Men who have sex with men (MSM) continue to be disproportionately affected by HIV globally [1]. In Asia, MSM account for 18% of all newly identified HIV diagnosis, and the rising prevalence has been documented in China [2-6]. The national HIV prevalence among MSM in China in 2015 was 8.0%, 200 times higher than 0.04% in China’s general population [7]. The low HIV testing rate continues to contribute to the spread of HIV among MSM. Although China has significantly expanded its efforts to control HIV transmission [8], systematic reviews suggest that the HIV testing rate among the Chinese MSM remains low [9-11]. Barriers to HIV testing include the lack of MSM community engagement, hesitancy to access facility-based services, and low level of trust toward facility-based services [12].

Influential individuals facilitate the spread of certain behaviors within a population, as demonstrated in both diffusion research [13,14] and social network research [15-17]. The social diffusion theory suggests that behavior change in a population can be initiated and diffused to others if a behavior is visibly endorsed by enough natural and influential opinion leaders within the population [14]. These influential individuals informed the concept of popular opinion leaders (POLs) and the development of POL-based HIV interventions that successfully identified and trained popular individuals to spread HIV prevention messages to peers [18,19]. Social networks can amplify the spread of behavior through interpersonal ties and have been used by public health interventions to promote a range of health behaviors [20-22]. The targeting of influential individuals also makes social network interventions more effective and efficient, possibly because of optimal properties associated with an influential individual’s network structure [17]. This body of evidence indicates that behavioral change in social networks may be accelerated through influential individuals.

In the internet era, MSM increasingly turn to Web-based networks and social media to look for sexual health information [23-25]. Popular MSM who frequently share sexual health information with their friends or social media followers may influence the health behaviors of people among their networks. However, understanding of such influential individuals in Web-based MSM networks remains incomplete. First, although social media is more available to MSM as a way to disseminate health knowledge [26], its use among Web-based influential individuals has not been fully explored [19]. Second, unlike POL interventions where the behaviors spread by influential individuals are purposefully established by POL-targeted training, Web-based networks also facilitate the spread of behaviors that are naturally endorsed by influential individuals themselves [15]. However, the sexual risk behaviors, HIV and syphilis testing behaviors of Web-based influential individuals have not been thoroughly described, creating uncertainty to the degree of these individuals’ positive influence.

Knowledge of both the use of social media and the set of health behaviors endorsed by Web-based influential individuals can inform the development of MSM-led, network-based interventions. The high rates of internet and mobile phone usage in China particularly provide a strong foundation for network-based interventions on social media [27-29].

In this study, we aimed to examine Web-based sexual health influencers (SHIs), defined as individuals with strong influence on spreading HIV and sexually transmitted infection (STI) information online, among Chinese MSM to describe their social-behavioral characteristics, including sexual behaviors, HIV and syphilis testing, social media engagement, community engagement, and HIV-relevant psychological profiles such as anticipated HIV stigma and HIV testing self-efficacy.

An online survey of 1031 MSM was conducted in August 2017 as a final follow-up of a step-wedged randomized controlled trial to improve HIV testing rates in 8 Chinese cities (Guangzhou, Jiangmen, Zhuhai, and Shenzhen in Guangdong province and Yantai, Jinan, Qingdao, and Jining in Shandong province). The cities of each province were randomly assigned the order of intervention and paired into 4 groups accordingly. The methods have been described in detail elsewhere [30]. Men were recruited through a gay social networking mobile phone app, Blued, by sending a survey invitation to registered users in the 8 selected cities. Men were included if they were born biologically male, aged 16 years and above, ever had sex with another man, and HIV negative or with unknown HIV status. All individuals completed informed consent. The intervention included individual- and community-level components. We sent campaign images and texts promoting HIV testing, along with local testing site information, to all participants privately on WeChat, an instant messaging tool (individual-level component). Local community-based organizations organized a crowdsourcing contest soliciting individual stories relevant to HIV testing among the MSM community with an aim of improving community engagement (community-level component). We followed the participants quarterly for 1 year. We conducted the final follow-up survey immediately after the completion of the final round of intervention. We conducted a secondary analysis of the final follow-up survey data for this study.

Due to the small cells of self-reported influence on others’ adoption of HIV test within influencers, we dichotomized the variable and grouped many and quite a lot into 1 category for further analysis. Descriptive analysis was conducted to describe the differences in sociodemographic characteristics, sexual behaviors, HIV and syphilis testing history, exposure to interventional materials, use of social media platforms, anticipated HIV-related stigma, HIV testing social norms, HIV testing self-efficacy, and community engagement between influencers and noninfluencers. Chi-square or Fisher exact test was used for categorical variables, and independent sample t test was used for continuous variables.

