This study was a 2-arm cluster randomized controlled trial; the trial protocol provides more details [25]. The Consolidated Standards of Reporting Trials (CONSORT) checklist can be found in Multimedia Appendix 1. This trial was clustered at the city level. Cities were eligible if they had an MSM sentinel surveillance site and had the capacity to recruit sufficient MSM. We first selected 10 cities from Shandong province, China, including Jinan, Weifang, Zibo, Jining, Liaocheng, Qingdao, Dezhou, Weihai, Binzhou, and Heze. Since Heze did not recruit enough MSM, we included the neighboring city Zaozhuang and regarded it as part of the same cluster (Multimedia Appendix 2).

Participants were eligible if they were born biologically male, were aged 18 years or older, currently lived in and planned to remain living in 1 of the 11 cities for the 12 months after enrollment, were HIV seronegative or had unknown serum status, had anal sex with a man in the past 12 months, had not been tested for HIV in the past 3 months, and provided informed consent. Participants were recruited by sending a link to the questionnaire to potential participants through Blued (a popular social app among Chinese MSM) and community-based organizations (CBOs) in each city. Additionally, eligible participants were invited to recommend at most 5 MSM in their city to participate in this study from their social networks.

The details of randomization are shown in Multimedia Appendix 3. According to the city’s number of people living with HIV in 2018, average GDP in 2017, cumulative number of HIV-positive people from 2012 to 2018, and population at the end of 2017, the cities with the most similar contexts were stratified into the same blocks. Ten clusters were stratified to 5 blocks (2 clusters per block). Then, 2 clusters in each block were randomly allocated 1:1 to either the intervention arm or the control arm with random numbers generated using SAS (version 9.4; SAS Institute) software. Participants and data analysts were blinded to the intervention assignment.

Participants in the intervention arm received a digital, crowdsourced, multilevel intervention developed through a series of crowdsourcing open calls tailored for MSM (Figure 1). Participants in the control arm received conventional intervention by their local center for disease control and prevention (CDC), including outreach activities and HIV testing services. All concomitant care was acceptable, with no restrictions on what participants may seek. The schedule of interventions across arms is shown in Multimedia Appendix 4.

The individual-level intervention included the digital intervention materials and HIV self-testing services delivered with digital tools. The digital intervention materials included 25 images and 4 videos, which were distributed by WeChat. After completing the 6-month follow-up survey, participants in the intervention arm could apply for the digital HIV self-testing services, including the free HIV blood self-test kits, use instructions, and counseling services, through WeChat. HIV self-testing counseling services were provided based on guidelines from the WHO, including pre-and posttest counseling [26]. The details of the counseling services can be found in Multimedia Appendix 4. Before testing, the staff explained the importance of regular testing and provided instructions on use of the kits. Participants who applied for and used the kits were reminded to return photos of the test results. Then, the staff helped them to properly interpret the meaning and implications of the results and provided counseling on safe sexual behaviors and regular testing for participants with negative results and referral services on confirmatory testing and further care for participants with positive results. After the 12-month follow-up, we also provided participants in the control arm with the opportunity to obtain the self-test kits.

For the community-level intervention, we set up 8 peer-moderated WeChat discussion groups in the intervention arm. All participants in the intervention arm received an invitation to join a group and made their own decision on whether to join. Each WeChat group consisted of 1 team member, 1 CBO volunteer, and about 20 participants. The messages about HIV testing and safe sexual behavior came from authoritative facilities (eg, CDCs, CBOs, and hospitals) and were shared through WeChat groups and WeChat moments (a function on WeChat that allows people to share their life with friends) biweekly. In addition, members in the WeChat groups were encouraged to discuss HIV prevention.

After the first intervention, follow-up surveys were conducted every 3 months for 12 months. Questionnaires were built using Sojump Survey Software to collect data. Demographic characteristics were collected at baseline. Sexual behaviors, HIV testing uptake, anticipated HIV stigma, HIV testing social norms, and HIV testing self-efficacy were assessed at baseline and at 4 follow-up visits, at 3, 6, 9, and 12 months. The anticipated HIV stigma, HIV testing social norms, and HIV testing self-efficacy were measured by Likert scales.

