Stoner, Bradley P. MD, PhD*; Whittington, W. L.*†; Hughes, James P. PhD*; Aral, Sevgi O. PhD†; Holmes, King K. MD, PhD*
MAINTENANCE AND SPREAD OF sexually transmitted infections (STI) within populations reflect complex interactions involving host susceptibility and pathogen virulence; availability and accessibility of health care; and sociobehavioral parameters of sexual mixing, partnership formation, and healthcare‐seeking responses to infection.1–7 Transmission‐dynamics models of STI have focused on the reproductive rate of infection (Ro) as a function of pathogen transmission efficiency (β), duration of infectiousness (D), and rates of sex‐partner change (c).8–10 Variability in each parameter may serve to promote differential transmission among subsegments of the at‐risk population.
Application of social and sexual network theory to studies of STI transmission patterns focuses on the importance of networks in defining reservoirs of infection, routes of transmission, and populations at risk of exposure.11–19 Sexpartner networks provide avenues by which STI spread through populations at risk, linking core and bridge groups through sexual activity among network members.20,21 Recent analyses have focused on the impact of sexual‐mixing patterns and preferences on STI epidemiology, with emphasis on selective mixing by geography, sexual activity level, and age.6,7,12,22–25 Enhanced understanding of the role of sexual networks in STI transmission can guide development of targeted intervention programs to reduce STI among the most susceptible sectors of a community.
We examined and compared sex partner networks that sustain and propagate gonococcal and chlamydial infection in Seattle, Washington. In‐depth evaluation of sociodemographic and behavioral characteristics of patients with STI (index cases) and their recent sex partners reveals important differences in network‐membership parameters, including patterns of sexual behavior and sex‐partner selection within transmission networks. These behavior patterns not only serve to sustain pathogen transmission within networks, but also provide opportunities for transmission linkages from core (high‐prevalence) to noncore (low‐prevalence) subgroups within the population.21,24
Study Design and Population
The study evaluated networks of heterosexually transmitted gonococcal and chlamydial infection in Seattle, Washington using sociometric data (i.e., data collected directly from index cases and from sex partners).19 Index cases were defined as persons with laboratory‐confirmed gonococcal or chlamydial infection diagnosed between February 1, 1992 and April 1, 1995. Participants were recruited from the following sources:
1. Randomly selected patients seeking evaluation for sexually transmitted infections at the Harborview Medical Center Sexually Transmitted Disease (STD) Clinic‐the major public STD facility in Seattle, with a volume of more than 13,000 patient visits per year. STD patients with gonorrhea or chlamydia who were not enrolled in the randomly selected group but who agreed to complete the initial research questionnaire were also included in the study.
2. Patients seeking evaluation for STI at the Columbia Health Center (CHC), a community adolescent‐medicine clinic serving minority and economically impacted residents in southeast Seattle.
3. Randomly selected laboratory‐based samples of patients with gonorrhea or chlamydia who sought care at sites other than the Harborview STD Clinic or CHC.
Eligible participants were heterosexual, between 14 to 45 years, and able to understand spoken English. Study participants provided written informed consent and completed face‐to‐face interviews to elicit detailed demographic, medical‐history, and sexual‐behavior information. In‐depth material was recorded with regard to partner‐specific sexual behaviors that extended up to 1 year before the time of the interview.
Sex partner‐locating information was obtained from all index cases by a research assistant or disease intervention specialist from the Seattle‐King County Department of Public Health, and extensive efforts were made to find and enroll all first‐generation partners said to have had sexual exposure to the index case within the past 3 months. All first‐generation partners contacted were strongly encouraged to undergo full clinical evaluation for STI. Among first‐generation partners with proven infection, efforts were undertaken to enroll their partners in the study (i.e., second‐generation partners to the index patient). Enrollment was carried out to the third generation if infection was identified in the second‐generation participants. In this manner, chains of infection were followed through contact tracing and data were collected on as many persons as possible, thereby establishing networks of sexual contact and disease transmission. At the time of the study, standard public health practice did not include partner notification for persons who did not test positive for either gonococcal or chlamydial infection. For this reason, contact tracing was not under‐taken for partners of those enrollees who did not have a documented infection on medical evaluation. All enrolled partners underwent detailed face‐to‐face interviews to elicit demographic, medical‐history, and sexual‐behavior information identical to that collected from index cases, thus permitting sociometric analysis of data collected from index patients and partners.19 All procedures were approved in advance by the University of Washington Human Subjects Committee.
