We designed E-PrEP based on formative work conducted by our team [51], in which peers used multiple social media platforms to promote HIV testing using creative messaging and led other social media–based HIV interventions [52,53]. Community-based participatory research methods [47-49,54] guided the development of E-PrEP, with input from a group of 6 YBLGBM intervention development partners experienced in HIV and PrEP outreach in New York City, whom we recruited through an existing HIV outreach initiative for YBLGBM. We selected Facebook and Instagram as the platforms for the intervention, as these are the 2 general social media sites most frequently used by the target population. We did not restrict the intervention to a single platform to provide flexibility, as not all YBLGBM use both sites equally. We developed all contents for both E-PrEP and the control condition (E-Health), and created a content posting and activity guide for all 6 weeks (see Multimedia Appendix 1 for additional details). We standardized E-PrEP in its mode of delivery, types of digital media and contents, and sequence of topics posted and discussed, but during the intervention each peer leader tailored the exact language of each post based on their individual communication style. All materials, including assessments, were mobile phone optimized.

We selected potentially modifiable targets to inform message content based on the DOI and IMB models. We also incorporated findings from a qualitative study of local YBLGBM [55], a systematic review of barriers to and facilitators of PrEP, which included a systematic content analysis of online posts about PrEP by men who have sex with men [56], and input from our intervention development partners. Messages were presented using digital media (eg, text, pictures, infographics, and video clips) and posted online based on findings from prior Web-based interventions[57,58] and on social media marketing principles [59]. We designed the content to engage participants in online discussions about HIV prevention, PrEP, and health care [57]. To retain YBLGBM and prevent intervention fatigue [7,60], peer leaders and participants were encouraged to post other items of interest, regardless of their relevancy to HIV or PrEP (eg, pop-culture posts, pictures from recent events, or discussion of current news) [10,61].

We used an iterative and participatory approach to inform the contents of each E-PrEP online post. First, the research team and our intervention development partners created a digital media library of PrEP educational contents by searching social media (Facebook, Instagram, Twitter, and YouTube) and websites with publicly posted and shareable information about PrEP and accessing health care in the United States. Next, we elicited feedback from the intervention development partners about the following attributes for each media item: their overall thoughts, the comprehensibility, aesthetic appeal, engagability, and informativeness of the item, whether the intervention development partners would actually share the media item, and whether they thought their YBLGBM friends would be likely to view or click on the media item. Then, we took the highest-rated items and mapped them onto a matrix including DOI and IMB domains, as well as barriers to and facilitators of PrEP, to ensure that all relevant topics were covered.

Table 1 [62-70] lists examples of E-PrEP intervention contents. The content, formats, and mode and timing of delivery of E-PrEP were informed by the intervention development partners during the intervention development phase, with ongoing input from the peer leaders during intervention implementation. Existing local resources, LGBTQ-friendly clinics, and an existing LGBTQ patient navigator in New York City were highlighted as part of the online content to ensure that people knew how to access care if desired. We only listed clinical resources that accepted new patients, accepted Medicaid or uninsured patients, and were already prescribing PrEP.

E-Health focused on a broad range of health topics prioritized by the peer leaders assigned to this arm, but did not include any contents about HIV or PrEP. The peer leaders randomly assigned to the E-Health control condition were informed at the first meeting that they would be creating a 6-week social media campaign focusing on health issues they viewed as a priority for YBLGBM within their communities. They chose to cover the following topics: depression, anxiety, suicide, intimate partner violence, drug use, social acceptance, and awareness of sexually transmitted infections (excluding HIV). We designed the E-Health timeline to match the E-PrEP intervention timeline for both time and day of posts and frequency of posts. Similarly to the development of E-PrEP, peer leaders compiled publicly available digital media contents addressing the selected health topics, and then as a group finalized materials to be posted during the online campaign. As with E-PrEP, standardized E-Health contents were posted by peer leaders, framed using their own words. At the end of the trial, peer leaders and participants randomly assigned to E-Health were exposed to all E-PrEP contents.

We established E-PrEP (intervention) and E-Health (control) private online communities (either a private Facebook group or a private Instagram feed) for each peer leader. We also formed 2 separate private Facebook groups (1 for each condition) for peer leaders, led by a peer facilitator. In these groups, peer leaders could share additional materials, troubleshoot potential issues, and communicate with other peer leaders, the peer facilitator, and a research assistant assigned to that condition. We used third-party content management software (Buffer [71]) to facilitate content posting so that all posts could be prescheduled by peer leaders for the 6-week intervention duration and published at the same day and time in both conditions.

