This study was designed, conducted, and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for cross-sectional studies (Checklist 1) [26]. The survey was implemented in Hangzhou, Zhejiang Province, China. Participants were recruited via digital outreach of the Sunshine Coast Public Welfare Center (a community-based local nongovernmental organization [NGO] providing HIV-related education and services to men who have sex with men) between January 2024 and August 2024. The inclusion criteria for men who have sex with men were male at birth, aged 16 years or older, and reporting a history of anal or oral sex with a man. The link was disseminated on the NGO’s official WeChat account, community WeChat groups, and WeChat Mini Program, and the participants completed a structured, anonymous, self-administered questionnaire online. No a priori sample size calculation was performed because the study used community-based recruitment and aimed to characterize latent subgroups rather than estimate a single parameter.

The study protocol and consent procedure were approved by the Medical Ethics Committee of the School of Public Health, Zhejiang University (project ZGL202306-9). Electronic informed consent was obtained at the beginning of the online survey; participants proceeded only after checking an “I agree” box. Survey responses were anonymous; no personal identifiers were collected. Data were stored on a password-protected server accessible only to study staff. No compensation was provided. No images containing identifiable participant information were included.

Descriptive statistics were used to describe the sociodemographic characteristics and sexual behavior profiles of the incorporated men who have sex with men participants. Latent class analysis (LCA) was conducted using the poLCA package (1.6.0.1) in R (version 4.3.3; R Foundation for Statistical Computing) to identify distinct behavioral and cognitive subgroups based on the main factors influencing the use of PrEP and PEP (ie, sexual behaviors and cognition of HIV preventive measures). LCA models with 2 to 5 latent classes were estimated, and model fit was assessed using multiple indices, including the Akaike information criterion (AIC), Bayesian information criterion (BIC), adjusted BIC (aBIC), entropy, Vuong-Lo-Mendell-Rubin likelihood ratio test (VLMR-LRT), bootstrapped likelihood ratio test (BLRT) [30], and average latent class posterior probability (ALCPP) [31]. Lower AIC, BIC, and aBIC values, and higher entropy values, indicate better model fit. Entropy values of more than 0.60 were considered indicative of acceptable model fit [32], while ALCPP values of more than 0.70 indicated reliable classification [31]. The VLMR-LRT and BLRT were used to compare the relative fit of K-class versus (K–1)-class models. Prior research suggests that among fit indices, BLRT exhibits the highest performance, followed by BIC and aBIC [33]. Additionally, selection of the optimal model also considered the interpretability, meaningfulness, and proportional size of latent classes [34]. In case of conflicting fit indices, we prioritized BLRT and BIC or aBIC for statistical fit, and then considered entropy, ALCPP, minimum class size (>5% a priori), and interpretability.

Following the identification of latent sexual behavior subgroups via LCA, further analyses were conducted to examine differences across classes. Multinomial logistic regression models were used to assess associations between sociodemographic variables and latent sexual behavior class membership. In addition, latent classes were used as the main independent variable in binary logistic regression models to explore their association with the PrEP and PEP uptake, while adjusting for the same set of sociodemographic covariates. All statistical analyses were conducted using SPSS (Statistical Package for Social Sciences; version 29.0; IBM Corp). A 2-sided P value <.05 was considered statistically significant. Results were reported as odds ratios (ORs) with corresponding 95% CIs.

The Sunshine Coast Public Welfare Center’s online platform serves approximately 10,700 men who have sex with men residing in Hangzhou. A total of 3267 participants were recruited (from January 2024 to August 2024), yielding an estimated response rate of 30.52%. The majority of participants were young, with 58.4% aged 16 to 29 years (Table 1). Most participants (n=2179, 66.7%) had resided in the city for more than 2 years. Educational attainment was generally high, with 46.8% (n=1529) holding a bachelor’s degree and 12.5% (n=407) attaining a master’s degree or higher. The income distribution suggested a moderate socioeconomic profile, with 45.1% (n=1473) earning between 5000 and 10,000 RMB (1 RMB=US $0.14) per month, and 29.2% (n=1043) reporting monthly earnings above 10,000 RMB.

A considerable proportion (68.1%) self-identified as homosexual, whereas 31.9% reported being bisexual. Nearly half of the respondents (44.5%) reported having one male sexual partner in the past 3 months, while 28.7% had 2 to 4 partners, and 3.9% reported 5 or more. Risk-related sexual behaviors were prominent, with 32% engaging in unprotected intercourse and 2.7% reporting engagement in commercial sex. Drug use during sexual activity was reported by 4.8% of participants. Only 48.5% of respondents reported being fully informed of their partners’ status. Only 2% of men who have sex with men reported being diagnosed with STIs, with 1% with syphilis. Most participants (75.6%) had received HIV testing within the previous six months.

