Abstract
HIV prevalence among men who have sex with men (MSM) has increased rapidly in China. Behavioral and biologic interventions are the key to control the spreading of HIV in the MSM population and the primary strategy for reducing the spread of AIDS in China. The objective of this study is to explore the relationship among HIV-related knowledge, the basic information of respondents, service utilization and risky sexual behavior. Respondent-driven sampling was used to recruit 225 MSM in Fushun and Huludao in China. The results of univariate analysis showed that condom use in the past 6 months was associated with age, being more knowledgeable about HIV, accepting lubricant distribution, and peer education (p<0.05). The structural equation modeling (SEM) results was as following, χ2=863.45 (p<0.01); root mean square error of approximation (RMSEA)=0.04; goodness of fit (GFI)=0.94, which indicated the model fitted the data well. The factor loads of HIV-related knowledge, the basic information of respondents, service utilization and risky sexual behavior was −0.06, 0.07, −0.27, respectively, which indicated that service utilization was likely to be a major factor negatively impacting risky sexual behavior. For service utilization, the greatest item load was for distribution of lubricants and peer education, 0.69, 0.68, respectively. The factor load of HIV-related knowledge and service utilization was 0.15, which suggested that risky sexual behaviors might be indirectly reduced by improving HIV/AIDS knowledge to increase service utilization. Basic information (age, income, marital status, age at first sexual intercourse) had a greater impact on service utilization, with a load factor of 0.26. For basic information, the greatest item load was age (0.96). In terms of the intervention strategies, it is essential that public health education is provided for the young, to ultimately decrease risky behaviors in MSM.
Introduction
In many countries, MSM behaviors are highly unacceptable, and there is strong social pressure against the behavior of MSM. 11 The traditional Chinese culture does not openly endorse MSM behaviors. 12 Approximately one half of MSM reported also had sex with a woman and one third had been married to a woman. 9 Therefore, MSM may play a bridging role in the spread of HIV and other STDs from the high-risk population to the general population. MSM have become one of the high-risk groups for HIV infection because of their sexual activities, which often occur without protection. Therefore, health interventions to reduce high-risk behavior are the key to control the spreading of HIV in MSM and the primary strategy to slow the spread of AIDS in China.
The research in this area is still in the exploratory stage in China, although some official behavioral surveillance and sexual health education programs have recently been piloted. However, official programs for MSM had little effect in promoting knowledge and healthy sexual practices among MSM. 13 –15 Therefore, effective strategies should be adopted to promote the HIV intervention in MSM. Behavioral and biologic interventions, such as pre-exposure prophylaxis (PrEP), have currently been studied as means to prevent HIV acquisition. 16,17 Many studies have investigated the factors associated with unprotected sexual behaviors. 18,19 However, due to inadequate research methods, the literature on the relationship between risk factors and high-risk sexual behaviors in MSM is limited. Since the 1960s and 1970s, structural equation modeling (SEM) has been a statistical analysis method that represents one of the three major advances in the field of statistics. 20 SEM has improved some of the shortcomings of other traditional statistical methods; e.g., both observed variables and nonobserved variables (latent variables) can be investigated using SEM, as can both the direct and indirect effects of variables through the development of a path map that shows the relationship between variables. In addition, it is possible to test and verify whether the relationships are reasonable with the help of SEM.
In this study, an SEM was built to model factors associated with unprotected anal intercourse (UAI) in a group of MSM in Liaoning Province, China.
Methods
Participants and procedures
A cross-sectional study was conducted in Fushun and Huludao, Liaoning Province, China, in October 2008. MSM were recruited using a respondent-driven sampling (RDS) method, and recruitment began with 10 MSM “seeds” (5 seeds from each city). Eight seeds extended the referral chain (Huludao City and Fushun City, with 4 seeds each). After the seeds completed the interview, each of them was given 3 coupons for recruiting their peers into the study. All new recruits were offered incentives for completing the interview and successfully recruiting up to 3 peers into the study.
