Both Wuhan and Changsha are provincial capital cities located in Central China, with an estimated web-based MSM population of around 20,000, which is among the highest in China [23]. In these 2 cities, there are respectively 2 and 3 social organizations dedicated to providing local MSM with rapid HIV testing, counseling, and sexual health education. This study selected 2 of these organizations with which we have established long-term collaborations as study sites. These 2 social organizations are located in Wuhan and Changsha, specializing in providing health-related services to the MSM community, and closely collaborating with the local Center for Disease Control and Prevention (CDC) to offer HIV-related counseling and rapid HIV testing services. Visitors who test positive during the initial rapid screening are referred to the local CDC’s HIV Voluntary Counseling & Testing clinic for confirmatory testing. We collaborate with these 2 social organizations to recruit study participants, as they have a strong reputation and are highly regarded within the local MSM community.

Participants were consecutively recruited from August 1 to October 30, 2020. To be eligible, participants had to meet the following criteria: (1) aged ≥18 years, (2) male, and (3) having engaged in anal intercourse with a man in the past 6 months. Participants who were cognitively impaired or unable to understand the questionnaire were excluded. The social organizations’ staff approached eligible seeds during their HIV counseling visits and explained to them our study purpose and procedure in detail. Those who were interested in participating in the study were then referred to our research team. After providing informed consent, the eligible participants were invited to complete a battery of web-based questionnaires with our research team.

In this study, respondent-driven sampling (RDS) was used to ensure that the participants were representative. RDS has been widely used in recruiting hidden populations including MSM [24-26] and reduces biases that are common in other nonrandom sampling methods, such as snowball sampling [27,28]. The study began by recruiting 8 participants (seeds) with different occupations, such as gay bar owner, student, etc, and larger social networks among MSM. The basic characteristics of the 8 seeds are shown in Table 1. To ensure diversity among participants, the seeds did not know each other, and their social networks did not overlap. After completing the questionnaire, the seeds were given 3 coupons and asked to invite a maximum of 3 eligible peers from their gay social network to participate. The flowchart of participant recruitment is shown in Figure 1. Participants would add the study’s official WeChat account using the information on the coupon. After confirming their identity, the investigator would instruct them to complete the web-based questionnaire. After completing the questionnaire and passing quality control, each participant received 20 ¥ (approximately converted to US $2.7) as compensation. The process would then be repeated, with participants receiving 3 coupons to invite 3 more peers until the desired sample size was reached.

Descriptive statistics, such as frequencies and percentages, were calculated for categorical variables, and means and SDs were calculated for continuous variables. The RDS method allowed us to control for selection bias and make statements about the population without being restricted to the sample [33]. To account for the size of participants’ social networks, we adjusted the analysis and calculated the corresponding 95% CIs. All analyses were performed by the RDS package in R (version 4.1.3; R Core Team).

The path analysis was conducted to assess the relationship between psychosexual characteristics (sexual compulsivity and sexual sensation seeking), gay app use, and high-risk sexual behavior (unprotected anal intercourse and multiple sexual partners). The mediating effect of gay app use on the relationship between psychosexual characteristics and high-risk sexual behavior was tested using the nonparametric percentile bootstrapping method. The statistical significance of the estimates was evaluated based on a probability criterion of P<.05. The data analysis was performed using Amos (version 23.0; IBM Corp).

The study was approved by the Ethics Review Committee of Xiangya Public Health, Central South University (XYGW-2020-49). By signing the electronic informed consent form, participants acknowledged that their participation was voluntary and that they could withdraw at any time without any negative consequences. All personal information collected from the participants were kept confidential and stored securely in encrypted storage spaces. The results of the study were presented in a way that ensured the participants’ anonymity and protected their privacy.

Figure 2 shows the tree diagram of RDS recruitment results. The 8 seeds recruited a total of 402 MSM through RDS, with the longest recruitment wave consisting of 10 waves. The maximum number of individuals recruited by a single seed was 102. The study showed that as recruitment progressed, the basic characteristics of MSM, such as sexual orientation, unprotected anal intercourse, and multiple sexual partners, reached a balance at the third, fifth, and fourth layers, respectively (with a mean absolute difference of less than 2%). Additionally, the median size of social networks in this survey was 24 (IQR 10-60)individuals. The adjusted composition ratio revealed that 90.9% (n=199) of MSM had social networks with 24 or fewer individuals, as shown in Multimedia Appendix 1. Furthermore, we have provided a regional map of the sample’s household registration locations. Please refer to Multimedia Appendix 2 for details.