A comparative summary of the study sample is provided in Table 1. A total of 141 persons with gonococcal infection (index cases) were approached and offered inclusion in the study. All index cases received full medical evaluation and treatment for STI, and 127 consented to participate in the study. The 127 enrolled index cases named a total of 246 sex partners (first‐generation partners) within the prior 3‐month period. Of these first‐generation partners, a total of 84 persons from the local Seattle‐King County area were located by the study disease intervention specialist, contacted, and offered study participation, 72 of whom agreed to enroll in the study. The 72 first‐generation partners named 110 sex partners (second‐generation partners) other than the index case. Of these second‐generation partners, 9 were contacted and offered study participation, and 6 agreed to participate. No third‐generation partners were traced or contacted. Twenty‐seven of the 72 first‐generation partners and none of the six second‐generation partners tested positive for gonococcal infection.
Similarly, a total of 215 persons with documented chlamydial infection only were medically evaluated, treated, and offered inclusion in the study; 184 of whom were ultimately enrolled. These index cases named a total of 291 first‐generation partners within the prior 3‐month period, of whom 134 were identified, contacted by the study disease intervention specialist, and offered enrollment in the study. Ultimately, 111 first‐generation partners agreed to enroll, and named 99 second‐generation partners other than the index case. Similar proportions of the approached second‐generation and third‐generation partners consented to study participation (19 of 25 persons and 3 of 4 persons, respectively). Chlamydial infection was documented in 41 of 111 first generation partners, 4 of 19 second‐generation partners, and one of three third‐generation partners. Thus, the final study group was composed of 205 persons in the gonococcal‐networks sample (index cases and all enrolled partners) and 317 persons in the chlamydia‐networks sample. Twenty‐six index cases were coinfected with gonorrhea and chlamydia; however, preliminary examination of this group demonstrated the greatest similarity with the gonococcal index case sample in terms of demographic and behavioral characteristics. For purposes of network analysis, these persons were included in the gonococcal‐networks sample.
Clinical and Laboratory Evaluations
All study participants from Harborview Medical Center and CHC received standard clinical evaluation for STI. Patients examined at other sites received standard medical care appropriate to those facilities. Documentation of gonococcal infection was based on a positive endocervical or urethral culture or the presence of typical Gram‐negative intracellular diplococci on a Gram‐stained urethral smear. Documentation of chlamydial infection was based on a positive endocervical or urethral culture.
All analyses were undertaken using sociometric data collected directly from index patients and partners during the face‐to‐face interview.19 Gonococcal and chlamydial networks were characterized by sociodemographic and behavioral variables including age, race or ethnicity, education, sexual activity level, drug use, and other features. Subjects were permitted to self‐identify with up to five racial or ethnic groups, but for the current analyses we used the first race or ethnicity stated. Age, education, and number of sex partners were analyzed as continuous variables using the t test. Other variables, such as drug‐use history and employment status, were analyzed using the chi‐square test. Between‐network analyses compared characteristics of gonococcal and chlamydial networks summed across index patients and enrolled partners. Preliminary analyses using generalized estimating equations suggested that assumptions of independence were reasonable (i.e., the estimated correlation between persons in the same network was small).26 Therefore, all tests and P values reported are based on standard statistical procedures that assume independence.