To pretest and refine E-PrEP, the intervention development partners each recruited 2 YBLGBM participants from their networks. These participants completed an online consent process and baseline survey, and then received an additional link to join an unlisted private Facebook group where all E-PrEP contents were posted over a 6-week period. The pretest participants provided feedback about contents, process, and acceptability, and also provided suggestions for improvement to all aspects of the intervention through three ways: (1) ongoing feedback elicited by the intervention development partners on the posted E-PrEP contents using open-ended questions in the Facebook group, (2) a brief online acceptability and usability survey at the end of the 6-week period, and (3) an in-person focus group with the intervention development partners and 8 of the pretest participants. Pretest participants received a US $25 debit card as an incentive for their participation. Based on feedback from pretest participants, we refined the E-PrEP intervention by modifying post contents (eg, replacing contents that elicited negative reactions or were considered stigmatizing).

We recruited 10 YBLGBM peer leaders through advertisements via emails to local youth and LGBTQ-focused community organizations, word-of-mouth through key informants in local YBLGBM communities, and targeted advertisements on Facebook and Instagram. All advertisements directed potential peer leaders to a brief online screening survey. Inclusion criteria for peer leaders were (1) having more than 500 online friends or followers on Facebook or Instagram, (2) using either Facebook or Instagram daily, (3) having positive attitudes about PrEP, (4) residing in New York City, (5) identifying as black or Latino, (6) being sexually active with men in the past year, (7) being fluent in English, (8) being between 18 and 34 years of age, (9) being willing to and feeling comfortable posting and discussing health issues (including sexual health and HIV) with friends on Facebook or Instagram, (10) being able to commit to meeting weekly for 12 weeks for training and intervention implementation, and (11) being able to provide consent. A study coordinator telephoned individuals meeting eligibility to provide further information about the study (eg, that this was a research study and they were study participants as well) and to assess interest and availability. After recruiting peer leaders over a 5-week period, we randomly assigned them to the 2 arms. Peer leaders were blinded to their study condition and were informed that their participation was to help refine and launch an online health promotion campaign for YBLGBM in their online social networks. Each arm was facilitated by a peer facilitator (who also identified as a black or Latinx GBM and was experienced in group facilitation for HIV prevention with YBLGBM in New York City) and supported by a research staff member.

We used a participatory process to refine and implement the study. Peer leaders met weekly over a 12-week period. Meetings took place for 3 hours in the evenings at a Wi-Fi–equipped community health center that was easily accessible by public transportation. As Figure 2 displays, the first 6 weeks were dedicated to training and intervention refinement activities (phase 1) and the second 6 weeks were the active intervention delivery period (phase 2).

At the first training session (phase 1), peer leaders randomly assigned to E-PrEP were informed that the overall goal of the campaign was to help prevent HIV in our communities by disseminating accurate information about PrEP to their networks and help link individuals to primary care or PrEP care through an online outreach campaign. During phase 1, peer leaders received training in online recruitment, HIV prevention outreach, PrEP, social media–based outreach and engagement, Buffer software (for scheduling the online posts), and research ethics. As part of the training and intervention refinement, peer leaders reviewed the previously selected and prepiloted E-PrEP digital media materials and made changes or additions as they deemed necessary. During the intervention period (phase 2), peer leaders posted content and provided ongoing feedback to the investigators (see Intervention Procedures, below). In addition to eliciting ongoing feedback during the intervention period (phase 2), we obtained feedback on peer leaders’ experiences participating in the study after study completion using a structured Web-based survey and a semistructured focus group discussion. We collected this information to guide future online outreach practices and capture any relevant issues that emerged during the intervention period but was not captured in our weekly discussions.

Peer leaders went through a community research ethics training adapted from a World Health Organization training for lay community members [72]. The training covered topics in the history of research ethics, confidentiality, vulnerable participants, ethical recruitment practices, protecting against risks, minimizing risks, and personal safety and conduct. This training had previously been used with community partners for whom the Collaborative Institutional Training Initiative course [73] was inaccessible due to literacy or language barriers. We also had ongoing discussions throughout the project period in online privacy and security issues. Peer leaders were informed that if there were any clinical questions or questions they did not know the answers to posted, they should send a message with a copy of the question to their peer facilitator and research staff associated with their group, who would obtain the answers to the questions.