The primary outcome was self-reported HIV testing uptake in the previous 3 months measured in each follow-up survey. The secondary outcomes included self-reported facility-based HIV testing uptake, HIV self-testing uptake, condomless sex behavior, social media engagement, anticipated HIV stigma, HIV testing social norms, and HIV testing self-efficacy in each follow-up survey.

Data from July 30, 2021, to April 28, 2022, were analyzed. A sample of 500 MSM from the 10 clusters (50 from each cluster), assuming a 30% dropout rate and an intracluster correlation of .020 (usually between .010 to .030) within 10 clusters, was planned to provide 85% power at 2-sided significance level of α=.05 to detect a 20% difference in the primary outcome. Descriptive analysis was used to describe baseline demographic characteristics for the total sample and by the intervention condition. The cumulative HIV testing uptake, incident testers (the first HIV test during the study period), HIV seroconversion, and HIV self-testing kit use were summarized. We assessed the difference in the proportion of HIV testing between study arms with 2-tailed χ² tests at each follow-up. The intention-to-treat approach was used to examine the cluster-level and individual-level effects of the intervention in the 12-month study period. The cluster-level effect was defined as the difference in the proportion of participants who tested for HIV in the previous 3 months between intervention cities and control cities. The individual-level effect was defined as the difference in the probability of participants having tested for HIV in the past 3 months between the intervention arm and control arm. Specifically, we fitted linear mixed models (LMMs) to investigate the cluster-level effect of the intervention, in which the HIV testing proportion of each cluster across the 4 follow-ups was used as the outcome, intervention status and time were considered as fixed effects, and the sites considered as random effects. To investigate the individual-level effect of the intervention, we used generalized linear mixed models (GLMMs) to evaluate the difference in probability of receiving HIV testing in the intervention arm and in the control arm, in which the test status of every participant across the 4 follow-ups was used as the outcome, intervention status and time were considered as fixed effects, and sites and individual participants were considered as random effects. The estimated intervention effects (risk ratio [RR] for binary outcomes and mean difference for continuous outcomes) are reported with the 95% CI and P value.

For the sensitivity analysis, we conducted a per-protocol analysis including only participants who self-reported that they had viewed the intervention materials or had participated in our WeChat groups. For the prespecified analysis, we tested the interaction effect of the intervention between the individual level and community level. After the 6-month follow-up, we evaluated the effectiveness of the digital materials using data collected from the 3-month and 6-month follow-ups, and we evaluated the effectiveness of providing digital materials plus self-test kits using data collected from the 9-month and 12-month follow-ups. We also evaluated the effect of every component of the overall intervention, including viewing intervention images, viewing intervention videos, receiving HIV self-testing kits from the study team, and participating in the WeChat groups. In addition, we performed a subgroup analysis based on age (<30 years vs ≥30 years). We compared the demographic characteristics between participants who completed the last follow-up survey or seroconverted and participants who were lost to follow-up, and we adjusted the factors that differed between them in the analysis evaluating the intervention effect. We also used multiple imputations by chained equations and 30 imputed data sets to impute missing HIV testing data at each follow-up period and examined the intervention effect. We conducted an intention-to-treat analysis to evaluate the secondary outcomes using GLMMs or LMMs. We performed all analyses with SAS.

This trial was approved by the institutional review board at the School of Public Health, Shandong University (20190210) and was registered in the Chinese Clinical Trial Registry (ChiCTR1900024350). All study participants provided informed consent at the baseline survey.

