As of 2005, there were a projected 650,000 persons living with HIV/AIDS in China.1 By transmission categories, injection drug users (IDUs) accounted for the largest share of persons living with HIV (44.3%), followed by sex workers and their clients (19.6%), partners of HIV-positive individuals and members of the general population (16.7%), and former blood donors and recipients of blood products (10.7%). Fewer data are available on men who have sex with men (MSM), but they are thought to account for 7.3% of persons living with HIV in China1 and the proportion may be growing. An estimated 11% of new HIV infections in 2003 were thought to be among MSM.2
A recent cause for concern is a report of rapidly rising HIV prevalence among the local MSM population of Bangkok, from 17.3% in 2003 to 28.3% in 2005.3 High HIV prevalence has been demonstrated among MSM in many areas of Asia in the past few years, including Phnom Penn, Cambodia (8.9%); Chiang Mai, Thailand (15.3%); and Andhra Pradesh, India (18.2%).4 HIV prevalence remains high and possibly rising in MSM populations across the Western world.4,5 So far, HIV prevalence measured among MSM in China has been less than 5.0% but is still relatively high compared with that in the general population.6-15 Meanwhile, measured levels of risk behavior or sexually transmitted diseases (STDs) among MSM have been consistently high.6-19
Surveys of MSM in China, as in virtually all parts of the world, face many limitations and challenges. Communities of MSM are relatively hidden and hard to reach because of the stigmatization of male-male sexual behavior. Participants of surveys are usually recruited by convenience sampling from gay bars or over the Internet. The representativeness of these surveys is therefore uncertain, and results are difficult to compare from location to location or from year to year. To our knowledge, there have been no rigorous, consistently implemented, serial, cross-sectional studies published to date that examine the trends in the HIV epidemic among MSM in China.
Because of the hidden nature of MSM populations, special methods are required to recruit MSM to participate in sentinel surveillance surveys. One effective sampling method is respondent-driven sampling (RDS).20-23 RDS relies on members of the target population to recruit other members of the target population while providing a theoretic basis to adjust statistically for the biases inherit in how persons associate and recruit from their social networks. The approach has been adopted to conduct HIV prevalence and behavioral surveillance in many hard-to-reach populations around the world, including MSM.22 In this article, we report on 3 consecutive years of applying RDS to measure HIV prevalence and risk behavior among MSM in Beijing.
Overall Study Design
We conducted 3 serial cross-sectional surveys of MSM in Beijing using RDS as part of HIV surveillance in Beijing, China in 2004, 2005, and 2006. The overall purpose was to track trends in the prevalence of HIV and other sexually transmitted infections (STIs) and related risk behavior in this population. In addition, we sought to field test the RDS methodology as part of ongoing surveillance for high-risk populations in China.
Study Population and Setting
Our target population was MSM residing in, working in, or living near Beijing. MSM were defined as ever reporting sex with another man. Participants were 18 years of age or older in 2004 and 16 years of age or older in 2005 and 2006. The age eligibility was lowered after 2004 so as not to turn away young MSM wishing to participate. The survey was conducted at the HIV voluntary counseling and testing clinic at the centrally located Beijing Centers for Disease Control and Prevention (CDC) office.