Peer leaders were considered to be research participants and were compensated for their time and incentivized to actively participate, by receiving increasing amounts of incentive for each week they attended the weekly in-person meetings (up to US $485 via a debit card) over the course of phases 1 and 2. Peer leaders could participate in makeup sessions if needed by meeting one-on-one with the research staff member assigned to their arm. Because peer leaders prescheduled all posts using the Buffer scheduling software, intervention delivery did not have to rely on peer leaders conducting core intervention activities (ie, daily posts in the private online group) outside of the weekly meetings. Peer leaders were encouraged, though, to foster discussions on each post by posing questions to their groups or sharing their thoughts and eliciting feedback from group members.

Peer leaders, after completing training, recruited network study participants via their existing online social networks to complete an online eligibility screener and baseline survey, using individualized links (Figure 2). Potential network study participants were directed to a Web-based informed consent form, followed by a screener that continued seamlessly into the baseline survey (for eligible network study participants) (using Qualtrics survey software) or an exit page indicating ineligibility. After completion of the baseline survey, network study participants were directed to join the private online intervention group of the peer leader who had recruited them (either a Facebook private group or a private Instagram feed as selected by the peer leader) and were considered to be enrolled after joining the group. Network study participants received an incentive of US $20 after joining the online group.

Inclusion criteria for network study participants were (1) identifying as or assigned male at birth, (2) self-identifying as black or Latino, (3) being 18 to 29 years of age, (4) being fluent in English, (5) being HIV uninfected or unknown by self-report, (6) residing in New York City, (7) having a Facebook or Instagram account, (8) having had insertive or receptive anal sex with a male partner in the past 12 months, and (9) having had 1 of the following in the past 12 months: condomless anal intercourse, anal sex with more than 3 men, bacterial sexually transmitted infection diagnosis (syphilis, gonorrhea, or chlamydia), or a sex partner who was at least 10 years older [74].

We used a cluster-randomized design [34] for several reasons. As we are testing a peer-based social network intervention, keeping study participants clustered with the peer leaders who recruited them maintains ecologic validity and approximates real-world circumstances. This approach also helps minimize contamination within peer networks and is consistent with the DOI model, which highlights social connections in the diffusion process [41,62]. The main drawback to this design is intracluster correlation, such that we cannot assume independence among participants within peer groups. To address this, we will conduct a series of sensitivity analyses (see Analytic Plan, below).

To address potentially fraudulent or duplicate responses, we excluded individuals from the same internet protocol address, recognizing that this approach may inadvertently exclude individuals who simply were sharing a Wi-Fi network. Additionally, we asked for Facebook and Instagram usernames to verify participant identity [75-80]. A research assistant reviewed and approved all requests for entry to the private groups by (1) verifying that the Facebook or Instagram account was already connected to the peer leader who recruited them, (2) asking participants to respond to a private direct message from the research assistant, and (3) insuring that participants had more than 50 friends or followers. We used this last criterion to avoid potentially fraudulent participants who may have developed a new social media account just for the intervention and would thus be unlikely to regularly log in and be exposed to contents being published during the intervention period.

After completing recruitment (over a 3-week period), YBLGBM peer leaders launched the intervention by posting materials according to the timeline developed during the training and intervention refinement period (Multimedia Appendix 1). We held weekly project meetings with peer leaders assigned to both E-PrEP and E-Health to ensure appropriate implementation and to discuss logistic, ethical, or other issues that may have arisen. Each group met on different days of the week with different research staff members to help minimize potential contamination. Content for both arms was posted over a 6-week period using Buffer [71], which allowed all the posts to be prescheduled and published in the private groups using the peer leaders’ existing accounts. This approach ensured standardized publication of contents (ie, at the same time and day for each post) and reduced reliance on peer leaders being asked to post contents every day. We collected outcome data from the network study participants via online surveys at baseline, 6 weeks, and 12 weeks. Participants were given US $20, $30, and $40 online gift cards as incentives after completion of assessments at baseline, 6 weeks, and 12 weeks, respectively.