We conducted multivariable linear or logistic regression analyses to examine the association between opinion leadership and various characteristics outcomes aforementioned. Mean differences in continuous variables between influencers and noninfluencers were evaluated, whereas odds ratios were reported for binary outcomes. Given the effects of trial interventions on these outcomes as well as its potential association with leadership, we controlled all previous interventions as predictors in the crude model (model 1). In addition, sociodemographic factors including age, education, income, and marital status were controlled in an extended model (model 2). We reported 95% CIs and P values. A P value of less than .05 was considered statistically significant. Data were analyzed with SAS software, version 9.4 (SAS Institute, Cary, USA).

Ethical approval was obtained from the institutional review committees at the Dermatology Hospital of Southern Medical University (Guangzhou, China), Shandong University (Jinan, China), University of North Carolina at Chapel Hill (Chapel Hill, North Carolina), and the University of California, San Francisco (San Francisco, California) before the launch of the survey.

Among 1031 men, 132 (132/1031, 12.80%) had a mean leadership score of greater than 3 and were categorized as influencers and 899 (899/1031, 87.20%) were noninfluencers (Multimedia Appendix 1). Most men were younger than 30 years (819/1031, 79.44%), were living in the sampling city (926/1031, 89.82%), were never married (907/1031, 87.97%), obtained college or above education (667/1031, 64.69%), and self-identified as gay (741/1031, 71.87%; see Table 1). Compared with noninfluencers, influencers were more likely to disclose their sexual orientation to others (100/132, 75.8% vs 580/899, 64.5%; P=.01), disclose sexual orientation to health care providers (37/132, 28.0% vs 178/899, 19.8%; P=.03), have a casual male partner in the last 3 months (63/132, 47.7% vs 315/899, 35.0%; P=.05), test for HIV in the last 3 months (73/132, 55.3% vs 337/899, 37.5%; P<.001), test for syphilis in the last 3 months (35/132, 26.5% vs 137/899, 15.2%; P=.001), and self-report that many or quite a lot of others had taken an HIV test after their intervention within their network (42/132, 31.8% vs 38/899, 4.2%; P<.001). There were no significant differences between the 2 groups in terms of the average number of male sex partners (1.32 vs 1.11; P=.16), condomless sex with male partners in the last 3 months (26/132, 19.7% vs 203/899, 22.6%; P=.46), and mainly meeting male sex partners online in the last 12 months (97/132, 73.5% vs 669/899, 74.4%; P=.82).

Table 2 shows exposure to the intervention trial and social media engagement by influencers and noninfluencers. Being in the same cohort, influencers were more likely to have seen our interventional materials, including images promoting HIV testing (125/132, 94.7% vs 759/899, 84.4%; P=.002), texts promoting HIV testing (120/132, 90.9% vs 672/899, 74.7%; P<.001), and a local crowdsourcing contest (63/132, 47.7% vs 312/899, 34.7%; P=.004). In terms of using social media platforms to give and receive any extra information relevant to HIV testing (except for the information delivered by the trial), influencers were more likely to give or receive extra information than noninfluencers on Weibo (42/132, 31.8% vs 112/899, 12.5%; P<.001), WeChat (85/132, 64.4% vs 275/899, 30.6%; P<.001), QQ (54/132, 40.9% vs 125/899, 13.9%; P<.001%), and mobile apps (42/132, 31.8% vs 169/899, 18.8% P=.001). As shown in Table 3, compared with noninfluencers, influencers had significantly more QQ followers (238 vs 159; P=.03), less Blued followers (172 vs 466; P=.003), a lower anticipated HIV stigma score (2.7 vs 2.9; P<.001), a higher HIV testing self-efficacy score (3.4 vs 3.1; P<.001), and a higher community engagement score (4.0 vs 2.5; P<.001).

As shown in Table 4, being influencers was associated with lower mean scores of anticipated HIV stigma (mean difference −0.22; P=.02), higher mean scores of HIV testing self-efficacy (mean difference 0.24; P=.01), and higher summed scores of community engagement (mean difference 1.48; P<.001). The differences remained significant after controlling for sociodemographic factors additionally in model 2.

Table 5 shows comparisons of binary outcomes between the 2 groups. Being influencers was found to be significantly associated with higher odds of HIV testing in the last 3 months (adjusted odds ratio, AOR 2.16, 95% CI 1.48-3.17), HIV self-testing in the last 3 months (AOR 1.64, 95% CI 1.10-2.46), HIV facility-based testing in the last 3 months (AOR 2.66, 95% CI 1.79-3.96), and syphilis testing in the last 3 months (AOR 1.95, 95% CI 1.25-3.03). Influencers were 1.90 (95% CI 1.19-3.02), 3.79 (95% CI 1.87-7.66), and 2.91 (95% CI 1.79-4.75) times more likely to use Weibo, WeChat, and QQ, respectively to give or receive extra HIV testing–relevant information than noninfluencers. In terms of self-reported influence on others’ HIV test uptake, influencers were 7.62 (95% CI 4.55-12.78) times more likely to report that many or quite a lot of people within their network have gone for an HIV test after their intervention. However, opinion leadership was found not to be associated with consistent condom use.