Mixing matrices for race or ethnicity and for sexual activity level were computed separately for males and females for each network. The matrices cross‐classify partnerships by race or ethnicity of the index case (rows) and first‐generation partners (columns). The Q statistic was computed for each mixing matrix as follows: EQUATION 1 where pii are the diagonal elements of the mixing matrix and n is the dimensionality (number of categories for each variable) of the matrix. Q varies between −1/(n − 1) to 1. Negative values indicate disassortative mixing (discordance), positive values indicate assortative mixing (concordance), and 0 indicates random mixing.6 Confidence intervals for Q are computed using the standard errors for pii. Assortativeness for each grouping was calculated as the percentage of all partnerships formed for which the index case and partner reported the same characteristic. The Fisher exact test was calculated for white versus black partnerships to compare mixing patterns by gender and race or ethnicity sexual activity level for each respondent was characterized as low activity (one partner during past 3 months), medium activity (two partners during past 3 months) or high activity (three or more partners during past 3 months).
Network size (Nt) was defined as the number of persons within two sex‐partner generations of an index case during the 3‐month period before study enrollment. Global network‐size comparisons were developed by calculating the individual network size for each index patient enrolled in the study and summing these sizes across study participants. For each index case enrolled, the network size (Nt) was computed as follows: EQUATION 2 where No represents the index case (i.e., No = 1), N1 represents the total number of first‐generation partners named by the index case, and N2 represents the total number of partners other than the index case of all first‐generation partners within the prior 3‐month period. No and N1 are known for all index cases, but N2 is not known for all index cases because of incomplete enrollment of first‐generation partners. Missing values for unenrolled partners were estimated in two ways. In the first model (observed‐data model), the number of partners of unenrolled first‐generation partners was assumed to be similar to responses of those persons who were enrolled in the study. Therefore, observed data on enrolled partners were used to estimate the total number of partners (N2) that would have been reported by the unenrolled first‐generation partners. In the second model (observed + imputed‐data model), a simple linear regression was calculated to compare index case perceptions of their partners' total number of partners with the actual number of partners stated by first‐generation partners who enrolled in the study (perceived versus stated: slope = 0.4, r2 = 0.072). The regression equation was then used to impute the likely number of partners that would have been reported by unenrolled partners, and network size was calculated using observed data on enrolled partners plus imputed data on unenrolled partners. In this model, N2 was equal to the total number of partners stated by enrolled first‐generation partners, plus the total number imputed for first‐generation partners who were not enrolled.
Sociodemographic and behavioral characteristics of heterosexual gonococcal‐ and chlamydial‐network members are provided in Table 2. The gonococcal‐network sample (N = 205) included slightly more females than males, with a mean age of 22.4 years. The chlamydial‐network sample (N = 317) demonstrated gender and age compositions similar to those of the gonococcal networks. The networks samples differed significantly with regard to race or ethnicity, with the gonococcal networks containing a significantly greater proportion of blacks and a smaller proportion of whites than chlamydial networks (P = 0.001). Number of years of education was also somewhat lower for gonococcal‐network members (11.4 versus 12.0, P = 0.005).
Persons within gonococcal networks were more likely than those in chlamydial networks to be unemployed (independent of student status), to have ever been jailed, to report using crack cocaine, to acknowledge a current drug problem, and to report a prior history of having been sexually abused or assaulted during childhood. There were no significant differences between networks samples in terms of student status, health insurance coverage, receipt of public assistance, and alcohol or tobacco use.
Compared with chlamydial networks members, persons within gonococcal networks reported younger mean age at time of first sexual intercourse (14.0 versus 14.7, P < 0.003) and higher numbers of sex partners during the past year (10.4 versus 8.0, P < 0.001) and in the past 3 months (4.9 versus 3.1, P < 0.001) as measured from the date of study enrollment. A trend was also noted for higher numbers of lifetime sex partners among gonococcal‐network members (50.1 versus 38.7, P = 0.06), as well as the number of partners within the immediate 30‐day period before study enrollment (2.2 versus 1.1, P = 0.10).
