Individual and Social Network Structure Characteristics Associated with Peer Change Agent Engagement and Impact in a PrEP Intervention

Abstract

Interventions that effective strategies for engaging utilize the influence of peer change agents (PCAs) have been shown to be key populations in HIV prevention. To date, little is known about the characteristics of PCAs associated with their effectiveness. Drawing on data from a peer leader PrEP intervention for young Black men who have sex with men (YBMSM) (N=423), we evaluated the effects of experiential (i.e., living with HIV, PrEP awareness, PrEP use), psychographic (i.e., self-perceived leadership, innovativeness), and network (i.e., degree centrality, eigenvector centrality, and brokerage) characteristics on three effectiveness outcomes: 1) recruiting peers into the study; 2) completing “booster” sessions; and 3) linking peers to PrEP care. For each outcome, multivariable regressions were performed. On average, PCAs recruited 0.89 peers, completed 1.99 boosters, and had 1.33 network peers linked to PrEP care.

Experiential factors:

Prior PrEP awareness was positively associated with booster completion.

Network factors:

Being a network broker (i.e., connecting otherwise disconnected communities) was positively associated with peer recruitment but negatively associated with linking peers to PrEP, and degree centrality (i.e., the number of network connections someone has) and eigenvector centrality (i.e., being connected to well-connected network associates) were positively associated with linking peers to PrEP. Psychographic characteristics were not associated with any outcome. These findings can be used to inform PCA selection and to identify subpopulations who require additional support to excel as PCAs.

INTRODUCTION

Pre-exposure prophylaxis (PrEP) is a critical and powerful prevention tool in efforts to eliminate HIV in the United States [1], particularly in key populations like men who have sex with men (MSM). Since its FDA approval in 2012, however, racial inequities in PrEP access have restricted its potential in most at-risk communities. Evidence of limited PrEP uptake among Black MSM (BMSM) is widespread. In San Francisco, data from the 2014 National HIV Behavioral Surveillance revealed that only 7.7% of eligible BMSM were using PrEP [2]. In PrEP demonstration and implementation projects in Washington, DC [3] and New York City, NY [4], less than 15% of PrEP clients identified as Black. And, in Birmingham, AL, a retrospective chart review occurring in the only Ryan White Clinic providing PrEP services from 2014–2016 found that Black MSM were least likely to access PrEP, despite comprising 18 percent of the clinic’s patients [5]. These trends are in the context of young BMSM (YBMSM) leading new infections in these cities and nationally [6]. In light of current racial inequities in HIV incidence and PrEP use, there is an urgent need to increase PrEP access and adoption among YBMSM [2, 5, 7].

Yet, the question remains: who is best poised to be agents of this change? Increasingly, behavioral interventions aimed at increasing HIV prevention engagement in at-risk populations like YBMSM are positioning peers in this role [8–12], and the popular opinion leader (POL) model of intervention [13, 14] is an exemplar of this approach. Informed by theories of social diffusion [15], POL intervention models involve recruiting and training change agents from within the target population (i.e., peer change agents (PCAs)), to promote HIV prevention messages and behavior change through their interpersonal networks [14]. As such, peer-led interventions not only offer the opportunity to reach larger portions of communities at risk by treating social networks as opposed to individuals in isolation, but through their emphasis on peer-driven leadership and support, they also help overcome socially derived barriers like stigma, homophobia, and institutional racism [12], which are well-known obstacles to HIV prevention and care engagement for MSM of color [16–18].

Despite their promise, POL interventions have yielded mixed results. When implemented in an intervention for BMSM in 3 southern US cities, researchers reported decreases in self-reported sexual risk behaviors including decreases in number of partners and episodes of condomless anal sex [19]. In contrast, in a randomized trial across 5 countries, the deployment of PCAs resulted in no significant differences in behavioral risk reduction or disease incidence between control and intervention groups [20].