Sampling and Recruitment
The RDS methodology is based on long-chain recruitment, whereby members of the target population participating in the study refer other members of the target population to the study. The sample is therefore established through successive waves of participant referral. An underlying assumption of RDS is that the long-chain recruitment represents a first-order Markov process that reaches a dynamic equilibrium between the tendencies of persons with similar characteristics to associate with each other, relative network sizes, and underlying makeup of the target population.20 By tracking who recruits whom and by recording the relative network sizes of participants, data are collected by means of which the biases inherent in the referral recruitment can be quantified and adjusted for in the analysis. These biases are quantified by the “homophily” of each variable (ie, the propensity of persons to recruit others similar to themselves) and by individuals' relative network sizes. For example, homophily on age would be gauged by how often young recruit young or old recruit old versus young recruiting old and vice versa or how often MSM recruit others of the same education level compared with those across different education levels. Each variable has its own homophily and is adjusted for in the analysis. Relative network sizes affect a person's probability of being recruited into the study and that person's ability to recruit others. Data for the adjusted analyses are sufficient when the sample reaches “equilibrium” (ie, when additional waves of recruitment do not substantially change the composition of the sample with respect to key variables). In practice, equilibrium is usually achieved in 4 to 5 waves for most variables.20-22
Recruitment chains begin with “seeds” or persons purposefully selected as members of the target population-in this study, men who self-identify as having sex with other men. In theory, achieving equilibrium is independent of the starting seeds as long as referral chains are permitted to progress long enough and equilibrium is achieved. For practical purposes, however, a diversity of starting seeds is chosen to ensure that equilibrium is achieved in a reasonable time frame. The seeds are given coupons used to recruit other MSM-in our study, up to 3 per person. Persons who present with a coupon and who are eligible MSM are consented, enrolled, and, in turn, given 3 recruitment coupons until the sample size and equilibrium are achieved on key variables tracked during the progress of the study.
In 2004, we began with a single 32-year-old college-educated but unemployed MSM seed recruited from a nongovernmental organization. He was not an official resident of Beijing. In 2005, a total of 10 seeds were selected with a median age of 27 years: 6 had a college education, 8 were unemployed, and half were not official Beijing residents. In 2006, 8 seeds were selected with a median age of 33 years: 6 had a college education, 1 was unemployed, and 3 were not official Beijing residents. For 2005 and 2006, additional seeds were selected to include a diversity of ages and education levels. In addition, we invited MSM with high levels of social connectedness recognized in the preceding survey waves to act as seeds. In other words, some seeds for the 2005 survey were selected from participants of the 2004 survey, and some seeds for the 2006 survey were selected from participants of the 2005 survey. Of note, seeds are removed from the final sample for analysis. For their own participation in the survey, MSM received free HIV and STD counseling and testing; for recruitment of their peers, MSM received a monetary incentive roughly based on the costs of transportation across Beijing (US $2.1 per successful recruitment up to $6.3 total).
The analysis of RDS data uses specific software (Respondent-Driven Sampling Analysis Tool [RDSAT]) that adjusts for the long-chain referral recruitment design to produce population estimates. The software is free and downloadable with documentation on its use (available at: http://www.respondentdrivensampling.org). Crude sample estimates (ie, the proportions of key variables in the sample itself) are adjusted to reflect the makeup of the target population based on tracking who recruits whom and the relative sizes of participants' networks. Unique numbers on the recruitment coupons link recruiter to recruitee. Network size is based on self-report in response to a specific question. In 2004, the network size question was: “How many men do you personally know who have sex with other men, who you at least know by face but do not necessarily know their name or phone number?” By mistake, however, this question was only added one third of the way through the survey after the omission was detected. For the first third of the sample, we imputed network sizes for the missing values based on variables that were significant correlates of network size found in the later two thirds. For 2005 and 2006, the network size question was: “How many MSM do you know personally who have sex with other men, for whom you know their name and phone number?” In 2006, we verified the social connection by asking how men knew the person who gave them the coupon. As noted previously, all seeds were excluded from analysis. In this report, we present crude and population-adjusted point estimates and 95% confidence intervals (CIs) for key demographic characteristics and risk behaviors for each survey year.
Because RDSAT does not directly perform statistical testing, adjusted 95% CIs were used to interpret differences within and between survey years. In other words, we concluded that differences were significant if there was no overlap in their 95% CI. This approach may be conservative in that some minimal overlap may still achieve statistical significance at P = 0.05. We further examined HIV and hepatitis C virus (HCV) prevalence for different demographic strata using RDSAT. RDSAT is currently limited in its ability to conduct multivariate analysis. In addition, population-adjusted point estimates and 95% CIs were not produced in many stratified analyses, because numbers of cross-recruitments were too small for calculations for some cells (ie, not enough subjects recruited others of different characteristics within certain subgroups).