Mixing matrices for index cases and first‐generation partners allow examination of patterns of partnership formation by race or ethnicity and by sexual activity level within gonococcal and chlamydial networks. Within gonococcal networks, the highest degree of race or ethnicity assortativeness was seen among black females (0.96), followed by black males (0.58) and white males (0.50) (Table 3). The lowest race or ethnicity assortativeness was seen among white females, for whom only 29% of all partnerships were with white males. Overall, female mixing patterns did not significantly differ from male patterns, as measured by the Q statistic (Qfemale = 0.29 versus Qmale = 0.08, P = 0.46). Fisher exact test analyses restricted to white and black partnerships demonstrated that black females were significantly more assortative than white females (P = 0.04), but no differences in assortativeness were seen between black and white males (P = 0.64).
Among persons in chlamydial networks, high levels of assortativeness were again seen for black females (0.91), followed by white males (0.70) and Asian or Pacific Islander females (0.56). Relative to gonococcal networks, white‐female chlamydial partnerships were more assortative (0.47), whereas black‐male chlamydial partnerships were somewhat less assortative (0.44). Similar to the gonococcal network pairings, female mixing patterns within chlamydial networks did not significantly differ from male patterns (Qfcmale = 0.38 versus Qmale = 0.23, P = 0.48). Fisher exact test analyses restricted to white and black partnerships demonstrated that black females were more assortative than white females (P = 0.02), and that white males were more assortative than black males (P = 0.03).
Sexual activity level matrices for gonococcal‐ and chlamydial‐network samples are presented in Table 4. Among members of gonococcal networks, the greatest degree of assortativeness was seen among low‐activity males (0.71), followed by low‐activity females (0.44) and high‐activity females (0.33). However, systematic assortativeness by sexual activity level did not occur (Qfemale = 0.02, Qmale = 0.00). Among persons in chlamydial networks, the highest levels of assortativeness were seen for low‐activity females (0.64), followed by low‐activity males (0.54) and medium‐activity females (0.41). As in the gonococcal‐networks matrix, Q calculations for the chlamydial matrix did not demonstrate systematic assortativeness by sexual activity level (Qfemale = 0.10, Qmale = 0.05).
We also calculated Q for mixing matrices derived from egocentric data collected from index patients alone (i.e., index patients provided information about themselves and their partners), and compared these data with the sociometric mixing matrices described previously. For the egocentric data, no systematic differences were noted in the direction or magnitude of Q values for mixing by race or ethnicity or sexual activity level (data not shown). This analysis lends further support to the argument of assortativeness by race or ethnicity but not sexual activity level in this cohort.
Within‐network analyses permitted assessment of the extent to which index cases and their partners were similar or different. For these analyses, index cases were compared directly with enrolled first‐generation partners. In both network samples, the majority of index cases recruited for the study were female, so the index groups contained a preponderance of females, whereas the partner groups were predominantly male.
Within gonococcal networks, index cases and partners were similar for race or ethnicity, age, education level, student status, employment status, jail history, crack‐use history, and current drug problem (data not shown). Although history of prior sexual assault was marginally higher among index cases than among partners independent of gender differences in the sample (25.3% versus 12.7%, P = 0.05), there were no significant differences observed in any of the sexual behavior measures, including age at first intercourse or numbers of partners in lifetime, past year, past 3 months, and past 30 days.
Within chlamydial networks, index cases were less likely than partners to have ever been jailed or to have used crack, independent of gender differences in the sample (data not shown). No differences were observed in race or ethnicity, student status, employment status, current drug problem or prior sexual assault. After controlling for gender, age at first intercourse and number of partners in lifetime, past year, past three months, and past 30 days was similar for both groups.
