ROTHENBERG, RICHARD B. MD, MPH*; STERK, CLAIRE PHD†; TOOMEY, KATHLEEN E. MD, MPH‡; POTTERAT, JOHN J. BA§; JOHNSON, DAVID BA‡; SCHRADER, MARK BA, MPA‡; HATCH, STEFANI MA*
INTERVIEWING PEOPLE WITH SYPHILIS, identifying their sex partners, and bringing those partners to evaluation and treatment have been cornerstones for syphilis control in the United States.1,2 According to current guidelines, the major target of the interview process is the infected person, but some attention is also given to people not found to be infected, but who may have been exposed or who have some other connection to an infected person (a "cluster" interview).3 During the standard interview, a "critical period" guides the elicitation of sex partners: The length of time for which partners are sought is determined by the clinical presentation of the respondent and is usually 3 months for those with primary syphilis, 6 months for those with secondary syphilis, and 1 year for those with early latent syphilis. Process evaluation includes a "source and spread" analysis of the direction of transmission, using the dates of first and last sexual encounter, date of the appearance of symptoms, and date of treatment as guides. Important evaluative measures of the adequacy of this process include examination of the number of contacts identified and brought to medical evaluation, and the number of new, previously undiagnosed people with syphilis brought to treatment.
In recent years, the potential for targeting sexually transmitted disease (STD) interventions by identifying groups of people who may be important in transmission ("core" groups)4–6 has led to a consideration of the role of social and sexual networks in disease transmission, and the potential use of social network and ethnographic concepts in disease investigation.7 We have used these concepts to provide concurrent and retrospective analysis of a syphilis outbreak in a small community near Atlanta, Georgia.
A small cluster of people with syphilis was first detected in the spring of 1996 by a nurse who staffs a part‐time STD clinic in a suburban county that is part of the Atlanta metropolitan area. The nurse reported these cases to the regional public health office staff that provides disease investigation services for the county and to the Georgia Division of Public Health. Initial assessment suggested that the outbreak involved a substantial number of people, with groups of people who interacted sexually on a regular basis. Additional help for disease investigation was mobilized from a larger nearby county STD control program. At the onset, an attempt was made to take a network‐oriented approach to the outbreak by interviewing as many people as possible who might be involved in transmission (whether infected or not), and by attempting to record information on standard epidemiologic forms (Centers for Disease Control and Prevention [CDC] 73.54) for use in network analysis. This report summarizes information on 99 people and 10 cases of syphilis, and presents an ethnographic assessment based on follow‐up interviews with key participants.
The standard epidemiologic interviewing form (CDC 73.54) contains all the information needed for network analysis. Relevant data (e.g., diagnoses, dates, contact information) were abstracted from the form to create a database with one observation for each respondent‐contact dyad. The 99 people in this group provided 203 observations. Conversion of routinely collected interviewing information to the format required for network analysis is a straightforward multistep procedure. Data were entered into one of the standard database managers (Microsoft Excel; Microsoft Corporation, Redmond, WA), then converted into a standard database analytic program format (Statistical Analysis Systems [SAS])8 using DBMSCopy.9 We wrote a short SAS program to create an ASCII file that could be used by network analysis software (UCINET10) to produce network measurements and statistics. UCINET also produced an ASCII file that is used by a network visualization program (KRACKPLOT11) to render network graphs. Further details of this process, including the relevant programming, will be provided by the authors on request.
We used network‐based measures to examine the temporal evolution of the outbreak, to visualize the interrelationship of subsets of people within the network (e.g., infected people, their sexual partners, and other ethnographically definable groups), to calculate the probability of direct sexual contact with an infected person as a function of selected characteristics, and to review in detail network connections, dates of sexual exposure, and clinical presentation of people with syphilis. We calculated the following statistics for the overall network and for comparison of subgroups12: components (the number of groups within which there was a connection of some length among all people); degree (the number of direct sexual connections each person has with others, or "partners per case" in STD parlance); betweenness (a measure of the frequency with which a person falls on the shortest line that connects two others); information centrality (a measure of the average of all possible pathways between individuals); and k‐core (the number of people with k contacts who are connected to each other). We examined the change in "microstructures" associated with the growth in sexual connections among people involved in the outbreak. These microstructures, defined in the notes for Table 3, quantify the intensity of interaction among participants. We used follow‐up ethnographic interviews to compare reported sexual and social behavior of selected participants 6 to 12 months after active involvement in syphilis transmission.