Participants presenting to the study were assessed for eligibility and, if eligible, consented. Participants completed a standardized structure questionnaire in a face-to-face interview with trained staff or the principal investigator (X. Ma). The questionnaire was based on the Family Health International instrument for MSM populations and on questions used in previous surveys of MSM in China and the United States.5,13,24 Items included basic demographic characteristics (eg, age, education, employment, marital status, whether participants were officially registered as a resident of Beijing [a status not necessarily attained by all persons living or working in the city]), sexual preference (for men only, for men and women, and for women), number of partners, condom use with insertive and receptive anal sex, use of drugs, and prior HIV testing. In 2006 only, MSM were asked their opinions on sources of information. On completion of the survey, blood was drawn for serologic testing for HIV, hepatitis B virus (HBV), HCV and syphilis.
HIV screening was conducted by using 2 enzyme-linked immunoassays (ELISAs; Shanghai Kehua Bioengineering, Shanghai, China, and BioMerieux bv, Boxtel, The Netherlands). If the results of one ELISA was positive, a Western blot test was used for confirmation (Abbott Laboratories, Tokyo, Japan). For 2004 and 2005, syphilis screening was done by ELISA (Jinghao Biotechnical Company, Beijing, China). Positive specimens were confirmed by rapid plasma reagin (RPR; Wantai Company, Beijing, China) testing, with any reactivity on both tests classifying subjects as having current syphilis. For 2006, all specimens were tested with ELISA (Jinghao Biotechnical Company) and RPR (Wulumuqi XinDi Company, Xinjiang, China) in parallel, and specimens with any reactivity on ELISA and RPR were considered as having current syphilis. HBV surface antigen was detected by ELISA (Wantai Company in 2004 and 2005 and Yingke Xinchuang Company, Xiamen, China in 2006), classifying subjects as having chronic HBV infection. HCV antibody was tested by ELISA (Murex Diagnostics, Dartford, United Kingdom). Test results disclosure and counseling were conducted on a scheduled return visit.
The protocol for this study is considered routine surveillance by the Beijing, China, and US CDCs. All information was collected confidentially, and participants were encouraged to use pseudonyms if they preferred. Patients testing positive for any infection were referred to treatment and follow-up. All subjects, including those testing negative, were provided individual risk reduction counseling, condoms, infectious disease prevention literature, and referrals to appropriate health services (eg, HBV vaccination) when needed.
The total numbers of MSM enrolled were 325 in 2004, 427 in 2005, and 540 in 2006. For each survey year, the time to recruit the total sample was between 3 and 4 months; from late September to early January in 2004, from early September to late December in 2005, and from early July to early October in 2006. For all years, recruitment chains progressed up to 15 or more waves depending on the particular branch. Equilibrium was reached in all years on all key variables examined, including age, employment, education, residence, marital status, and sexual preference within 5 generations in the latest case. In 2006, 8.5% of participants said they received their coupon from a stranger; this information was not recorded in 2004 and 2005.
Characteristics of the crude sample and population-adjusted estimates for all 3 survey years are presented in Table 1. Although the population-adjusted estimates for age and employment status remained stable over the 3-year period (ie, the population-adjusted 95% CI overlapped), other demographic characteristics changed. Most substantially, education level was higher in 2004 compared with 2005 and 2006. In 2004, most (76.9%, 95% CI: 67.0 to 85.5) MSM had more than a high school education compared with minorities in 2005 (30.7%, 95% CI: 24.4 to 36.1) and 2006 (30.7%, 95% CI: 25.7 to 36.7). Being officially registered as a resident of Beijing decreased from 38.0% (95% CI: 30.1 to 46.5) in 2004 to 17.3% (95% CI: 13.4 to 22.1) in 2006. Being currently married increased from 9.3% (95% CI: 5.2 to 14.8) in 2004 to 29.2% (95% CI: 24.8 to 34.4) in 2006. In 2004, more men reported a male-only sexual preference (56.1%, 95% CI: 49.7 to 65.7) compared with 2005 (39.5%, 95% CI: 33.2 to 44.4) and 2006 (43.2%, 95% CI: 37.9 to 49.2).