We administered online surveys to collect self-reported data at the 3 time points (baseline, 6 weeks, and 12 weeks). We collected data on PrEP use and intention to use, PrEP knowledge and attitudes, self-efficacy for PrEP care, sexual behaviors, sexually transmitted infection and HIV testing, and other covariates listed in Table 2. At each assessment period, participants were sent automated email reminders, followed by direct social media or text message reminders (sent by study staff 24 hours later if surveys were not completed). We repeated these reminders every 2 days until assessments were completed, for a period of 2 weeks. To assess possible contamination between study arms, the 6-week assessment displayed random samples of campaign posts from both E-PrEP and control arms to all participants and assessed recall [34,81,82].

We collected number of posts viewed, comments, and likes for each participant at 6 weeks. For each post in each of the private peer Facebook groups, we manually collected view data in a spreadsheet by documenting whether a participant had viewed the post or not, coded as 0 (not viewed) and 1 (viewed). For participants using Instagram, we used the same approach to document likes and comments for each of the posts. We also extracted additional engagement data (comments, likes, and reactions) using Grytics software (1339 SAS) for the Facebook groups (not available for Instagram). Grytics is a third-party analytics app that extracts Facebook group engagement data. At the time of the intervention, Grytics did not extract group members’ profile data such as age or gender. Sharing of posts on Facebook was disabled for the trial to reduce potential contamination.

Our primary outcomes are (1) self-reported PrEP uptake or intention (measured by indicating either current use of PrEP or intention to use PrEP in the next month) [83], and (2) change on the PrEP motivational cascade [84] at 12 weeks. Secondary outcomes are PrEP knowledge [85], PrEP-related stigma [86], attitudes about PrEP, and access to primary or sexual health care [87]. We will also explore potential changes in social network factors (eg, social support) and whether social media engagement correlated with the primary and secondary outcomes.

To test feasibility, we will assess online process measures, including participation rate (number of individuals screening eligible and then ultimately joining the study) and retention metrics (number of participants actively leaving or “unjoining” a social media study site and number of respondents completing follow-up assessments).

We will evaluate acceptability by assessing engagement activity (eg, number of posts viewed, number of individuals commenting or liking posts), satisfaction with the intervention, and willingness to continue participating if it were an option, joining a similar study again, and likelihood of recommending friends to join this study if it were an option.

First, we will compare groups for equivalence at baseline, using chi-square test, t tests, or nonparametric tests as appropriate. In addition, we will determine whether subgroups (eg, grouped by social media platform used, peer leader group, race/ethnicity, and sexual orientation) differ with respect to the primary and secondary outcomes. Relevant differences will be used as covariates in subsequent models. To compare differences in outcomes between arms over time, we will use repeated-measures mixed-effects logistic models, with the treatment arm as a fixed effect, peer leaders as a random intercept, repeated measures from a same participant as another random intercept, and time (as a linear term) as an independent variable. We will also look at residual plots to determine whether we need to include nonlinear terms for time.

Next, we will use hierarchical models to examine the role of secondary outcomes as potential mediators of change over time; for example, are there differences by condition in PrEP knowledge or attitudes that account for differences in the outcomes? Finally, we will assess the impact of the exposure arm (E-PrEP or E-Health control) and selected covariates (significant in the bivariate analysis at P<.15) on the primary and secondary outcomes using mixed-effects models.

To assess the impact of online process and engagement metrics on the outcomes, we will assess in the E-PrEP arm whether intervention exposure and engagement metrics correlate with the primary outcomes of uptake of PrEP and intention to use PrEP. In exploratory analyses, we will assess whether group activity (ie, total number of network participant reactions and posts) correlate with the primary outcomes. To overcome possible limitations due to intracluster correlations, we will also include a modified sensitivity analysis that examines outcomes in relation to participants’ network size and perceived affinity to the peer leaders.

Unlike in prior internet-based interventions, we proposed to use online venues already frequented by YBLGBM and to push intervention components to private home pages, visible only to participants, thus obviating the need to return repeatedly to specific study sites and potentially facilitating engagement. Peer leaders were familiar to participants and served as both recruiters and messengers, delivering contents framed in the peer leader’s communication style. We believe these design considerations will help mitigate attrition, observed in prior Web-based interventions.

The internet’s strength is that it delivers multidimensional intervention components [88]. Social media’s strength in diffusing innovations is that it facilitates information sharing within and between networks at a pace not previously possible. The nature of social media means there may likely be contamination between study arms. We minimize this to the extent possible by randomizing by peer leader, blinding peer leaders and study participants, and limiting access to the private study sites and contents. Additionally, peer leaders all agreed to not share any contents of materials being posted in their private Facebook or Instagram study groups until after the 12-week assessment.