The participants underwent a face-to-face interview with a standardized questionnaire, which was initially drafted based on the contents formulated by relevant previous studies. 21 –23 The questionnaire was divided into two parts. The first part included questions dealing with some basic information and the second part was about knowledge, service utilization and sexual-related behaviors. Before formal investigation, the investigators had been trained together to make them understand the purpose and meaning of the investigation, be familiar with the contents, the courses, and be skilled in the interview. They were also trained to explain any question in the questionnaire on the spot. The interviewers signed a form pledging that they had clearly explained the details of the study to the respondents. The interviews were anonymous, no identifying information was collected. Verbal informed consent was obtained from respondents before the anonymous face-to-face interview. Participation in the study was completely voluntary and the participants had the option of declining to answer specific questions or of leaving the entire questionnaire blank if they did not wish to participate. During the investigation, to ensure the quality of the questionnaire, the questionnaires had to be filled out face to face. After completion of the questionnaires, specially trained personnel would inspect the questionnaires, identify the questionnaires that were filled out with nonstandard and ambiguous answers, and find investigators and respondents for timely verification. An incentive of 50 Yuan (equivalent to approximately $7 US) was given to respondents as compensation for the time spent being interviewed. The study protocol was approved by the bioethics advisory commission of the China Medical University.
Data analysis
HIV/AIDS-related knowledge was measured with an eight-item set of questions covering the basic characteristics of the disease, routes of viral transmission, and measures of prevention. The knowledge questionnaire was reviewed for content validity by five experienced HIV/AIDS Chinese epidemiology professors. Respondents were given the option to reply that each statement was true or false or that they were uncertain. The service utilization was measured with six items. Each item was attached to a two-point response scale, “Yes” or “No.”
Respondents were asked whether they had ever received voluntary counseling and testing (VCT) services in the past 12 months. Risky sexual behavior was measured with two items. Respondents were asked if they had used a condom in the past 6 months and the frequency they used. Condom use in the past 6 months was attached to a two-point response scale, “yes” or “no.” The frequency of condom use was attached to a three-point response scores that ranged from ‘‘1’’ for “always” to ‘‘3’’ for “seldom.”
RDSAT software version 5.6 (RDS Software Inc., Nashua, NH) was used to generate the weighted population proportion estimates to make population-based prevalence estimates from data accumulated. Basic information, HIV/AIDS-related knowledge and service utilization variables were analyzed in relation to condom use in the past 6 months as the main outcome of interest. RDSAT also provided individual HIV outcome weights for use in multivariable analyses following methods outlined by Deiss and Abramovitz 24,25 We used the SEM to assess the multivariable association between our variables of interest, adjusting for the RDS process by using individual RDS-generated weights generated on the dependent variable (condom use in the past 6 months) as a covariate in the SEM.
The SEM that was used included the measurement model and structural model. 26,27 The whole model was based on the relationships between latent variables. Latent variables are ubiquitous in some research domains, while in other contexts, they are seldom used. In this study, the variables were the complexity of the HIV/AIDS-related knowledge, service utilization and behavior and underlying traits rather than any specific behavior.
The measurement models for the p endogenous observed variables represented by the vector y and the q exogenous observed variables contained in the vector x relate the observed (manifest) variables to the underlying factors (latent variables) and may be expressed as:
In the above equations, X is the p×1 vector of the exogenous observation variable; Λx is the vector of factor loadings relating X to ξ; ξ is an m×1 vector of exogenous latent variables; δ is a vector of measurement errors for X; Y is the q×1 vector of the endogenous variable; Λy is the factor loadings relating Y to η; η is the n×1 vector of endogenous latent variables; ɛ is the vector of measurement errors for Y.
The SEM mainly indicates the relationship between the latent variables. It provides the causality between the assumed exogenous potential variables and the inherent potential variables in the system. The equation is expressed as:
where B is vector of the coefficients relating η to η; Γ is the vector of coefficients relating ξ to η; ζ is the vector of residuals in equations.