Table 2 shows the sample characteristics with and without adjustment. Half of the participants (53.73%, n=216) were under 23 years old, with 98.26% (n=395) of them being single and only 7 participants being married. Most of the participants (89.05%, n=358) had a bachelor’s degree or higher and 47.52% (n=191) had a monthly income of more than 3000 ¥ (~US $411). Most participants (81.09%, n=326) self-identified as homosexual. In terms of gay app use, most participants (67.42%, n=271) have used gay apps for the past 6 months, including 35 (8.71%) who used it everyday. A total of 69 (17.16%) participants reported significant sexual compulsivity and 113 (28.11%) reported significant sexual sensation seeking. Regarding high-risk sexual behaviors, 149 (37.06%) participants reported engaging in unprotected anal intercourse and 218 (54.23%) reported having multiple sexual partners in the past 6 months.

The average scores of sexual compulsivity and sexual sensation seeking were 17.43 (SD 6.63) and 24.03 (SD 7.67), respectively. The average scores of multiple sex partners and unprotected anal intercourse were 2.51 (SD 3.61) and 4.29 (SD 1.13), respectively. The average score of gay app use was 3.17 (SD 2.02). Correlation analysis showed that sexual compulsivity was positively correlated with gay app use (r=0.11; P=.03) and multiple sex partners (r=0.25; P<.001). Sexual sensation seeking was positively correlated with gay app use (r=0.11; P=.31) and multiple sex partners (r=0.24; P<.001). Gay app use was positively correlated with both multiple sex partners (r=0.24; P<.001) and unprotected anal intercourse (r=0.15; P=.004), which were correlated with each other (r=0.24; P<.001). The detailed results are presented in Table 3.

Figure 3 and Table 4 show the results of mediation analyses of gay app use on the relationship between sexuality traits (sexual compulsivity and sexual sensation seeking, respectively) and high-risk sexual behaviors (multiple sexual partners and unprotected anal intercourse, respectively). As shown in Table 4 and Figure 3, there was a significant direct (effect=0.227, 95% CI 0.070-0.358; P=.002) and indirect effect (effect=0.023, 95% CI 0.004-0.067; P=.01) between sexual compulsivity and multiple sexual partners. Gay app use acted as a partial mediator, accounting for 9.2% of the total effect. Additionally, there was a significant indirect effect (effect=0.015, 95% CI 0.004-0.031; P=.01) between sexual compulsivity and unprotected anal intercourse, although the direct effect (effect=0.082, 95% CI 0.011-0.150; P=.06) was not found to be significant. These results suggest that gay app use acts as a complete mediator in this relationship.

As shown in Table 4 and Figure 3, there was a significant direct effect (effect=0.218, 95% CI 0.098-0.305; P<.001) and indirect effect (effect=0.023, 95% CI 0.003-0.068; P=.03) between sexual sensation seeking and multiple sexual partners. Gay app use acted as a partial mediator, accounting for 9.5% of the total effect. Additionally, there was a significant indirect effect (effect=0.016, 95% CI 0.001-0.038; P=.03) between sexual sensation seeking and unprotected anal intercourse, but no significant direct effect (effect=–0.005, 95% CI –0.101 to 0.094; P=.93). These results suggest that gay app use plays a complete mediating role in this relationship. In addition, no stratified heterogeneity was found in the samples from both Wuhan and Changsha. Please refer to Multimedia Appendices 3 and 4 for details.

Drawing from the uses and gratifications theory, this study explored the mechanisms that underlie high-risk behaviors among MSM at the behavioral and psychosocial levels. The study’s findings indicated that sexuality traits (including sexual compulsivity and sexual sensation seeking) had a direct impact on high-risk behaviors (including multiple sexual partners and unprotected anal intercourse) among MSM and that the use of gay apps mediated the relationship between sexuality traits and high-risk behaviors.

A major finding was that sexuality traits had a direct influence on high-risk sexual behaviors. Specifically, MSM with higher levels of sexual compulsivity and sexual sensation-seeking were more likely to have multiple sexual partners. This finding was consistent with prior studies showing that sexual compulsivity and sexual sensation-seeking were positively associated with multiple sexual partners, unprotected anal intercourse, sexual acts under drug influence, and infections of HIV and other STIs [34-37]. Individuals with high sexual compulsivity cannot control their sexual thoughts and behaviors, they engage in frequent sexual activities and nonparaphilic activities, such as sex with anonymous partners, which all contribute to an increased risk of high-risk sexual behaviors [38,39]. Moreover, sexual sensation seekers are more likely to engage in sexual-related activities and behaviors that require increasing stimulation and adventure, leading to high-risk sexual behaviors [40]. Our findings suggest that sexuality traits are important risk factors for high-risk sexual behaviors among MSM and should be incorporated into HIV/AIDS prevention and intervention programs.

