Social network methodology for studying HIV epidemiology in men having sex with men

1. Introduction 

The rising incidence of human immunodeficiency virus (HIV) infection in men having sex with men (MSM) is a cause for concern in Americas, Europe as well as Asia [1], [2], [3], [4]. Safe sex fatigue, treatment optimism, use of recreational drugs, for example, methamphetamine, and the practice of serosorting have been suggested to be the prevailing attitudes that have led to changes of behavioural practice [5], [6], [7]. The use of internet is changing the communication patterns of the human society. Notably seeking sex partners through the internet has been shown to be associated with unprotected intercourse and a higher number of partners [8]. Apparently, the rise of HIV transmission in MSM does not appear to be a universal phenomenon, as the transmission risk may vary with the specific setting of exposure. The exposure risk to HIV can be situational, when, for example, there is a higher chance of infection in the presence of drug use, or when a sex partner is acquainted through cyberspace [7], [9]. HIV transmission can also be socio-culturally determined, as evidenced by a higher number of black MSM with HIV [10], [11], or through like-minded MSM who select partners according their own preference [6].

Against the background of the changing landscape of sex partnership, conventional means of surveillance that considers MSM as a homogeneous community may be insufficient for describing HIV epidemiology effectively. More innovative approaches are needed which could incorporate the dynamics of networking of MSM in the community [12], [13]. In this pilot study, we set out to track HIV epidemiology by examining the affiliation of MSM with locations where they seek sex partners, on the principle of the duality of place (where MSM seeks partner) and person (the MSM). We considered that that people affiliated with the same place were closer in relationship with one another [14]. Using a social network analysis (SNA) framework, exploration was conducted on data collected from a behavioural survey administered on MSM with recently diagnosed HIV infection.

2. Methods 

In collaboration with the Integrated Treatment Centre (ITC), a self-administered questionnaire survey was conducted. With an active caseload of over 1500, ITC is the largest HIV specialist clinic in Hong Kong with a predominantly Chinese patient population. The clinic provides standard antiretroviral therapy in accordance with clinical indications, and is affiliated with a voluntary counseling and testing service (VCT) [15]. HIV patients were invited to join the study if one (1) was 18 years old or above, (2) had contracted HIV via homosexual transmission, hereafter defined as MSM and (3) was able to read and speak Chinese. Research staff assessed the eligibility of each patient attending the clinic over the study period, explained the protocol to potential participants, and obtained written consent. Approval was obtained from the Ethics Committee of the Department of Health of the Hong Kong Special Administrative Region Government, and the Survey and Behavioural Research Ethics Committee of the Chinese University of Hong Kong.

The questionnaire was constructed covering (1) respondent's demographics including age, education level, employment status, and residence; (2) location for seeking sex partners and for sex in the 3-year period prior to one's speculated year of infection; (3) practice of sexual activities, including masturbation, oral sex and anal sex, and the frequency of condom use for each form of sexual activity. For those admitting to have used a specific location for seeking partner, a six-point Likert Scale (ranging from “1—rarely” to “6—always”) was administered to assess the frequency of such visits as well as the frequency of their sexual behaviour practice [16]. After field testing, six categories of partnership locations were specifically defined for respondents’ selection, which were: public toilet, bars, saunas, gymnasium, beach, parties and the use of internet. The questionnaire was pilot tested and deemed applicable for the present study.

On the assumption that MSM using a specific location for partner-seeking was more closely connected to one another, a matrix was constructed by tabulating Likert Scale scores against respective location category. The networking characteristics of the respondents were assessed by (a) creating two-mode affiliation networks, which served to show the linkages between people and locations, and (b) converting 2-mode networks into 1-mode networks, to assess the linkages between individuals [17]. UCINET version 6.174 and NetDraw 2.083 (Analytic Technologies, Lexington, USA, www.analytictech.com) were used for performing the network analyses and constructing the diagrams.