Exploratory factor analysis (EFA) was first applied to summarize the factor structure. The analysis was performed using principal axis factoring with varimax rotation on the correlations of the observed variables. The context and meaning of items were also considered during the EFA. An SEM was used to estimate the magnitude and direction of the interdependent effects between the related factors that were identified. This general covariance structure contains both observed and latent variables. The factor scores from regressions obtained from the EFA were used to compute the composite variable regression coefficients and measurement error variances for the SEM (error variances not accounted for by the latent factors). This process ensures that the estimation of the scale/composite variable is proportionally weighted by the actual contribution made by each observed variable, thereby minimizing the measurement error in the items contributing to each scale. This also has the advantage of increasing the reliability (and validity) of the computed scale scores. For the SEM, assessing the model's adequacy may be based on various measures. In the present study, we considered (1) the root mean square error of approximation (RMSEA), which should be below 0.05 for a good fit, (2) the Satorra–Bentler χ2, (3) the standardized root mean square residual (SRMR), (4) the goodness-of-fit index (GFI), an index of the amount of variance and covariance explained by the model (5) the adjusted goodness-of-fit index (AGFI), adjusted for the degrees of freedom in the model, and (6) the incremental fit index (IFI). Values greater than 0.80 for indices (4) to (6) indicate an adequate fit of the model. 28
The preliminary EFA of the data was performed using SPSS version 13.0 (SPSS Inc., Chicago, IL) and the SEM were undertaken using LISREL 8.5 (Scientific Software Internation, Lincolnwood, IL).
Results
Participant characteristics
A total of 225 MSM were selected for the study, and 198 MSM completed the questionnaire (response rate, 88.0%). The participants were between 18 and 70 years old, with a mean age of 27.6 years (standard deviation [SD]=10.8). In adjusted analysis of nonseed respondents, 58.6% identified as gay, 41.4% as bisexual. Their marital status was as follows: married, 19.2%; unmarried, 74.2%; cohabiting, 0.5%; and divorced or widowed, 6.1%. Their monthly income levels were no income, 39.9%; below 1000 Yuan, 19.7%; 1000–2000 Yuan, 23.2%; 2000–3000 Yuan, 12.6%; and 3000 Yuan and above, 4.6%. The interviewees had first begun to engage in sexual intercourse from ages 13 to 37. Ages are significant for condom use in the past 6 months (p<0.05), and the first sexual intercourse ages are not significant for condom use in the past 6 months (p>0.05). Adjusted bivariate associations between participants' marital status, monthly income, sexual orientation and condom use in the past 6 months (using RDSAT-generated weights) are included in Table 1.
RDS Software Inc., Nashua, New Hampshire.
MSM, men who have sex with men; CI, confidence interval.
HIV/AIDS knowledge
A total of 95.5%, 94.9%, and 93.4% of the subjects were aware that HIV was transmitted by “transfusions tainted with HIV,” “sharing intravenous needles,” and “from mother to child,” respectively. The proportion of subjects that incorrectly selected the nontransmission routes was low, e.g., “mosquitoes or other insects' (32.8%).” Knowledge of the items “people who look healthy might possibly carry HIV,” “blood transmission,” “needle transmission,” and “correct use of condoms could reduce the risk of HIV spreading” are significant for use of condoms in the past 6 months (p<0.05). Details of HIV/AIDS knowledge are shown in Table 2.
RDS Software Inc., Nashua, New Hampshire.
p<0.05.
HIV/AIDS service utilization
Results showed that the ratio of sexually transmitted disease (STD) screening, AIDS counseling, and informational materials are 23.3%, 36.4%, and 55.6% respectively. A total of 49.0%, 46.5%, and 46.0% of the subjects received condoms, lubricant, and peer education, respectively. Accepting lubricant and peer education were significant for the use of condoms in the past 6 months (p<0.05). The results of social service utilization were illustrated in Table 3.
RDS Software Inc., Nashua, New Hampshire.