Inconsistent findings like these suggest greater attention to the criteria used to select PCAs is needed [21, 22], as the selection method used will have downstream effects on the success of a study. In their review of the subject, Valente and Pumpuang [21] identify a repertoire of techniques used for identifying PCAs, including selecting people on the basis of: (1) their desire to help, (2) aspects of their personality (i.e., psychographic qualities), (3) their nomination by community experts or study staff, and (4) their perceived or actual positions in salient social networks among community members (i.e., sociometric selection). Others have surmised that drawing on community members who have personal interest in and/or experiences with the behavior of interest (i.e., experience-based selection) is optimal, as this may increase their enthusiasm and engagement in the study [23, 24].

What remains to be seen, however, is how well these different qualities of an individual’s leadership potential stand up to one another in terms of their ability to effectively identify PCAs who will be more engaged in a study and who will have the greatest impact on its outcomes. To these ends, we perform an analysis to determine which aspects of peer leadership potential predict a PCA’s study engagement and intervention impact in a PrEP intervention among YBMSM. Specifically, we focus on three types of PCA qualities: (1) those that reflect life experiences that may increase their knowledge of and interest in PrEP adoption (experiential factors); (2) those that reflect personality traits associated with peer leadership (psychographic factors); and (3) those that reflect an individual’s social position among other YBMSM (network factors). The post-hoc identification of characteristics of effective PCAs is a critical step toward the development of future HIV prevention interventions. This information can help maximize the impact of interventions by selecting individuals who will be more engaged, while also identifying subpopulations among PCAs who may require additional support.

METHODS

Study Setting and Population

Data for this study were collected March 2016 – March 2018 as part of PrEP Chicago, a randomized controlled trial (RCT) PrEP for prevention intervention among 423 young BMSM (YBMSM) living in Chicago. Cook County (Chicago) is an ending the epidemic jurisdiction and ranks 3^rd in total HIV burden behind Harris and Miami Dade Counties [25]. Participants were considered eligible if they met the following criteria: 1) 18–35 years of age, 2) identified as a person of color, 3) assigned male sex at birth, 4) had sex with a man in the past 12 months, and, because the intervention emphasized social media as a communication tool, 5) had an active Facebook profile. Once deemed eligible, individuals were assigned randomly to one of two treatment sequences: (1) receives PCA training in year 1 of the study or (2) receives PCA training in Year 2. As such, all 423 YBMSM participants enrolled in the study were ultimately trained as PCAs and followed for 12 months in that roll. Once participants were randomized, they were scheduled for a baseline visit. All participants provided written consent during that baseline visit.

Study Procedures

This study was approved by the Institutional Review Boards at the Biological Sciences Division and the National Opinion Research Center (NORC) at the University of Chicago. As described elsewhere [12], participants were recruited using respondent-driven sampling (RDS), a procedure well suited for identifying members of “hard-to-reach” populations like MSM [26]. A variant of snowball sampling [27, 28], RDS draws on peer referral chains, beginning with a set of initial “seeds” that meet study eligibility. Because seeds should have large social networks (i.e., are popular) and have ties to a diverse array of people belonging to different subpopulations [28–30], we selected YBMSM seeds based on their central or boundary spanning positions (i.e., structural signatures of popularity and diversity, respectively) within a previously derived Facebook friendship network among the target population [31]. Once a seed was enrolled and completed their baseline assessment, they were instructed to recruit up to 6 peers (or “sprouts”) who also met the eligibility criteria. Following enrollment, sprouts were also instructed to recruit peers, and the process continued until the recruitment target was reached. Participants received a $20 cash incentive for each peer whom they successfully referred into the study.

Peer Change Agent Training and Engagement

The peer change agent training is designed to develop a participant’s knowledge about PrEP, their willingness to discuss it with others, and to build motivational skills for engaging clients in PrEP care. The training workshop adapts the peer educational and mentoring program developed as part of the HIV Prevention Trials Network [32, 33] and is conducted in a single half-day session. The first half of the workshop focuses on building participants’ HIV and PrEP knowledge, particularly in the context of how each effects the Black MSM community. The second half of the workshop develops participants’ communication and motivational interviewing skills in order to increase their effectiveness as peer change agents. Specifically, participants practice initiating conversations about PrEP with at-risk peers and learn to identify strategies for supporting at-risk peers through the PrEP adoption process. More details about the training can be found elsewhere [12].