3. Results 

The study was conducted over an 8-week period between November 2006 and February 2007. A total of 114 MSM were recruited. About 30% of the respondents reported that their infection took place before 1997. The date of laboratory diagnosis was not used in the analysis as there was often a variable period between infection and HIV testing. To minimize the potential recall bias from our respondents, we only included participants whose reported infection took place in or after 1997 for the present analysis. Excluded respondents were generally older in age (t=−3.16, p<0.01), while education level, employment status and residency were not significantly different between the 2 groups. A total of 91 respondents were subsequently evaluated, who had a median age of 37.5 years (range: 22–71 years). Most of the MSM had received secondary or higher education (92.9%), and a majority were living with family (40/91, 44%) and were currently employed (73/91, 80.2%).

The 91 respondents were classified into 3 groups according to their speculated year of infection (1997–2000, 2001–2003, and 2004–2006). Through tabulating the affiliation of MSM with individual location category, 2-mode matrices were created, which showed the linkages between individuals (one mode) and locations (another mode) [17]. Fig. 1 shows the 2-mode networks thus constructed, which reveal a shift from an affiliation with bar and sauna in 1997–2000 to internet in 2004–2006. One-mode networks (Fig. 2) were constructed after conversion from the affiliation matrices. This was done firstly by dichotomizing Likert's score, followed by determination of cross products and re-dichotomizing the results [16], [17]. At the end the binary variables were used to construct one-mode networks that show the linkages between individuals. For the first and the second time periods (1997–2000, and 2001–2003), the networks were less dense, in which 9 and 7 individuals were not connected to any of the networked members. Between 2004 and 2006, only 3 individuals were not connected to the rest of the network members. Network density was measured, which reflected the ratio of actual to possible linkages. There was increased density over time, from 0.26, through 0.36, to 0.53 for 1997–2000, 2001–2003, and 2004–2006 respectively. The degree centrality, which measured the importance of each person as regards his linkage to others, has likewise changed from a median of 9 in 1997–2000, 12 in 2001–2003, to 16 in 2004–2006. Condom use was assessed to determine the level of risk during sexual intercourse. Its use for anal and oral sex in the 3-year period prior to HIV infection was compared. None of the comparisons showed differences which reached statistical significance (Table 1), though there's a rising tendency for not using condom for oral sex. The frequency of behavioural practice of masturbation, oral sex, and anal sex had remained similar over time (results not shown).

View Large Image Download to PowerPoint

Fig. 1. Two-mode networks of HIV-positive MSM (n=91) over the 3 time periods of (a) 1997–2000, (b) 2001–2003 and (c) 2004–2006. Venues for sex partnership are shown in squares while the individuals are in circles. The networks were constructed from the two-mode matrices that showed the strength of linkages between individuals and location using a six-point Likert's Scale. The number for each person denotes the year of infection and the reference code.

View Large Image Download to PowerPoint

Fig. 2. One-mode networks of HIV-positive MSM (n=91) over the 3 time periods of (a) 1997–2000, (b) 2001–2003 and (c) 2004–2006. The networks were constructed after conversion of 2-mode affiliation matrices to one-mode matrices. This was done firstly by dichotomizing Likert's score, followed by determination of cross products and re-dichotomizing the results. Each square represents one MSM in the study population. The number for each person denotes the year of infection and the reference code.

Table 1. Respondents’ reported use of condom for sex in the 3-year period before HIV infection.

		Anal sex		Oral sex
		Seldom use	Always use	Seldom use	Always use
	1998–2000	7 (30.4%)	16 (69.6%)	18 (78.3%)	5 (21.7%)
	2001–2003	9 (29.3%)	22 (71.0%)	27 (87.1%)	4 (12.9%)
	2004–2006	10 (34.5%)	19 (65.5%)	28 (93.3%)	2 (6.7%)

Seldom use—Likert scale 1–3; always use—Likert scale 4–6.