Network size (Nt) calculations permit comparison of the number of individuals within two sex partner generations of an index case during a 3‐month period. Individual network sizes were calculated based on the number of partners reported by each index case during the 3‐month period before study enrollment, plus the number of partners reported by all first‐generation partners. As described above, missing values for unenrolled partners were derived from observed data from enrolled partners (observed‐data model), or were estimated based on regression analysis of index case perceptions of partners number (observed‐ + imputed‐data model). These results are presented graphically in Figure 1. Based on the observed‐data model, gonococcal networks were significantly larger during a 3‐month period than chlamydial networks (median Nt, 4.0 versus 3.0; P < 0.05). Using the observed‐ + imputed‐data model, the differences in network size were increased, with median Nt for gonococcal networks of 4.7, compared with 3.0 for chlamydial networks (P < 0.05). Both models suggest that gonococcal networks contain more persons per unit time (and by extension may exhibit higher rates of partner change) than chlamydial networks.
These findings suggest that gonococcal and chlamydial infections spread differently through networks of sex‐partner interaction, and involve different sectors of the larger population at risk. Persons within gonococcal networks, for example, were more likely to report an earlier age of sexual debut and higher rates of sexual activity in the recent past than those in chlamydial networks. Further, gonorrhea‐network members had a greater number of markers of social marginalization, underachievement, and rule‐breaking behaviors (e.g., unemployment, education level, past and current drug use, jail history) than persons within chlamydial networks. Members of gonococcal networks were also more likely to be victims of sexual abuse (independent of gender), perhaps consistent with a marginalized and peripheralized status in society. Conversely, chlamydial‐network members were more socially integrated, with lower reported rates of drug use, jail, unemployment, and fewer recent sex partners.
Race or ethnicity mixing matrices suggest that gonococcal partnerships were somewhat more discordant than chlamydial partnerships, with the lowest level of assortativeness in our sample for partnerships involving white females. While Fisher exact test analyses demonstrated differences in assortativeness between white and black females in gonococcal networks and between white and black males and females in chlamydial networks, use of the more robust Q statistic for comparing male and female mixing patterns within networks demonstrated no significant differences by gender in either networks sample. Further, there was no systematic assortativeness of mixing by sexual activity level in either networks sample.
These findings may be considered within the context of transmission‐dynamics models for the spread of STI, which predict higher rates of sex partner change (c) for gonorrhea relative to chlamydia based on prior estimates of efficiency of transmission (β) and duration of infectiousness (D). Brunham and Plummer27 estimated β for gonorrhea at 0.5 and D for gonorrhea at 0.15 in the setting of an effective disease‐control program, leading to an expected value for c of 13 partners/year at disease equilibrium (Ro = 1). For chlamydia, β was estimated at 0.2 and D at 1.25, leading to a predicted c of 4 partners/year. Although our study did not measure rate of partner change per se, the number of partners reported was higher for gonococcal‐network members than for chlamydial‐network members, and calculations of Nt were larger for the gonococcal networks‐findings that are seemingly consistent with model expectations.28 However, acquisition of gonococcal infection may be a proxy marker for high‐risk social network activities on a number of fronts, and care must be exercised in interpreting the findings because mixing matrices showed only modest network differences in partner‐selection preferences.
The analysis is further complicated by the fact that disease rates were not at equilibrium during the course of the study, because reported gonorrhea and chlamydia infections were decreasing in the Seattle area. In this sense, gonococcal and chlamydial infections are at different stages of dynamic topology.29 Gonorrhea‐control programs have been in place for a longer period, and gonococcal infection is arguably closer to a state of endemic equilibrium, with infection restricted to a smaller subsegment of the population at large. However, Seattle has experienced intensive community‐wide chlamydial screening through the Region X chlamydial screening program, which has been in place since the mid‐1980s. Screening reduces the number of prevalent chlamydia cases, reduces the duration of infectiousness (D), and may over time limit the transmission of chlamydial infection to networks characterized by higher rates of sex‐partner change. From a transmission‐dynamics standpoint, such control programs may affect Q by eliminating assortative and disassortative extremes within the population, driving Q closer to a random value. Further research will clarify the extent to which disease‐control programs and the dynamic topology of STD epidemics may affect sexual‐mixing patterns among persons at risk for STD.