Table 1 also shows HIV risk-related variables by year. The overlap in 95% CI for unprotected insertive and receptive anal sex indicates no significant increase in consistent condom use over time. The proportion of MSM reporting multiple partners did increase, with those reporting 10 or more different sex partners in the preceding 6 months increasing from 2.4% (95% CI: 1.2 to 3.9) in 2004 to 17.4% (95% CI: 13.3 to 21.0) in 2006. More MSM had only 1 sex partner in the past 6 months in 2004 (36.6%, 95% CI: 29.3 to 43.6) compared with 2006 (17.4%, 95% CI: 13.3 to 21.0). History of ever having an STD increased from 15.1% (95% CI: 10.5 to 20.8) in 2004 to 27.9% (95% CI: 22.6 to 32.7) in 2006. No MSM reported injection drug use in any year (data not shown), and history of blood transfusion remained low and unchanged. In 2005 and 2006, less than 1 in 5 MSM reported having tested previously for HIV.
Figure 1 shows trends in the prevalence of HIV, current syphilis, chronic HBV, and HCV infection. The crude and population-adjusted prevalences of these infectious diseases are also presented in Table 1. HIV seropositivity was 0.4% (95% CI: 0.1 to 0.8) in 2004, 4.6% (95% CI: 2.2 to 7.6) in 2005, and 5.8% (95% CI: 3.4 to 8.5) in 2006. Syphilis infection increased from 4.5% (95% CI: 2.5 to 7.5) in 2004 to 12.4% (95% CI: 8.4 to 17.2) in 2005 and slightly decreased but remained high in 2006 (9.9%, 95% CI: 7.3 to 13.6). There was no evidence of a change in the prevalence of chronic HBV infection, ranging from 7.5% (95% CI: 3.7 to 11.6) in 2004 to 10.3% (95% CI: 6.9 to 13.9) in 2006. HCV seropositivity increased from 0.4% (95% CI: 0.1 to 0.8) in 2004 to 5.2% (95% CI: 2.3 to 8.2) in 2006.
Further investigation of HIV prevalence was conducted by stratified analysis (Table 2). Unfortunately, population-adjusted point estimates could not be calculated for HIV prevalence in substrata for 2004 because of the small number of HIV-positive MSM, although RDSAT generated the 95% CI. Comparing the 95% CI for 2004 and 2006, HIV prevalence rose among young and old MSM and among those with higher and lower education levels. HIV prevalence rose among MSM with and without official Beijing residence status but more substantially among MSM without. HIV prevalence rose among all income groups, but only the rise in the highest income group had no overlap in the 95% CI comparing 2005 and 2006 (data not available for 2004).
Stratified analysis of HCV infection in 2006 found prevalence higher in MSM not officially registered as residents in Beijing compared with those with official resident status. In 2006, HCV prevalence among unofficial residents was 5.0% (95% CI: 1.9 to 7.8) compared with 0.4% (95% CI: 0.1 to 0.5) among official residents. No MSM reported a history of injection drug use, and few reported a history of blood donation or transfusion (see Table 1).
In 2006 only, MSM were probed on their opinions toward sources of HIV information and services (Table 3). When asked which services or HIV prevention activities for MSM they considered most important or beneficial, the top response was having MSM-friendly HIV testing sites (74.4%, 95% CI: 69.8 to 79.2), followed by HIV/STD counseling services (71.1%, 95% CI: 65.6 to 75.9) and having an STD clinic (63.7%, 95% CI: 58.0 to 68.4). Fourth was having an HIV care clinic, dropping to 35.2% (95% CI: 29.6 to 40.4). Internet-based education was mentioned by 19.4% (95% CI: 15.1 to 23.6), and peer education was mentioned by 12.0% (95% CI: 9.2 to 15.6). The most trusted source for HIV prevention services was health professionals (59.4%, 95% CI: 53.6 to 64.7), followed by the government (47.0%, 95% CI: 41.7 to 52.2) and peer volunteers (21.0%, 95% CI: 17.1 to 26.0).
