit-dial surveys. Sex Transm Dis 2005; 32:710–717.
© Copyright 2009 American Sexually Transmitted Diseases Association
27.SAS for Windows
. Cary, NC: SAS Institute Inc.