To maintain contact with participants after the initial session, staff administer a total of eight monthly telephone check-ins (i.e., “boosters”) with each participant, each lasting 10–20 minutes. Each booster has four components: 1) learning about specific peers that the participant wants to approach about PrEP, 2) devising personalized conversational strategies for approaching those peers, 3) troubleshooting communication barriers, and 4) setting personal outreach goals. The same staff mentor conducts all boosters with a given participant. Taken together, the PCA training workshop and the eight monthly booster calls (if completed) comprise the maximum dose of intervention actions to which a PCA study participant could have been exposed over the course of their 12-month enrollment.

Upon completion of the training, PCAs have the opportunity to engage with the study in three ways: 1) to recruit up to six peers into the study; 2) to participate in the aforementioned monthly boosters calls; and 3) to refer individuals to a PrEP warmline called the “PrEPline”, from which non-participant peers can learn more about PrEP and schedule an initial PrEP appointment with a provider. The PrEPline includes linkage to health care providers, health insurance, and low-cost or free medication, and is part of a wider citywide PrEP campaign [34].

Data Collection Modalities

Data collection occurred at Baseline, 12-month, and 24-month follow-ups. We use baseline data in the analysis featured here. In total, participants consented to three types of data collection. A computer-assisted self-administered survey captured information about PrEP knowledge and attitudes, sexual health behaviors, psychographics (i.e., measures of cognitive traits like personality, attitudes, and beliefs), and demographics. Blood tests determined the HIV and Syphilis status of each participant. And, to evaluate the impact of social connectivity on intervention outcomes, a list of each participant’s Facebook friends was collected using Facebook’s manual data download feature. A waiver of consent from the IRB for third party (non-participant) network members was obtained given the minimal risk to these individuals. Additional data protections to secure third party identities (e.g., hashing, numeric de-identification) were also established.

Outcome Measures

This study focuses on three outcome measures that relate to a PCA’s study engagement (peer recruitment and booster completion) and their impact on the community (peer linkage to PrEP care).

Recruitment.

Recruitment is a count of the number of eligible peers that a participant successfully recruited into the study. Successful recruitment occurs when recruited peers enrolled and completed the initial baseline visit. Each participant was allowed to recruit up to 6 peers.

Booster completion.

Booster completion is a count of the number of scheduled monthly check-in calls (8 in total) completed by each participant. The majority of check-ins were completed telephonically, however, they could also occur via SMS text or social media messaging according to client preference. A booster was counted as “complete” if all goals of the booster were achieved. Staff mentors assigned to each participant made three attempts to contact a participant before marking the booster “incomplete.”

Peer linkage to PrEP care.

The PrEP linkage outcome is an indirect measure of a PCAs impact on the target community to which they belong and the overall success of the intervention. Specifically, it is a count of the number of non-participant peers who were connected to a PCA on Facebook (i.e., via a Facebook “friend” tie) who met at least one of two conditions: 1) they were referred from the PrEPline, and/or, because some YBMSM bypassed the PrEPline and were referred directly to a PrEP provider by a health service agency, 2) they attended at least one PrEP consultation appointment at Howard Brown Health, the largest PrEP provider in Chicago. We view both linkage modalities, PrEPline referrals or attending a PrEP consultation, as critical early steps in the adoption process. To count as a PrEP-linked network peer, the peer had to have called/visited after the participant’s baseline visit and before the end of their 12-month post-training observation period. An honest broker managed the process of linking the PrEPline callers and PrEP clinic visitors to a feasible Facebook profile in the pooled list of study participants’ Facebook friends.

Referrals from the PrEPline were linked to participants in one of two ways, either the caller directly identified a study participant when asked from whom they heard about PrEP or the name of the caller appeared in the Facebook friend list of a given participant. Although the latter may not reflect direct communication about PrEP with a study participant, being exposed to the online social interactions and communication of a study participant may have an indirect effect on their interest in PrEP. Because some PrEPline referrals could have also visited the PrEP clinic, the list of PrEPline referrals were matched with the list of clinic visitors and any duplicates were excluded.