In 3 years of conducting repeated RDS surveys, we found prevalence of HIV among MSM in Beijing was <1% in 2004 and >5% in 2006. Supporting this possible rise is an accompanying increase in self-reported STDs and high rate of current syphilis infection by serology. Moreover, our data suggest that HIV prevalence is rising in younger MSM, which is a marker for more recent infection. Furthermore, there was no increase in consistent condom use, with more than half of MSM reporting unprotected insertive anal sex and more than two fifths reporting unprotected receptive anal sex in the past 6 months. This lack of consistent condom use, coupled with a significantly rising proportion reporting 10 or more recent sex partners, speaks to the behavioral conditions that may foster further spread of HIV. Our findings are particularly worrisome in light of the rapidly increasing HIV prevalence recently documented among MSM in Bangkok and the high HIV prevalence among MSM in diverse areas of Asia.3,4 The coincidence of such findings suggests there may be common societal and behavioral trends if not direct or indirect sexual interconnections.13
Of course, the observation of a rapid increase from such a low initial level merits careful examination, because small biases can greatly influence the relative change in HIV prevalence. For example, an underestimation of HIV prevalence in 2004 might drive the apparent upward trend. Indeed, the 2004 estimate is based on only 5 HIV infections detected. Moreover, we do note many differences in the 2004 sample compared with 2005 and 2006 samples. First, there was a much lower level of education among MSM surveyed in 2005 and 2006 compared with 2004. Nonetheless, we found no difference in HIV prevalence by education level, and there was a rise in HIV prevalence in both groups. Moreover, most starting seeds had a college education in all 3 survey years. Second, there were more unofficial Beijing residents in 2006, coupled with a more substantial rise in HIV prevalence among the unofficial residents. As with education, however, we found a rise in HIV prevalence in both groups, and the difference in HIV prevalence between the groups was not significant. It is noteworthy that there was an increase of unofficial residents in the final sample despite a decreasing proportion of initial seeds being unofficial residents (from 1 of 1 in 2004 to 3 of 8 in 2006). This finding illustrates how the chains of recruitment do not necessarily depend on the attributes of the starting seeds. The relations between HIV risk behavior, internal migration, and education level among MSM in China merit further examination in more in-depth quantitative and qualitative studies. Third, there were more single men in the 2004 sample compared with the 2005 and 2006 samples. The 95% CI for HIV prevalence overlap between single and married MSM does not support a difference between the groups, however. Thus, although such a dramatic shift in the demographic profile of MSM seems unlikely in the short time frame, stratified analysis suggests that these trends do not account for the apparent rise in HIV prevalence. We also acknowledge more logistic errors in 2004, our first year of implementation: the omission of the network size question for the first third of the sample, a difference in the wording of the network question from 2005 and 2006, a higher age limit in 2004, and the use of a single starting seed in 2004. Although equilibrium is theoretically independent of the starting seed20 and we observed that equilibrium was not necessarily correlated with the makeup of the starting seeds, it is possible that the RDS methodology did produce different samples in the different survey years.
Several other factors speak against a bias fully accounting for the rise in HIV prevalence. First, as mentioned previously, the rise in HIV prevalence is accompanied by a rise in syphilis (although with a leveling off in 2006), a rise in reported STDs, a rise in multiple partners, and no increase in always using condoms. These conditions are consistent with a fast rise in HIV transmission. Second, also mentioned previously, the increase in HIV prevalence is pronounced among younger MSM, in whom infection is more likely to be recent.25 Finally, even if you discount the 2004 HIV prevalence estimate, there was still an elevated prevalence from 2005 to 2006. For these years, we saw no substantial differences in the demographic makeup of the samples or in the implementation of the survey. Furthermore, HIV prevalence in 2005 and 2006 in our surveys is higher than the 3.1% HIV prevalence reported in 2002 in an independent survey by Choi et al.15,19 We believe the weight of evidence points to a high level of recent HIV transmission occurring among MSM in Beijing in the past few years, and prevalence may have crossed the 5% threshold for the first time in China.