In a network of this size, a single display of all participants is not visually informing, but does confirm the impression of substantial interaction provided by network statistics (graph not shown). Subgraphs were used to specify temporal, ethnographic, and epidemiologic associations that may be important in syphilis transmission.
Overall Ethnographic Assessment
Syphilis was diagnosed in six white female subjects (four of whom were younger than 16 years of age), two white male subjects (both 17 years of age), and two African‐American male subjects (ages 19 and 16 years). Based on routine contact interviewing and investigation of infected and uninfected people, and special ethnographic interviews with key participants, a complex picture of sexual interaction, starting at least 1 year before the diagnosis of the first syphilis cases, emerged. At the center of this outbreak was a group of young white girls (two thirds of whom were 16 years of age or younger) who, in various combinations, met periodically to use drugs and have a variety of sexual interactions with several groups of slightly older boys (Figure 1). The venue was usually the home of one of the girls whose parents were out for the evening. The two major groups of boys differed in their ethnic and economic background. One group was a more affluent set of whites 17 to 21 years of age; the other was a predominantly African‐American group of similar age but of less affluent background. The two groups did not comingle at the parties. The drugs of choice were blunts (short, mild cigars), to which marijuana or cocaine had been added, and alcohol. Multiple accounts corroborated the fact that injectable drugs were not used. Sex was usually public and communal: the girls would have sequential or simultaneous sex partners, experiencing vaginal, anal, and oral sex, occasionally at the same time, and occasionally with more than one partner at a particular orifice. The girls also had sex with each other, and numerous sexual encounters outside the party environment were also documented. During the initial outbreak investigation, we were unable to document the extent of parents' knowledge or understanding (or possible participation) in these activities.
The overall group contained only two connected components, one of 95 people and one of 4 people. In the overall network, the average degree for each person was three (range, 1–16; Table 1). Those with syphilis had an average of 7.4 sex partners, compared with 2.4 for those without syphilis. This difference was an artifact of interviewing, however, because those without syphilis who were interviewed also had an average of 7.4 sex partners. Similarly, the average overall information centrality was 0.53, but was 0.72 for those with syphilis and 0.73 for those without syphilis who were interviewed. The pattern for betweenness deviated somewhat, however, in that those without syphilis who were interviewed had an average betweenness of 1.73, compared with 4.06 for those with syphilis, suggesting that those with syphilis were in fact more central within this network than those without, and that their greater centrality was not an artifact of interviewing. The largest k‐core had a k of 8, and consisted of 19 people (that is, there were 19 people who each had 8 contacts and were connected to others with 8 contacts). There were two k‐cores each with a k of 7, and there were 13 people in each of these groups.
Probability of Contact With an Infected Person
Based on graph connections, an infected person had a probability of 0.33 of being connected sexually to another infected person. The probability that an uninfected person was connected to an infected person was also 0.33. Based on interviews, the proportion of infected contacts named by people with syphilis was 0.26 and the proportion of infected contacts named by people without syphilis was 0.27 (Table 2). This proportion varied somewhat with the underlying characteristics of the interviewed person, but in only one subgroup was the difference large (0.48 of contacts to white, uninfected people were infected; 0.30 of contacts to white infected people were infected). This difference, and other variations among subgroups, was related to the differential sexual mixing patterns among subgroups.
The intensity of interaction between the African‐American men and the white girls was greater than that between the white men and white girls, although this intensity did not appear to lead to greater transmission to the African‐American men (Figure 1). Visualization of these groups and all their sex partners uncovered the importance of several people not specifically identified with these groups (e.g., N43 and S30) who served as bridges between the two groups of men.
The temporal evolution of network formation, using the date of first sexual contact with anyone in the network as the date of entry, demonstrates a substantial buildup of sexual activity during the second half of 1995, with further intensification of such activity into the first half of 1996 (Figure 2). The sexual genesis of this outbreak had occurred a year or more before the period of intense interviewing and investigation in the spring of 1996, at which time all the cases of syphilis were actually diagnosed. Several people diagnosed with syphilis had been active during the early period (S21, S9, S1, S11, S32) and of these, three had early latent syphilis and two had secondary syphilis. Their moment of infection and subsequent infectivity cannot be known with certainty, but would suggest that important sexual activity antedated the cutoff periods used as criteria for interviewing.