Participants were recruited from August 6, 2019, to January 25, 2020, and followed until the last participant completed the 12-month follow-up survey on April 8, 2021. A total of 935 participants were enrolled, including 404 in the intervention arm and 531 in the control arm (Figure 2, Multimedia Appendix 5). Most participants were younger than 30 years (n=601, 64.3%), were never married (n=681, 72.8%), had a college degree or higher (n=629, 67.3%), identified as homosexual (n=668, 71.4%), and disclosed their sexual orientation to others (n=606, 64.8%). One-third of participants had engaged in condomless sex (n=314, 33.6%) in the past 3 months, and 785 participants (84%) had an HIV testing history (Table 1). After the 9-month follow-up, 10 participants HIV seroconverted (Multimedia Appendix 6). Among the 925 HIV-negative or unknown participants after the 9-month follow-up, 262 (28.3%) did not complete the last follow-up survey (Multimedia Appendix 7). The participants who completed the last follow-up survey or seroconverted differed in educational attainment and sexual orientation compared to participants who were lost to follow-up (Multimedia Appendix 8).

A total of 751 participants (751/935, 80.3%) completed at least one follow-up survey and were analyzed to evaluate the effect of the intervention (Multimedia Appendix 9). Among these 751 participants, 451 (60%) reported that they were tested for HIV at least once during the follow-up period (Multimedia Appendix 10). Incident HIV testers across arms are shown in Multimedia Appendix 11. Overall, the proportion of participants testing for HIV in the past 3 months at the 3-, 6-, 9-, and 12-month follow-ups was higher in the intervention arm (139/279, 49.8%; 148/266, 55.6%; 189/263, 71.9%; and 171/266, 64.3%, respectively) than the control arm (183/418, 43.8%; 178/408, 43.6%; 206/403, 51.1%; and 182/397, 48.4%, respectively), with statistically significant differences at the 6-, 9-, and 12-month follow-ups (Figure 3, Multimedia Appendix 12).

For the measure of the cluster-level effect of the intervention, the proportion of individuals receiving an HIV test within a city in the intervention arm was 11.62% (95% CI 0.74%-22.50%; P=.04) higher than in the control arm (Figure 4). Analysis after multiple imputation produced a similar result. In the per-protocol analysis, the estimated effect size was larger (mean difference 14.69%, 95% CI 4.04%-25.33%; P=.01). For the individual-level effect of the intervention, results from the intention-to-treat analysis showed that participants in the intervention arm had 69% higher odds for testing for HIV in the previous 3 months compared with control participants, but the result was not statistically significant (RR 1.69, 95% CI 0.87-3.27; P=.11). Per-protocol analysis and sensitivity analysis adjusting for education level and sexual orientation showed similar results. An interaction was found between the individual-level intervention and the community-level intervention, but this was not statistically significant (interaction test P=.08) (Figure 5).

In addition, the estimated effect of the provision of digital materials alone was not statistically significant (mean difference 5.58%, 95% CI –8.51% to 19.68%; P=.46; RR 1.27, 95% CI 0.59-2.70; P=.50). After the provision of digital materials plus HIV self-testing kits, the probability that an individual had tested for HIV in the previous 3 months increased (mean difference 17.66%, 95% CI 7.36%-27.97%; P=.008; RR 2.18, 95% CI 1.23-3.87; P=.02) (Figure 4). As for the components of intervention, viewing intervention videos (RR 1.49, 95% CI 1.08-2.07; P=.02) and participating in peer-moderated discussion WeChat groups (RR 3.18, 95% CI 1.90-5.31; P<.001) increased the HIV testing uptake. A total of 287 HIV self-testing kits were applied and used by participants during the study period (Multimedia Appendix 13).

The subgroup analysis showed that the effect of the intervention was similar among MSM who were aged 18 to 30 years (RR 1.44, 95% CI 0.62-3.36) compared to MSM who were aged more than 30 years (RR 2.29, 95% CI 0.93-5.67; interaction test P=.20) (Figure 5).

The digital, crowdsourced, multilevel intervention did not improve the secondary outcomes (Multimedia Appendix 14).