We also found unexpectedly high and rising levels of HCV in this MSM population. These findings are puzzling, given that no MSM reported injection drug use and few reported a history of blood donation or transfusion. Anecdotally, examination of veins while collecting blood specimens (all done by X. Ma, the principal investigator) suggested that only a few MSM might have a possible history of injection drug use. Our estimate of HCV prevalence in 2004 is within the confidence limits of a 2004 survey of MSM in Guangzhou, which found an HCV prevalence of 1.0% (95% CI: 0.1 to 3.6).13 The same survey also found a low level of injection drug use among MSM (0.5%). Nondisclosure of injection drug use remains a possibility that should be explored with techniques that may reduce reporting bias, such as audio computer-assisted self-interview (ACASI). One clue for potential risk for HCV is that prevalence was higher among MSM who were not registered as official Beijing residents. This suggests that HCV infection may originate from exposures or practices done in other provinces or rural areas, such as injections for treatment of illnesses. Unfortunately, we did not collect data on such activities. Verification of the prevalence of HCV in this population and more detailed investigation of potential modes of transmission are needed in the next survey round and in other studies.
Although we acknowledge limitations in interpretation of our data and errors in implementation of our surveys, it is nonetheless important to remember that there exists no “gold standard” with which to compare our findings. In fact, most data on MSM worldwide originate from convenience samples or from random selection of venues and time periods when MSM congregate.5-19 The former approach has no theoretic basis for representativeness and does not ensure consistency over time. The latter approach, known as time-location sampling, presents a viable consistent alternative to RDS for sampling MSM5,26 but also lacks validation against a gold standard and misses MSM not attending the identified venues. In its favor, RDS presents a theoretic basis for statistical adjustment of samples obtained through long-chain referral, a means of reaching less visible segments of the target population, and a methodology that can be consistently applied from year to year and from location to location. Conversely, the analysis of RDS data is currently complicated and only produces point estimates of proportions, 95% confidence intervals, and the same within strata of the data. Although these basic functions are usually sufficient for wide use in surveillance, they are limited in testing specific hypotheses. Moreover, at the time this article was written, the multivariate analysis version of RDSAT was still under development.
Our efforts also show that RDS can efficiently recruit large samples of MSM in a short period. In Beijing, we were able to recruit several hundred MSM within a few months with 2 to 3 staff members and at relatively low cost compared with venue-based convenience samples. More detailed comparisons of relative costs of RDS with those of other methods are provided elsewhere.21,22 RDS has been widely adopted for HIV surveillance in many hard-to-reach populations worldwide, including sex workers and IDUs at high risk for HIV.21,22 We believe, although it remains to be tested or reported, that RDS may be efficient at sampling other hidden populations in diverse areas of China.
To our knowledge, our study is the first to report on a basic HIV surveillance function using RDS: tracking trends in the epidemic in the same population over a minimum of 3 years.24 The results of our study present an opportunity to demonstrate a second basic HIV surveillance function: generating a rapid response to prevent the further spread of infection. Findings from our study suggest the desire and need for MSM-friendly HIV and STD testing and treatment programs, health care provider sensitization and education, and expanded prevention programs.
The authors acknowledge Dr. Keith Sabin and Dr. Jinkou Zhao for their guidance on the study.
1. Ministry of Health, Peoples' Republic of China and Joint United Nations Programme on HIV/AIDS and World Health Organization. 2005 Update on the HIV/AIDS epidemic and response in China. 2006. Available at: http://www.unchina.org/unaids
. Accessed November 15, 2006.
2. State Council AIDS Working Committee Office and UNT Theme Group on HIV/AIDS in China. A joint assessment of HIV/AIDS prevention, treatment, and care in Chian (2004). Beijing, China, 2004. Available at: http://www.chinaids.org.cn
. Accessed November 15, 2006.