It is likely that sexual‐mixing assortativeness by social and demographic characteristics will vary in different locales and under different conditions and circumstances. Rothenberg et al30 found heterogeneity in sexual‐mixing and substance‐use patterns in a comparative study of drug users in two US cities. Study participants in Flagstaff, Arizona demonstrated disassortative mixing for race or ethnicity, whereas participants from Atlanta, Georgia exhibited high levels of assortativeness for this characteristic. Additional studies in various geographic and socioeconomic settings will help determine the extent to which mixing patterns are affected by local conditions and practices.
This study has several limitations. A major concern is the generalizability of the study sample. Although index patients were randomly selected from several sources‐including the municipal STD clinic, an urban adolescent health clinic, and community‐based pool of patients diagnosed with STI‐the network compositions identified may inadequately reflect all networks of disease within the community. This is especially important for homosexual disease‐transmission dynamics because the study was limited to heterosexual index patients and their partners. Further, participation in the study may have been biased toward enrollment of accessible and compliant partners, because partners who were difficult to contact or who declined voluntary participation were not included in the analysis. This would be an important consideration in evaluating the mixing matrices in which only a subset of partners named by index cases were ultimately enrolled. The fact that the study was limited to partners of infected persons limited our ability to draw conclusions about the behavior of persons within gonococcal and chlamydia sexual networks but who did not contract disease. The fact that the networks samples in Seattle were rather heterogeneous with regard to race or ethnicity may limit the applicability of these findings to other cities, where disease may be more concentrated in minority race or ethnicity communities. Finally, the study depends on self‐reported behavior by participants and the assumption that participants' responses to questions adequately reflected their actual behavioral patterns.
Our calculations of network size are also based on certain assumptions about missing partners (i.e., those persons who were not enrolled in the study). For example, the observed‐data model implicitly assumes that enrolled partners are a random sample of all named partners, whereas the observed‐ + imputed‐data model assumes that index case perceptions of partner sexual activity apply equally to enrolled and unenrolled partners. Although we recognize these inherent limitations, we anticipate that any existing biases in partner study enrollment or index‐case perceptions of partner behavior will apply equally to gonococcal and chlamydial networks, thereby strengthening our interpretation of relative differences in network size.
In conclusion, STI networks are characterized by significant social and behavioral variability. Despite common modes of transmission, gonorrhea and chlamydia are spread through distinct yet overlapping networks of sexpartner interaction. These findings have important disease‐control implications. Targeted intervention activities directed at the level of the network, rather than at the individual, may allow greater access to persons at highest risk of future infection by virtue of their network membership. For example, community‐based programs that focus on socially marginalized populations may have the best chance to reduce gonococcal transmission. Further research is required to discern key predictors of network membership and to develop effective intervention strategies that are informed by network models.
1. Anderson RM, May RM. Infectious Diseases of Humans: Dynamics and Control. New York: Oxford University Press, 1991.
2. Hethcote HW, Yorke JA. Gonorrhea Transmission Dynamics and Control: Lecture Notes in Biomathematics No. 56. Berlin: Springer-Verlag, 1984.
3. Gupta S, Anderson RM, May RM. Networks of sexual contacts: implications for the pattern of spread in HIV. AIDS 1989; 3:807–817.
4. Anderson RM. The transmission dynamics of sexually transmitted diseases: the behavioral component. In Wasserheit JN, Aral SO, Holmes KK, eds. Research Issues in Human Behavior and Sexually Transmitted Diseases in the AIDS Era. Washington: American Society for Microbiology, 1991:38–60.
5. Garnett GP, Anderson RM. Contact tracing and the estimation of sexual mixing patterns: the epidemiology of gonococcal infections. Sex Transm Dis 1993; 20:181–191.
6. Garnett GP, Hughes JP, Anderson RM, Stoner BP, Aral SO, Whittington WL, Handsfield, HH, Holmes KK. Sexual mixing pattern of patients attending STD clinics. Sex Transm Dis 1996; 23:248–257.
7. Aral SO, Hughes JP, Stoner BP, Whittington WL, Handsfield HH, Anderson RM, Holmes KK. Sexual mixing patterns, including linking and bridge populations, in spread of gonococcal and chlamydial infections. Am J Public Health 1999; 89:825–833.