Another major finding was that gay app use mediated the relationship between sexuality traits and high-risk behaviors. Specifically, MSM with higher levels of sexual compulsivity and sexual sensation-seeking tend to spend more time and energy on gay apps, which in turn, lead to more high-risk sexual behaviors, such as having more sexual partners and engaging in more unprotected anal intercourse. This finding was in line with the uses and gratifications theory proposing that individuals follow their interests to choose media to satisfy their needs. According to the uses and gratifications theory, MSM’s gay app use (communication behavior) is determined by their sexual traits such as sexual compulsivity and sexual sensation-seeking (needs and motives). In order to satisfy their excessive sexual needs, they will engage in a series of high-risk behaviors such as multiple sexual partners and unprotected anal intercourse (the consequences of such behavior) [22]. Our findings suggest that the use of gay apps increases the likelihood of high-risk sexual behaviors among MSM with more prominent sexuality traits. Therefore, it is important to develop intervention programs that are based on gay apps for MSM. Moreover, interventions conducted through specially designed gay apps may also help reduce cultural and communication barriers between MSM and health care providers [41].

Our study showed that approximately 67% (n=271) of MSM used gay apps in the past 6 months to seek potential sexual partners in their local areas. This result confirms previous findings that MSM are increasingly using mobile internet to find sexual partners in the current era [42]. GPS-enabled social networking apps on smartphones allow users to find nearby sexual partners in a flexible and real-time manner and provide more opportunities for anonymous partner-seeking, which also increases the risk of high-risk behaviors among MSM [43,44]. However, other studies have demonstrated that gay apps may increase social support among this population, facilitate the provision of HIV testing services, and reduce related risky behaviors [45]. In recent years, with the advancement in artificial intelligence, some new apps have been developed to promote behavioral changes for HIV control among MSM. For example, Chen et al [46] proposed a chatbot for HIV self-testing to offer personalized, engaging, and needs-based health interventions to MSM based on human-machine interactions in Hongkong, which is expected to be equally, if not more, efficacious as HIV self-testing through live chat app. The use of gay apps can potentially have both favorable and unfavorable health effects, and it is imperative to emphasize the motivations and objectives of MSM using these apps, as they may result in distinct health outcomes.

This study reveals that in the past 6 months, 37% (n=149) of MSM have engaged in unprotected anal intercourse, and 54% (n=218) of MSM have had multiple sexual partners. These findings suggest a high prevalence of high-risk sexual behavior among MSM. The prevalence of unprotected anal intercourse among MSM in our study was higher than the reported 33% in a meta-analysis in China in 2014 [47], indicating an increasing trend of unprotected anal intercourse that warrants attention and action. A possible reason for this rise in prevalence may be related to the changes in the COVID-19 pandemic and related control measures over time. Studies have shown that high-risk sexual behaviors among MS decreased during the COVID-19 pandemic due to strict lockdown and social isolation measures, and then rebounded to the level before the COVID-19 outbreak after the pandemic received initial control in China [48]. The prevalence of multiple sexual partners in our study was consistent with that of previous studies. Unprotected anal intercourse and multiple sexual partners are identified as significant risk factors for HIV and other STIs among MSM [49,50]. It is important to continue monitoring these high-risk sexual behaviors in this population and take appropriate warning measures and contingency plans.

Wave-to-wave balance can be used to assess the balance of the recruited samples in the RDS process, which is crucial for estimating completion and evaluating the effectiveness of RDS [51]. The results showed that the primary indicators of this study eventually achieved balance over time. While the seeds chosen in the early stages of this study were not randomly selected, representative samples were recruited by selecting seeds from different cities and professions and ensuring that the social networks of seeds did not overlap, thereby achieving the goal of probability sampling.

There are several limitations to this study. First, the participants were recruited through the internet, and most of them were young and experienced with social media use. Therefore, this study may not represent MSM who are older or not familiar with social media, resulting in sample selection bias. The sample representativeness may be further limited by the RDS method used in this study, which may introduce statistical inference error due to the specific population, the way of sampling, and the method of estimation, referred to as the “statistical trinity” or the “spatial statistical trinity” [52]. However, RDS remains a valuable tool for researching hidden, hard-to-reach populations, and small subgroups that lack a well-defined sampling frame [53]. Second, all behavioral and psychosocial variables were self-reported, which could introduce social desirability bias. Finally, the study design was cross-sectional, making it difficult to establish causal relationships. In the future, more comprehensive and rigorous longitudinal studies should be conducted among MSM to determine the causal relationships between sexuality traits, gay app use, and the occurrence of high-risk behaviors.

The study shows that high-risk sexual behaviors are common among MSM. Therefore, public health initiatives should prioritize prevention efforts that target these behaviors. Most MSM rely on gay apps to find sexual partners, which, when combined with higher levels of sexual compulsivity and sexual sensation seeking, can increase the likelihood of engaging in high-risk sexual behaviors. Therefore, interventions aimed at reducing these behaviors among MSM should focus on addressing the use of gay apps, while also considering the influence of their sexuality traits on gay app use.