3. Centers for Disease Control and Prevention. HIV prevalence among populations of men who have sex with men-Thailand, 2003 and 2005. MMWR Morb Mortal Wkly Rep
4. United Nations Program on HIV/AIDS. AIDS epidemic update: special report on HIV/AIDS. December 2006. Available at: http://www.unaids.org/en/HIV_data/epi2006
. Accessed January 23, 2006.
5. Centers for Disease Control and Prevention. HIV prevalence, unrecognized infection, and HIV testing among men who have sex with men-five US cities, June 2004-April 2005. MMWR Morb Mortal Wkly Rep
6. Qu S, Zhang D, Wu Y, et al. The epidemiologic study of sex behaviors and HIV infection among male homosexual persons in somewhere north-east China. Chin J STD/AIDS Prev Cont
7. Yang Z, Fang S, Cai W, et al. A study of syphilis, HIV infected and sex behaviors among men who have sex with men. Chin Journal of Public Health
8. Wang J, Tong W, Li R, et al. A survey of HIV, STD serological and behaviors among men who have sex with men. Chin Journal of Public Health
9. Tao X, Cai W, Cai Y, et al. A high-risk behaviors study among 114 male homosexual persons in Shenzhen, China. Modern Preventive Medicine
10. Cai W, Feng J, Jingguang T, et al. A survey of behavior characters and STD/HIV infection among male homosexual persons, China. Modern Preventive Medicine
11. Jiang J, Cao N, Zhang J, et al. High prevalence of sexually transmitted diseases among men who have sex with men in Jiangsu Province, China. Sex Transm Dis
12. Gu Y, Qu P, Su L, et al. The KABP analysis related with STD/AIDS among male homosexual persons in Shenyang, China. Chin Journal of Public Health
13. He Q, Wang Y, Lin P, et al. Potential bridges for HIV infection to men who have sex with men in Guangzhou, China. AIDS Behav
14. Liu H, Yang H, Li X, et al. Men who have sex with men and human immunodeficiency virus/sexually transmitted disease control in China. Sex Transm Dis
15. Choi KH, Liu H, Guo Y, et al. Emerging HIV-1 epidemic in China in men who have sex with men. Lancet
16. Zhang BC, Li XF, Hu TZ. A survey of men who have sex with men: mainland China. Am J Public Health
17. Zhang BC, Li XF, Hu TZ. AIDS-related high-risk behaviors and affecting factors of men who have sex with men (MSM) in mainland China. Chin Journal of Sexually Transmitted Infection
18. Zhang BC, Shi TX, Cao NG, et al. Survey on the high risk behaviors and other AIDS/STI related factors among men who have sex with men (MSM) in mainland China (2001). Chin Journal of Dermatology
19. Choi KH, Lui H, Guo Y, et al. Lack of HIV testing and awareness of HIV infection among men who have sex with men, Beijing, China. AIDS Educ Prev
20. Heckathorn DD. Respondent-driven sampling: a new approach to the study of hidden populations. Soc Probl
21. Magnani R, Sabin K, Saidel T, et al. Review of sampling hard-to-reach and hidden populations for HIV surveillance. AIDS
. 2005;19(Suppl 2):S67-S72.
22. Abdul-Quadar AS, Heckathorn DD, Sabin K, et al. Implementation and analysis of respondent-driven sampling: lessons learned from the field. J Urban Health
. 2006;83(Suppl 7):1-5.
23. Ramirez-Valles J, Heckathorn DD, Vasquez R, et al. From networks to populations: the development and application of respondent-driven sampling among IDUs and Latino gay men. AIDS Behav
24. Family Health International. Behavioral Surveillance Surveys: Guidelines for Repeated Behavioral Surveys in Populations at Risk for HIV
. Arlington, VA: Family Health International; 2000.
25. Razani N, Schwarcz S, Klausner JD, et al. How well do trends in HIV prevalence in young people reflect HIV incidence? Results of 10 years of HIV serosurveillance in San Francisco. AIDS
26. MacKellar D, Valleroy L, Karon J, et al. The Young Men's Survey: methods for estimating HIV seroprevalence and risk factors among young men who have sex with men. Public Health Rep
. 1996;111(Suppl 1):138-144.