Despite significant achievements in HIV prevention worldwide, HIV testing uptake among MSM is still suboptimal in China. It is crucial to develop effective interventions to improve HIV testing uptake among MSM. In this cluster randomized controlled trial among Chinese MSM, the digital, crowdsourced, multilevel intervention was able to effectively enlarge HIV testing coverage, with an 11.62% increase in the proportion of HIV testing in the intervention arm. Our study extends previous research on HIV prevention interventions by using crowdsourced methods to develop tailored interventions, focusing on digital methods for intervention promotion and dissemination and addressing both individual- and community-level factors related to HIV testing.

Our findings are consistent with a previous cluster randomized controlled trial among MSM in China, which found that a crowdsourced intervention increased the number of individuals receiving an HIV test by 8.9% [23]. In addition to its use of a crowdsourced approach, this trial extends previous studies by implementing individual- and community-level interventions through digital platforms simultaneously. The fact that Chinese MSM widely use social media, mobile phone apps, and the internet to find sexual partners and search for HIV prevention information provides a basis for digital interventions. Digital tools could allow MSM to access HIV information and services without disclosing their sexual orientation, which might decrease stigma when seeking services and improve community engagement. Our findings confirm previous studies demonstrating the feasibility and effectiveness of digital-based intervention in improving HIV testing among MSM [8,20]. This trial could guide digital health intervention organization and promotion in HIV prevention programs and even other health care programs.

Most previous HIV prevention programs implemented interventions addressing factors related to HIV testing at a single level [27,28], and few studies evaluated the effect of multilevel intervention in improving HIV testing. Although there is scarce evidence on the effects of multilevel intervention for HIV testing among MSM, it has been shown to be effective in reducing sexual risk behaviors and increasing access to antiretroviral treatment among other key populations [20,29]. In this study, according to the socioecological theory [19], we developed a multilevel intervention addressing individual- and community-level factors related to HIV testing. The individual-level intervention included digital educational materials (eg, images and videos) and free HIV self-testing kits with instructions for how to use them and interpret the test results provided through WeChat, which could increase knowledge, training, and support for HIV testing services from the individual perspective. The community-level intervention included peer-led, moderated discussion and communication through WeChat groups, as well as messages shared through WeChat moments, through which MSM could obtain peer education and community support. Our findings suggest that multilevel intervention is an effective tool to expand HIV testing among MSM, highlighting the need for more multilevel interventions to promote HIV testing or other health services among key populations.

We also found that providing HIV self-testing services through digital tools effectively improved the probability of individuals receiving an HIV test. HIV self-testing services are an effective and cost-saving tool for promoting HIV testing among MSM [30,31]. Previous studies also showed the feasibility of digital HIV self-testing services and the potential effect on promoting early diagnosis in MSM [32]. Recently, the WHO has called for evidence in digital innovations to enhance HIV self-testing coverage [33]. This study provides evidence on digital HIV self-testing services to improve HIV testing among MSM in China. The digital approach might be a useful tool to make HIV self-services more available, convenient, and discreet.

Our study is subject to several limitations. First, the data on recent HIV testing uptake and results were self-reported by participants, which may cause information bias or recall bias. Second, sampling using digital approaches cannot reach people who do not use digital tools frequently. However, offline recruitment and peer recommendation were also used as auxiliary methods, which might alleviate this problem to some extent. Third, 28.3% (n=262) of participants were lost at the last follow-up. We performed a sensitivity analysis adjusting for factors that differed between participants lost to last follow-up and those who finished the last follow-up or seroconverted, and we used multiple imputation to examine the intervention effect. We found the results did not change. Fourth, participants in the intervention arm and the control arm came from different cities in Shandong province. However, the study cities were pair-matched according to their populations, socioeconomic contexts, and HIV burden, which could balance the participants in the 2 arms. Fifth, participants from the intervention and control cities might communicate with each other, resulting in contamination. We conducted per-protocol analyses to measure whether our results were affected by contamination and found the results did not change.

This digital, crowdsourced, multilevel intervention is an effective approach to improve HIV testing uptake among MSM. Our findings highlight that such interventions should be made more widely available for HIV prevention and other public health issues.