Despite high exposure to HIV-related education campaigns (76.4%), only 35.7% demonstrated awareness of PrEP and PEP, with only 7.8% having used PrEP and 4.7% having used PEP.

LCA was conducted to compare models comprising 2 to 5 latent classes (Table 2). The ALCPP declined from 0.892 in the 2-class model to 0.757 in the 5-class model, and entropy values were consistently low—particularly for the 3-class (0.591) and 5-class (0.619) models—indicating reduced classification precision with increasing numbers of classes. Both the VLMR-LRT and the BLRT yielded statistically significant results (P<.001), indicating improved model fit with the inclusion of additional classes. Specifically, the 2-class model exhibited the most distinct classification structure but demonstrated poor overall model fit, whereas the 3-class model showed improved fit but suffered from the lowest entropy value. The 4-class model provided the optimal tradeoff among model fit (as reflected by lower AIC and BIC values), entropy (0.630), and classification reliability (ALCPP=0.787), and was therefore selected as the final solution. The 5-class model demonstrated signs of overfitting and diminished clarity in latent class differentiation.

Accordingly, 4 distinct latent classes (ie, low-risk and well-informed class, high-risk and well-informed class, moderate-risk and poorly informed class, and high-risk and poorly informed class) were identified, characterized by varying sexual behaviors and cognition of HIV preventive measures (Table 3). The low-risk, well-informed group exhibited the most protective behavioral profile: only 2.4% reported having 2 or more partners in the past 3 months, no participants engaged in condomless sex, and 74.9% had undergone HIV testing within that period. Awareness of PrEP and PEP was moderate (42.3%), and a substantial proportion (87.4%) reported recent exposure to HIV-related public health messaging. The high-risk, well-informed group reported elevated sexual risk behaviors—38.1% had two or more partners and 67.2% engaged in condomless sex—yet demonstrated strong prevention engagement: 77.3% were fully aware of PrEP and PEP, 88.5% had undergone recent HIV testing, and 97% knew their partners’ HIV status. The moderate-risk, poorly informed group displayed intermediate levels of behavioral risk—24.1% reported multiple partners and 23.5% reported condomless sex—but exhibited the lowest levels of prevention knowledge: 99.8% lacked full awareness of PrEP or PEP, 80.4% had not encountered HIV-related messaging, and only 48.5% had recently undergone testing. The high-risk, poorly informed group represented the most vulnerable subgroup: 72% reported multiple partners, 55.9% engaged in condomless sex, and 76.3% lacked comprehensive awareness of PrEP and PEP. Despite a relatively high rate of HIV testing (85%), prevention literacy remained consistently low across this group.

As shown in Table 4, men who have sex with men aged 16 to 29 years had significantly lower odds of belonging to the moderate-risk, poorly informed class compared with those aged 50 years or older (OR 0.452, 95% CI 0.255-0.798; P=.006). Household registration characteristics were significantly associated with latent class membership. Participants with registered residence in Zhejiang province had lower odds of being in the moderate-risk, poorly informed class (OR 0.745, 95% CI 0.596-0.931; P=.01) and the high-risk, poorly informed class (OR 0.759, 95% CI 0.631-0.912; P=.003) compared with those from other provinces. Similarly, individuals from urban areas had reduced odds of being classified into the high-risk, poorly informed class compared with those from rural areas (OR 0.808, 95% CI 0.671-0.972; P=.02). Full-time employment in the public sector was associated with lower odds of membership in the moderate-risk, poorly informed class compared with other occupational categories (OR 0.559, 95% CI 0.317-0.987; P=.045). Educational attainment was strongly associated with class membership. Compared with those holding a master’s degree or above, participants with primary school or below (OR 3.496, 95% CI 1.268-9.639; P=0.02), junior high school (OR 2.482, 95% CI 1.380-4.463; P=0.002), senior high school (OR 2.279, 95% CI 1.379-3.766; P=0.001), vocational or technical education (OR 2.922, 95% CI 1.551-5.504; P<0.001), and associate degree (OR 1.880, 95% CI 1.253-2.822; P=0.002) had higher odds of belonging to the moderate-risk, poorly informed class. Monthly income was inversely associated with membership in the high-risk, poorly informed class. Compared with those earning more than 10,000 RMB per month, participants earning less than 3000 RMB (OR 0.620, 95% CI 0.416-0.923; P=.02), 3000 to 4999 RMB (OR 0.726, 95% CI 0.529-0.996; P=.047), and 5000 to 10,000 RMB (OR 0.682, 95% CI 0.550-0.847; P<.001) had lower odds of belonging to the high-risk, poorly informed class.