4. Discussion 

Our study presents a new approach to describing HIV epidemiology, using social network data collected from a partnership-based questionnaire self-completed by HIV-positive patients following simple instructions. The study was conceptualized in an attempt to answer the urgent question of why HIV has spread in the MSM community in Hong Kong, a metropolitan city with a predominantly Chinese population in Southern China. The HIV spread in MSM in other Chinese cities, including Beijing, Chongqing, Jiangsu, has likewise increased in recent years [18], [19], [20]. Results in our study is therefore useful in offering an explanation to the situation in the same ethnic community. Clearly, our analysis suggests that the networking pattern of a cohort of HIV-positive MSM prior to their infection has changed. The higher network density means that MSM were, in general, more closely linked with one another through sexual activities within their common social environment. Concurrently, some MSM held a more central position as reflected by the higher centrality, which would have facilitated the spread of HIV if the virus has entered the very community. The increased HIV transmission in Hong Kong and nearby Chinese cities could have resulted from an increased connectivity among MSM, rather any change in the practice of unprotected sex, a phenomenon also proven in our study.

Popularity of internet use could be an important predisposing cause for the rising efficiency of HIV transmission [21]. The convenience of internet use has turned it into a virtual location for sex partnership. MSM using internet were closer to one another, as illustrated by the denser network diagram and the changes in networking properties, including centrality and density. Distinctive clustering was however not observed. Using phylogenetic analysis, genetic clusters have in fact been identified in MSM in Hong Kong [22], a phenomenon testifying to the presence of closely knit networks in the community. The reason for the absence of clear-cut clustering in our study could have been methodological, as our networks were indirectly constructed by converting affiliation matrices and Likert Scale score into linkages. The connectivity of MSM shown in the 1-mode network diagrams was inferred from the respondents’ linkages with sex partnership locations, which did not mean that the individuals actually had sex with the others. The processes of data transformation may have reduced the sensitivity of the network methods to detect clustering [14]. Our approach also carried other limitations. The speculated year of infection, for example, is not a perfect means of defining the time of infection. Any errors in reporting the year of infection may have placed MSM in a different position as regards his affiliation with sex partner-seeking venue. Unfortunately the precise time of infection is often not determined through standard clinical investigation. One alternative would be to study the phylogenetic relationship of the virus strains in each individual, or that a combination of social and genetic methods can be used for exploring HIV transmission dynamics. The relationship between social and genetic clusters would need to be examined in specially designed study in which both types of data are available for assessment [16].

With the ever-changing dynamics of HIV transmission, it appears that conventional approaches to HIV surveillance, comprising passive reporting, prevalence/incidence testing and behavioural monitoring, would hardly be adequate to describe the prevailing epidemiology pattern. As a socio-behaviorally driven epidemic, SNA analysis offers a new angle for enhancing the robustness of epidemiologic investigation in public health context. The application of SNA in exploring HIV epidemiology in MSM shows how this can be done in the field. By the same token, HIV transmission dynamics can be investigated in other vulnerable communities. Injection drug users (IDU) constitute another HIV risk population the networking of which can be investigated using a SNA approach. IDUs are brought together by their shared habit of drug injection. Though HIV can easily spread through needle-sharing, a high HIV prevalence is not inevitable and transmission depends significantly on the network property of those having the same habit. An earlier study in Colorado Springs revealed that HIV spread did not explode when the virus had not reached the central core of the network [23].

In conclusion, we piloted an epidemiologic investigation of HIV infection in MSM by incorporating a 2-mode approach in the application of SNA methodology. This is a feasible alternative to full network or egocentric network methods. Full network investigations would require the identification and interviewing of all MSMs and their partners, which is impossible in field studies. There is the inherent difficulty of identifying each and every sex partner over an extended period prior to one's HIV infection. Egocentric SNA involves collection of information from individuals (ego) on their connections to alters, without investigating the complete network [24]. This is less complex but the issues of data sensitivity and confidentiality still remain. Our strategy of integrating ranked behavioural profile and the application of the principle of the duality of person and people can be an adjunct to behavioural surveillance, which would allow timely and meaningful interpretation to be made. Its application in prediction modeling may further add values to the interdisciplinary approach in epidemiology.

Conflict of interest 

Funding: The study was funded as a commissioned project by the Department of Health of the Hong Kong Special Administrative Region Government, Hong Kong. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: None declared.

Ethical approval: Approval had been obtained from the Ethics Committee of the Department of Health of the Hong Kong Special Administrative Region Government, and the Survey and Behavioural Research Ethics Committee of the Chinese University of Hong Kong.