The increasing sexual interaction demonstrated by visualization of the changing network (Figure 2) is substantiated by the occurrence of microstructures‐small groupings of people that can heighten transmission.12 Such structures include cliques, n‐cliques, k‐plexes, and k‐cores (see Figure 2 for definitions). There was a dramatic increase in the number of these microstructures over the period of observation (Table 3), indicating a rapidly increasing potential for transmission. All cases of syphilis were diagnosed during the last period depicted, although, as noted, some probably contracted the disease earlier.
When only people with syphilis and their sex partners are visualized (Figure 3), it is apparent that at least one infected person (S30) had no known contact with another infected person. Perhaps even more important, a substantial number of people (N41, N26, N18, N10, N42, N13) had contact with two or three infected people but were not found to be infected at the time of evaluation, suggesting that syphilis may have been prevented by epidemiologic treatment of these people. The diagram also pinpoints a subcluster of cases (S1, S2) for whom few contacts were elicited and no further cases emanated. These cases suggest a deficiency in the case‐finding process, because, unlike other cases, they do not appear to be part of this complex web of sexual interactions.
Simultaneous examination of the sexual connections among people with syphilis only, including their clinical presentation and their dates of first and last contact with each other (Figure 4), highlights the considerable concurrency that occurred. For example, S11, who had secondary syphilis at the time of diagnosis, had conconcurrent sexual relationship with five other infected people, and for at least four of these, the period of contact overlapped. Similar situations apply to S9 and S17. In view of the presumed duration of secondary and early latent syphilis, the source of infection would be impossible to define.
Follow‐up interviews, focusing on the current situation and network changes, were conducted between 6 and 12 months after the initial interviews with a subsample of eight of the adolescent women. The interviews were held outside the clinic setting, and subjects voluntarily participated in the 30‐ to 60‐minute session with one of us (C.S.).
Based on their personal history and their comments about others', these young women revealed that many had continued to be sexually active with multiple partners in the context of drug and alcohol use. There were, however, some important changes. A few young women no longer participated because they had moved or because of stricter parental supervision. In the past, gatherings had tended to be at the home of one of the young women, but had moved to local motel and hotel rooms. The originally identified social network had fragmented into several smaller groups. For example, one of the African‐American men who was central in the original network was apparently no longer willing to engage in group sex and tended to consort with a single (changeable) partner at a time. Several women also claimed to demur from group sex, although group use of alcohol and drugs had continued for them. Two women stated that they would "get high" with the group and have sex with one of the male partners, either in their car or in a hidden public setting. Other women stated that they were in a steady sexual relationship with a man from the original group involved in the outbreak. One of the main motivators for sexual risk reduction was the unconfirmed rumor that one of the men in the central group had been identified as HIV positive. The women interviewed all appeared to agree that most parents had not taken action in response to the outbreak, nor were there increased levels of communication at home regarding drug use and sexual activity.
Thus, the ethnographic data indicate that some social network changes had occurred that might inhibit continued disease transmission (e.g., fewer sexual exposures in groups), but that the basic behavior patterns had not changed greatly. The clinic staff indicated that they maintained more contact with the men than with the women, many of whom began attending the family planning clinic. Two women in the original cluster of 18 were pregnant at the time of their treatment for syphilis. Clinic record and verbal reports indicated that an additional 13 women, 8 from the original cluster and 5 others involved in the outbreak, became pregnant subsequent to completion of their treatment.
In this outbreak, a small number of syphilis cases illuminated a network of intense, concurrent sexual activity with some unusual ethnographic features. In fact, the overall characteristics of this outbreak deviate markedly from the stereotypic patterns now thought to dominate in the United States,13 wherein minority populations and the exchange of sex for drugs play dominant roles. The use of network and ethnographic tools highlights these characteristics and challenges some of the precepts that form the basis for syphilis control strategies.
Network statistical measures document that this group was highly interactive, with a considerable degree of centrality among its members. The existence, for example, of a single connected component that included all but 4 of the participants, and a subgroup of 19 connected people, each of whom had 8 direct contacts, provides direct evidence of this interactivity. The dramatic increase in microstructures during the later phase of the outbreak underscores the type of network structure that facilitates disease transmission. Our approach demonstrated that uninfected people are often as central (i.e., as potentially important in transmission, based on their egocentric networks) as infected people. Because intervention occurred only during a narrow window, compared with the temporal extent of actual sexual activity, the potential importance of these uninfected people for illuminating transmission dynamics is undervalued by traditional methods. Furthermore, the network approach also demonstrated that the boundaries of this network had probably not been reached, as evidenced by the existence of cases that appeared at the margin (S1, S2) or unconnected to other cases (S30).