Reliability and validity analysis
The investigation had a Cronbach α coefficient of 0.85. The analysis of EFA was performed using principal axis factoring with varimax rotation on the correlations of the observed variables. The Bartlett's test for sphericity was found to be significant (p<0.01), while the Kaiser–Meyer–Olkin measure of sampling adequacy was 0.78, justifying the application of a factor-analytic procedure. Eight factors with four values greater than 1 were extracted, accounting for 88% of the variance. The remaining four factors explained about 12% of the variability and were therefore dropped from further analysis. After a substantial review of the literature, these factors were classified into four domains: basic information, HIV/AIDS knowledge, service utilization, and risky sexual behavior. The latent variable of basic information included “age,” “marital status,” “income per month,” and “age at first sexual intercourse.” The knowledge latent variable included “no apparent symptoms,” ‘blood transmission,” “needle transmission,” “condom can prevent HIV,” “mother to child transmission,” and “mosquito transmission.” The service utilization latent variable and risky sexual behavior included all items.
Relationship between HIV/AIDS knowledge, basic information, service utilization, and risky sexual behavior
To assess the factor structure underlying the 21 items of the survey, a confirmatory factor-analytic measurement model was fitted to the data using the scaled covariance matrix of the polychoric correlations. In the solution, four correlated factors are labeled: basic information (η 1), HIV/AIDS knowledge (η 2), service utilization (η 3), and risky sexual behavior (η 4). To facilitate the interpretation of Fig. 1, the four factors are bounded by ellipses and the item descriptions within each first-order factor are bounded by rectangles. For example, the loading of the items for the “HIV/AIDS knowledge” factor (η 2) ranged from 0.55 for “condom can prevent HIV” to 0.90 for “blood transmission.” It is evident that all the factor loadings were high (ranging from 0.55 to 0.90), indicating a strong association between each of the latent factors and their respective items. The factor loadings also made substantive sense, as they were meaningful and statistically significant. The relationships between the categories of basic information, HIV/AIDS knowledge, service utilization, and risky sexual behavior must also be significant. The results showed that convergent validity was achieved.

The structural equation modeling of the factors that influence risky sexual behaviors among men who have sex with men (MSM). Results of the study of the relationship between factors and risky sexual behavior regarding human immunodeficiency virus/acquired immune deficiency syndrome (HIV/AIDS) in Chinese MSM (n=198) showed that basic information and HIV/AIDS-related knowledge had little influence on risky sexual behavior. Service utilization and risky sexual behavior were the two most closely related factors with a factor load of −0.27, which indicated that the risky sexual behavior could be constrained, in part, by increasing the acceptance of service utilization.
To assess the relationships between the latent constructs and items obtained from the preliminary EFA and to determine whether there was a construct common to these factors, a hypothesized recursive SEM was used to estimate the magnitude and direction of the interdependent effects between the factors identified to be related using the maximum likelihood estimation method. This general covariance structure includes both observed and latent variables. The best fit solution is illustrated in Fig. 1. The goodness of fit for the full model showed that it fit the data well (χ2=863.45 [p<0.01]; RMSEA=0.04; GFI=0.94). The AGFI index indicated that 95% of the relative variance and covariance of the data was accounted for. This implies that the hypothesized casual model of the associations among the constructs is tenable. In the analyses reported here, no further modifications were performed on the model to achieve a better fit. All other indices and measures were satisfactory. The path coefficients proximal to the unidirectional arrows (from ellipse to ellipse) shown in Fig. 1 are the standardized estimates of the direct effects of one factor score on the other. These may be interpreted as standardized regression coefficients. For example, an increase of one standard deviation in service utilization (η 3) can lead to a corresponding decrease of 0.27 standard deviation units in risky sexual behavior score (η 4). All estimated path coefficients were significant. For service utilization, the greater item loads for accepting lubricants and peer education were, 0.69, 0.68, respectively. Therefore, it may be effective to vigorously strengthen the programs for lubricant distribution and peer education to reduce risky sexual behavior. HIV/AIDS-related knowledge had a minor influence on the risky sexual behavior of the study subjects (factor load –0.06), however, it had an important influence on the service utilization with a factor load of 0.15, which suggested that risky sexual behaviors might be indirectly reduced by improving HIV/AIDS knowledge to increase service utilization. The factor load of the basic information of respondents and risky sexual behavior was 0.07. Basic characteristics had a greater impact on service utilization, with a load factor of 0.26. For basic information, the greatest item load was for age (0.96), which indicates that age was likely to be a major factor impacting basic information.