Table 5 revealed substantial disparities in the uptake of PrEP and PEP across distinct HIV risk profiles. Overall, PrEP use was 7.81% (255/3267; 95% CI 6.93%-8.78%) and PEP use was 4.74% (155/3267; 95% CI 4.07%-5.53%). Compared with the low-risk, well-informed reference group (PrEP: 37/1186, 3.12%; PEP: 35/1186, 2.95%), individuals in the high-risk, well-informed class reported significantly higher use of PrEP (76/564, 13.48%; OR 4.890, 95% CI 3.224-7.418, P<.001) and PEP (53/564, 9.40%; OR 3.035, 95% CI 1.930-4.771, P<.001). Similarly, participants in the high-risk, poorly informed class also showed higher PrEP uptake (129/952, 13.55%; OR 5.419, 95% CI 3.690-7.959, P<.001) and higher PEP use (61/952, 6.41%; OR 2.574, 95% CI 1.666-3.978, P<.001) compared with the reference group. In contrast, the moderate-risk, poorly informed group showed no significant difference in PrEP use (13/565, 2.30%; OR 0.891, 95% CI 0.463-1.713, P=.73) and had significantly lower PEP uptake (6/565, 1.06%; OR 0.385, 95% CI 0.158-0.939, P=.04) relative to the reference group.

In contrast, the moderate-risk, poorly informed group showed no significant difference in PrEP use (2.30%; OR 0.891, 95% CI 0.463‐1.713; P=.729) and had significantly lower PEP uptake (1.06%; OR 0.385, 95% CI 0.158‐0.939; P=.036) compared to the reference group.

This study identified 4 latent subgroups among men who have sex with men in Hangzhou based on sexual behaviors and prevention-related awareness (low-risk, well-informed; high-risk, well-informed; moderate-risk, poorly informed; and high-risk, poorly informed groups). PrEP and PEP use were primarily concentrated among the 2 high-risk groups, whereas the moderate-risk but poorly informed group exhibited particularly low uptake (PrEP: 2.30%; PEP: 1.06%), indicating a substantial mismatch between behavioral exposure and coverage of HIV prevention services.

Sociodemographic factors play a pivotal structural role in differentiating men who have sex with men subpopulations according to their HIV risk profiles. Education and income emerged as the most consistent predictors among the high-risk and moderate-risk but poorly informed groups. Lower educational attainment—particularly junior high school and vocational training—was significantly associated with membership in the high-risk, poorly informed group, while individuals with unstable residence and nonlocal household registration were more likely to fall into the moderate-risk, poorly informed group. These findings align with existing literature indicating that lower socioeconomic status is strongly associated with reduced access to HIV-related knowledge, rather than knowledge-use behaviors. For example, research from Brazil and China suggests that men who have sex with men with lower income and education levels tend to have lower HIV knowledge scores and weaker risk perception [35,36]. This phenomenon may be explained by structural factors such as informational exclusion, institutional discrimination, and unequal access to community-based education programs. Studies from both high- and middle-income countries also indicate that health communication tends to favor those with higher educational attainment, reinforcing knowledge gaps within socioeconomically marginalized men who have sex with men subgroups [37,38]. These findings highlight the need to design differentiated health promotion strategies that address the cognitive disadvantage created by broader structural inequities.

Both high-risk subgroups—irrespective of their cognition of HIV preventive measures—exhibited significantly elevated rates of PrEP and PEP use. This aligns with well-documented associations between high-risk sexual behaviors and the uptake of biomedical prevention tools observed among men who have sex with men globally, including in China [39,40]. Notably, PrEP usage among the high-risk, poorly informed group was comparable to that of their well-informed counterparts, despite markedly lower levels of HIV-related knowledge. This pattern suggests that PrEP uptake in this subgroup may be less influenced by comprehensive prevention literacy and more reflective of reactive, event-driven decision-making—for example, seeking PEP following perceived exposure, or initiating PrEP due to a partner’s HIV status or prior testing experiences. Similar behavioral dynamics have been reported in other men who have sex with men populations, where PrEP use is often shaped more by perceived vulnerability and episodic risk than by an in-depth understanding of biomedical interventions. For instance, a study in Belgium found that men who have sex with men modulated PrEP adherence based on fluctuating risk exposure, indicating a largely reactive usage model [41]. Likewise, research in India reported that men who have sex with men with higher perceived HIV risk were more willing to use PrEP, regardless of their actual knowledge of the intervention [42]. However, the rate of PEP use in the high-risk, poorly informed group (6.41%) was substantially lower than in the high-risk, well-informed group (9.40%). This discrepancy likely reflects the time-sensitive nature of PEP initiation—typically within 72 hours of exposure—which demands a higher degree of health awareness. The comparatively lower uptake in the low-literacy group underscores the urgent need for targeted health education to bridge this critical knowledge-to-action gap.