The network approach demonstrated that the results of interviewing infected and uninfected people were virtually identical for purposes of discovering infected contacts. In this outbreak situation, the probability of being sexually connected to an infected person was the same regardless of whether one was infected. In a closer examination of the clinical presentation and temporal data of people with syphilis, the concept of "source and spread" cases has little meaning. The likelihood of having contracted syphilis from a particular person might appear to be based more on the probabilities of transmission from multiple infected partners than on single transfer of the treponeme.
The traditional presentation of epidemic evolution‐the epidemic curve‐would fail to capture the growth in size and complexity of the sexual substrate from which these cases of syphilis came. The sexual context that generated these cases of syphilis became more intense and more interactive during the latter half of 1995, and intensified even further, both with new participants and new connections, during the first half of 1996. Although there may have been some impact on the transmission of syphilis through active case investigation and epidemiologic treatment, later ethnographic information suggests that disease intervention had little impact on the underlying network behavior that generated these cases.
Implications for Evaluation
These observations suggest that some of the traditional approaches to syphilis control may not be well founded when dealing with outbreaks associated with highly interactive groups. Interviewing only infected people distorts the extent and nature of sexual interaction, misses the role that some people may play in forming a bridge between subgroups, and fails to identify some infected cases. Using a fixed critical period, based only on the presumed clinical epidemiology of the stage of syphilis, ignores important sexual interactions outside that period and does not give investigators access to individuals who may still be important in transmission, despite their remote contact. Finally, in a situation of considerable concurrence, the goal of establishing source and spread cases may be misplaced because multiple infections to multiple infected partners preclude clear delineation of a chain of infection.
On the other hand, a network analysis suggests several important approaches to evaluation. Formulating network diagrams in real time permits disease investigators to identify critical people (e.g., for follow‐up, reinterview) and critical subgroups (for selective mass treatment). The network approach permits rapid assessment of the type of group, the specific ethnographic characteristics, and the potential for continuing transmission. With ongoing network evaluation, the boundaries of the network and the connections to wholly different geocultural groups can be determined.
As an epidemic investigation proceeds, the changes in the network over time permit a rational estimate of the size of the outbreak and the underlying sexual activity, of the structural characteristics of the group, and of placement of risk behavior (what people do) in the context of risk configuration (how they interact with others). Visualization of the network permits a firmer basis for identifying prevented cases‐for example, the six uninfected people who had contact with two or more infected people‐because the likelihood of infection would have increased with multiple concurrent exposures. The use of a network‐based "epidemic curve" permits better correlation of case‐finding activity with secular trends. The changing configurations of networks are likely to be harbingers of transmission, and a movement away from dense, sexually interacting networks provides a basis for tying program intervention activity to declining transmission. Similarly, routine assessment of networks can be used as a surveillance tool for providing warning of adverse epidemiologic change.
Although, as noted, this outbreak has some unusual ethnographic features, it is not clear whether it represents an unusual network configuration for the transmission of syphilis. Much of the experience of STD control programs in dealing with sporadic and epidemic syphilis over the past several decades is undocumented, and a distribution of the network configurations associated with syphilis is not available. It has been postulated that the more typical syphilis outbreak in the current era‐that associated with the exchange of sex for drugs or money‐may result from differences in the density of sexual networks.14 Inner city networks of people involved in the exchange of sex for drugs are likely to be as dense and interactive as the network described here (Rothenberg RB, Sterk C, Pach A, unpublished observations from an ongoing study of urban social networks), and the outbreak described here may well be typical of current syphilis occurrence. If so, the small number of syphilis cases, given the amount of sexual activity and the extended time period, suggests that a high degree of interactivity may be necessary to sustain transmission, a finding that has been suggested on more theoretical grounds.15
With appropriate training16 and utilization of a database such as CDC's STD Management Information System (Mattock L, Rothenberg RB, personal communication), the process described here can be adopted as part of the preventive activities of many STD control programs. We would caution those who work in STD control, however, against interpreting these findings as simply an admonition to do more "cluster" interviewing. The latter, as noted, is the process of interviewing people who are not directly connected sexually to those with syphilis but who may be important in transmission. As such, cluster interviewing is very much in the spirit of a network approach. Our data suggest, however, that partner notification, its current variants, and its administrative edifice are a subset of a broader approach that examines the complex context of social networks. Use of network and ethnographic methods can be of direct benefit in epidemic control, as was the case here, and may provide insight into the long‐term dynamics of transmission.