Influential individuals may help promote health behaviors within their networks; however, there is insufficient understanding of the characteristics of Web-based influential individuals who are increasingly important to health promotion in the internet era. By examining social media engagement and health behaviors of Web-based SHIs among Chinese MSM, this study extends the literature by illuminating the degree to which existing influential individuals may be useful agents in the Web-based virtual space and by informing the development of network-based social media interventions. Our study found that influencers had higher social media engagement for HIV testing, higher likelihood of HIV and syphilis testing, and did not have increased sexual risk behavior. This group could become key leaders within network-based social media MSM HIV interventions.

We found that influencers had a higher HIV testing rate than noninfluencers. The HIV testing rate of influencers was higher than that of Chinese MSM in another postintervention study [35]. After adjusting for the intervention, influencers in this study also had a higher likelihood of HIV testing. Their higher HIV testing may be related to increased community engagement in sexual health [36] and HIV testing–related social media use [37]. The higher rates of HIV testing among influencers may also be related to lower HIV stigma and higher testing self-efficacy, which are 2 important contributors to test uptake [32,38]. Influencers also had a higher rate of syphilis testing than noninfluencers in the last 3 months (26.5% vs 15.2%). This is consistent with previous studies that found community engagement in sexual health to be associated with increased syphilis testing [36]. This suggests that influencers could be helpful in promoting dual HIV and syphilis testing, given that these related infections often co-occur among MSM in China [39].

In terms of influence on others’ adoption of HIV test, we found influencers were significantly associated with reporting that many people within their Web-based social network have taken an HIV test after their intervention. This may be explained by greater social media engagement and larger social network sizes among influencers. Influencers had a greater exposure to our trial intervention materials, including seeing any images or texts promoting HIV testing and local testing sites information. They were also more active in using various social media platforms to communicate with others about HIV testing. Being more active in receiving and disseminating sexual health information indicated that natural influencers may be more central to information flow, hence facilitating healthy behavior spread within their social network.

Finally, there was no significant difference between influencers and noninfluencers with regard to sexual behaviors. Specifically, rates for condomless sex, having a regular male partner, meeting sex partners online, and the mean number of male sex partners were similar. These findings are relevant to the growing body of research on internet use and sexual risk behaviors among MSM. On the one hand, internet use and social media are believed to be an avenue for meeting MSM, who then engage in risky behaviors associated with transmission of HIV and other STIs [40,41]. However, our study revealed that influencers had less followers than noninfluencers in Blued, the most popular gay social networking app in China. This indicates that influencers may in fact use this app less often for finding dates online. Furthermore, social media use could also potentially decrease sexual risk behaviors as it allows MSM to discuss with others about sexual health and learn about HIV and STI prevention [42]. Influencers in this study had a higher degree of online usage and communication with other MSM; however, they do not have increased sexual risk behaviors.

We noted some limitations of the study. First, the study tended to focus on describing the influencers’ characteristics and behaviors. Yet, descriptive studies on influencers are valuable, given that health behaviors are known to spread from person to person in social networks [43,44]. Our study may also provide a mechanism (ie, centrality in information flow) explaining why naturally existent influencers may facilitate behavioral change within their own network. Second, we only evaluated opinion leadership in the final follow-up survey and were not able to examine the longitudinal effects of the intervention on the relationship between opinion leadership and testing behaviors. Instead, we controlled for the intervention in the models to eliminate the confounding effects of interventions on the outcomes. Third, we used self-reported data to measure leadership and their influence on others’ HIV test uptake. More reliable measurement and more robust research are necessary to evaluate the effect of influencers within MSM social networks with regard to behavior change. Incorporating social network measurement (eg, eigenvector centrality) and personal influence (eg, opinion leadership) and measuring their effects on positive behavior spread are worth consideration. Finally, our cross-sectional design makes it difficult to determine causal relationships.

Our findings have implications for strengthening HIV and syphilis testing interventions. Web-based SHIs could be useful in a range of testing promotion models, particularly network-based social media approaches. Vulnerable populations such as MSM may distrust outside authorities but find advice from known influencers who carry credibility [45]. For example, influencers could serve as steering committee members of crowdsourcing contests that aim to promote testing [30,46]. Influencers could also serve as leaders in network-based social media interventions to allow health messages to reach more MSM [37]. We found a higher rate of social media engagement about HIV testing and a higher likelihood of having used social media for HIV testing in influencers compared with other MSM, suggesting influencers may be readily incorporated into social media HIV interventions [37]. Influencers additionally contribute by being well positioned in social networks to spread behavior change. Interventions that operate through existing influencers’ social networks hold promise for reaching vulnerable communities, particularly when formal prevention infrastructure supports are limited [17,47].