8. Anderson RM. Mathematical and statistical studies of the epidemiology of HIV. AIDS 1989; 3:333–346.
9. May RM, Anderson RM. Transmission dynamics of HIV infection. Nature 1987; 326:137–142.
10. May RM, Anderson RM. The transmission dynamics of human immunodeficiency virus (HIV). Philos Trans R Soc Lond B Biol Sci 1988; 321:565–607.
11. Anderson RM, Gupta S, Ng W. The significance of sexual partner contact networks for the transmission dynamics of HIV. J Acquir Immun Defic Syndr Hum Retrovirol 1990; 3:417–429.
12. Ghani AC, Swinton J, Garnett GP. The role of sexual partnership networks in the epidemiology of gonorrhea. Sex Transm Dis 1997; 24:45–54.
13. Haraldsdottir S, Gupta S, Anderson RM. Preliminary studies of sexual networks in a male homosexual community in Iceland. J Acquir Immun Defic Syndr Hum Retrovirol 1992; 5:374–381
14. Jacquez JA, Simon CP, Koopman J, Sattenspiel L, Perry T. Modelling and analyzing HIV transmission: the effect of contact patterns. Math Biosci 1988; 92:119–199.
15. Klovdahl AS. Social networks and the spread of infectious diseases: the AIDS example. Soc Sci Med 1985; 21:1203–1216.
16. Klovdahl AS, Potterat J, Woodhouse D, Muth J, Muth S, Darrow WW. HIV infection in an urban social network: a progress report. B Method Sociol 1992; 36:24–33.
17. Klovdahl AS, Potterat JJ, Woodhouse D, Muth J, Muth S, Darrow WW. Social networks and infectious disease: the Colorado Springs study. Soc Sci Med 1994; 38:79–88.
18. Morris M. A log-linear modeling framework for selective mixing. Math Biosci 1991; 170:349–377.
19. Morris M. Epidemiology and social networks: modeling structured diffusion. Sociol Methods Res 1993; 22:99–126
20. Morris M, Podhisita C, Wawer MD, Handcock MS. Bridge populations in the spread of HIV/AIDS in Thailand. AIDS 1996; 10:1265–1271.
21. Thomas JC, Tucker MJ. The development and use of the concept of a sexually transmitted disease core. J Infect Dis 1996; 174(suppl. 2):S134-S143.
22. Wallace R, Traveling waves of HIV infection on a low-dimensional “socio-geographic” network. Soc Sci Med 1991; 32:847–852.
23. Woodhouse DE, Rothenberg RB, Potterat JJ, et al. Mapping a social network of heterosexuals at high risk for HIV infection. AIDS 1994; 8:1331–1336.
24. Aral SO. Patterns of sex partner recruitment and types of mixing as determinants of STD transmission: limits to the spread of sexually transmitted infections. Venereology 1995; 8:240–242.
25. Rothenberg RB The geography of gonorrhea-empirical demonstration of core group transmission. Am J Epidemiol 1983; 117:688–694.
26. Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986; 73:13–22.
27. Brunham RC, Plummer RA. A general model of sexually transmitted disease epidemiology and its implications for control. Med Clin N Am 1990; 74:1339–1352.
28. Nagelkerke NJD, Brunham RC, Moses S, Plummer FA. Estimating the effective rate of sex partner change from individuals with sexually transmitted diseases. Sex Transm Dis 1994; 21:226–230.
29. Wasserheit JN, Aral SO. The dynamic topology of sexually transmitted disease epidemics: implications for prevention strategies. J Infect Dis 1996; 174(suppl 2):S201-S213.
30. Rothenberg R, Trotter R, Baldwin J, Sterk C, Maxwell C, Pach A. Heterogeneity of risk: variability in social and risk networks of drug users (abstract no. 23177). In, Abstracts of the 12th World AIDS Conference. Geneva, 1988.