Discussion
At present, MSM have become one of the high-risk groups for HIV infection and other STDs for their active sexual activities without protection. 4,29 There are 2–8 million MSM in China and the rapidly rising prevalence of HIV/AIDS among this high-risk population draws worldwide concerns. 30 Our results showed that 19.2% of MSM were married, raising the possibility of the MSM population being a bridge population that spread AIDS from high risk populations to average-risk populations. 31 Therefore, an effective strategy for prevention and control is needed for this specific population.
We assessed the relationship and the interdependent effects between four latent factors. The results provided some further information on the interrelationships between the factors affecting MSM risky sexual behavior. Service utilization and risky sexual behavior were the two most closely related factors with a factor load of −0.27 (Fig. 1), which indicated that the risky sexual behavior could be constrained, in part, by increasing the acceptance of service utilization. Service utilization consists of the following factors: condom, lubricants, peer education, STD screening, AIDS counseling. Lubricant distribution and peer education had larger item loadings (0.69, 0.68, respectively) indicating that they had a greater impact on AIDS social service utilization. The results of univariate analysis also showed that lubricant distribution and peer education were significant for the use of condoms in the past 6 months (p<0.05). It was reported that peer education activities had a significant impact on sexual behavior at the community level. 32 –35 Peer-implemented education was better accepted, as the messages were more adapted to individuals' needs. Peer-based interventions centered on MSM practices were more adapted to the needs of MSM and offered them a rare chance to have access to lubricants. However, peer-based actions and lubricant distributions have not been effectively reaching the target in the MSM population in China. Therefore, it might be more effective to vigorously strengthen the program of lubricant distribution and peer education in reducing high-risk sexual behavior. Studies have suggested a positive association between condom use in male-to-male anal intercourse and access to HIV prevention actions. 36 –40 In our study, the item load of frequency of condom use was 0.96, which indicated that the risky sexual behavior associated with HIV infection could be constrained, in part, by increasing condom use.
This study also found that HIV/AIDS knowledge played a protective role against risky sexual behaviors, but that its impact was small (only −0.06). HIV/AIDS knowledge had a greater impact on service utilization, with a load factor of 0.15, which suggested that the risky sexual behaviors might be indirectly reduced by improving HIV/AIDS knowledge to increase service utilization. Some studies showed that age, income, and other basic characteristics influenced risky sexual behaviors. 41,42 In our study, the impact of basic characteristics was small (only 0.07). Basic characteristics had a greater impact on service utilization, with a load factor of 0.26. For basic information, the greatest item load was for age (0.96), which indicates that age was likely to be a major factor impacting basic information. Different interventions might be required for different age groups. In terms of the intervention strategies, public health education provided for the young people is essential to ultimately decrease risky behaviors in MSM.
There were also some limitations in this study. First, MSM are a hidden population. Because of privacy issues and because certain MSM behaviors may be discriminated against or be illegal, MSM might not will to fully cooperate or maybe just give some false answers to protect their privacy. Second, it is difficult to determine the sampling frame and to obtain a representative sample of MSM. Thus, this study used a respondent-driven sampling method. However, this method may have introduced some sampling bias because it included subjects who were willing to cooperate. Third, our study was conducted in a limited urban area in China. Further work that focuses on extending the research into the broader areas in China would be valuable.
Footnotes
Acknowledgment
This study was supported by the National Natural Science Foundation of China (Grant Number 30700690), and we thank Ming Ren, Center for Disease Control and Prevention of Funsun City for his excellent help with the survey.
Author Disclosure Statement
No competing financial interests exist.