Interestingly, no significant difference in PrEP or PEP uptake was observed between the high-risk, well-informed and high-risk, poorly informed groups, challenging the assumption that cognition of HIV prevention measures alone directly translates into preventive behavior. Several factors may account for this disconnect. First, both groups may encounter similar structural barriers, such as limited access, financial constraints, and stigma [43-45], that hinder service use regardless of knowledge levels. Second, general awareness of HIV prevention may not equate to a heightened perception of personal risk, thereby weakening the motivation to initiate PrEP or PEP [46,47]. Third, the knowledge indicators used in the LCA may capture general informational exposure rather than specific, service-oriented literacy or navigational capacity. Fourth, men who have sex with men populations often demonstrate event-driven, reactive patterns of biomedical prevention use, such as initiating PEP following perceived exposure, rather than sustained, proactive engagement [48]. Finally, social norms and peer influences may exert a stronger effect than individual knowledge [43,45,49], especially in the first context.

The moderate-risk, poorly informed group emerged as a notably underserved population in this study. Despite reporting non-negligible behavioral risks—such as a 24.1% prevalence of multiple concurrent sexual partnerships—this group exhibited the lowest uptake of PrEP (2.30%) and PEP (1.06%). Strikingly, 99.8% of individuals in this class lacked adequate knowledge of PrEP or PEP (Table 3), revealing a profound gap in biomedical prevention coverage for what may be termed an “invisible risk group.” This misalignment between behavioral vulnerability and prevention engagement echoes findings from prior research in China. For instance, Wang et al [9] reported rising PEP usage in Shenzhen—from 3.92% in 2018 to 10.29% in 2020—yet noted that such uptake remained concentrated among high-risk populations, with limited reach among moderately at-risk individuals with low prevention literacy [50]. Importantly, the disconnect between moderate-risk individuals and biomedical prevention tools reflects not only gaps in personal awareness but also deeper structural and informational asymmetries. As Lei [51] highlights in a qualitative study, knowledge dissemination about PrEP and PEP within men who have sex with men communities in China is highly networked and stratified. Information circulates primarily through specific digital platforms and community-based networks, often excluding individuals from rural backgrounds, those with limited education, or recent migrants [51]. These mechanisms reinforce their “invisibility” in existing prevention infrastructures, rendering them unable to access timely information or interventions despite engaging in risk behaviors. This phenomenon is supported by quantitative evidence: studies in China have shown that low health literacy, rural or nonlocal household registration, recent migration, and limited access to public sector resources are independently associated with restricted access to HIV prevention services [52-54]. Traditional intervention strategies, which often prioritize those at the highest observable risk, may inadvertently neglect populations with moderate but sustained behavioral risk. For this structurally disconnected group, a multilevel intervention framework is urgently needed. This includes peer-led community education, embedding PrEP or PEP counseling into routine HIV testing, and implementing culturally responsive outreach for rural and migrant populations [52,55]. Enhancing both institutional design and the communication ecology will be essential for achieving equitable, targeted service delivery across diverse men who have sex with men subpopulations.

This study has several limitations. First, its cross-sectional design limits the ability to infer causal relationships between HIV risk patterns and PrEP or PEP uptake. Second, self-reported data may be subject to recall bias or social desirability bias, potentially affecting the accuracy of behavioral and prevention-related indicators. Third, participant recruitment was conducted exclusively online through a NGO’s digital platforms, which may have excluded men who have sex with men who are less digitally connected or more structurally marginalized. Fourth, PrEP and PEP usage were measured as binary variables, without capturing duration, adherence, or regimen type, thereby restricting the interpretability of biomedical prevention engagement. Finally, this study did not collect or report participants’ HIV serological results, and thus the HIV status of respondents could not be objectively verified, which may introduce uncertainty regarding the actual infection status. Despite these limitations, the identification of distinct behavioral-knowledge typologies offers meaningful insights for refining HIV prevention strategies.

This study identified that the 4 distinct subgroups of HIV-related risk profiles among men who have sex with men and PrEP or PEP use were concentrated among high-risk individuals. By integrating behavioral and prevention-cognition indicators through LCA, the study revealed an underserved subgroup, men who have sex with men with moderate behavioral risk but very low prevention literacy, which traditional behavior–only frameworks fail to detect. These findings demonstrate that knowledge alone is insufficient to drive PrEP or PEP uptake and highlight the need for differentiated, equity-focused strategies that pair education with improved access and supportive service environments. Targeted community-based education, embedding PrEP or PEP counseling into low-threshold HIV testing, and outreach to men who have sex with men with lower socioeconomic status emerge as practical priorities to close prevention gaps. By offering a multidimensional, person-centered perspective, this study contributes novel evidence to inform real-world HIV prevention planning in China and other middle-income settings facing similar structural challenges.