Covariate Measures

Experiential Factors.

Three aspects of a participant’s knowledge and life experiences — living with HIV, being aware of PrEP, and using PrEP — were evaluated as possible factors associated with PCA engagement and impact. To determine HIV status, we used confirmatory HIV testing data

or self-reported status if the participant opted out of HIV testing. Second, we account for a participant’s awareness and use of PrEP measured at Baseline. PrEP awareness was assessed by a single item, “Before joining the project, had you heard about PrEP?” (no/yes). PrEP use was assessed by asking those who were aware of PrEP whether they were currently taking PrEP (no/yes) or had taken PrEP in the past (no/yes).

Psychographic Factors.

Two personality constructs believed to be indictive of a PCA’s influence potential and their willingness to adopt new ideas and behaviors — leadership [35, 36] and innovativeness [15] — were included in each model. Leadership was assessed using a 9-item scale [37], the items of which inquire about the nature of a respondent’s interaction with friends and reflect the extent to which individuals give information about a topic and the extent to which others seek information from them. Innovativeness was assessed using an 8-item scale [38], the items of which inquire about the nature of a respondent’s original thinking and openness to new ideas. Items on each scale were measured on a 4-point agreement response scale (1=“Strongly disagree”, 2=“Disagree”, 3=“Agree”, 4=“Strongly agree”). Further details about our use of each scale are discussed elsewhere [39].

Network Factors.

Sociometric characteristics include three commonly used measures of an individual’s network position — degree centrality [40], eigenvector centrality [41], and network brokerage [42] — in a Facebook friendship network constructed from the Facebook friend lists downloaded from participants’ personal data archives. The network on which the sociometric measures are based represents observed friendship ties between study participants only; ties to non-participant Facebook friends are excluded from this portion of the analysis. Degree centrality represents the number of network connections that an individual has with other network members [40]. Eigenvector centrality measures the degree to which an individual is connected to other well-connected network members and, hence, their relative nearness to others in terms of the “overall” structure of the network [41, 43]. Finally, to account for the degree to which an individual connects disparate subcommunities within a network as well as the degree to which they may be receptive to new ideas [44], we measure an individual’s brokerage in the Facebook friendship network using Everett and Valente’s EV-brokerage measure [42].

Controls.

All models adjusted for demographic factors, including self-reported measures of a participant’s age, educational attainment (“Less than high school”, “High school diploma or GED”, “More than high school”), employment status (“Employed”, “Not employed”, and “Disabled”), and sexual orientation (“Gay”, “Straight”, “Bisexual”, and “Other”). Additionally, each model adjusts for the effects of date of enrollment in the intervention arm, given that participants enrolled in the study and were subsequently trained as PCAs on a rolling basis throughout the duration of the study. Those completing the intervention on the first day of the study were coded as “1” and those completing the intervention on the final day were coded as “760”.

Data Analysis

Descriptive and multivariable regression analyses were performed using STATA version 15 [45]. Each of three outcomes were modeled separately using a generalization of Poisson regression called negative binomial regression. Negative binomial regression is designed for count outcomes that are over-dispersed [46], as is the case for all three outcomes featured in this analysis. We interpret the coefficients of each negative binomial regression model in terms of incidence rate ratios (IRRs). IRR values are interpreted as the increase (when the IRR value is greater than 1) or decrease (when the IRR value is less than 1) in the outcome, of a magnitude equal to the IRR value, given a one unit increase in the covariate while holding constant all other variables experiential, psychographic, and sociometric factors as described above, while in the model. Each multivariable regression model includes covariates representing also adjusting for a series of demographic control variables. An alpha < 0.05 was used for all significance testing.

RESULTS

Characteristics of YBMSM Peer Change Agents (PCAs)

In total, 384 of 423 study participants completed the PCA training and provided Facebook friendship data and were, therefore, included in the sample. Characteristics of the PCAs in our sample are shown in Table 1.

Table 1.

Characteristics of Young Black Men Who Have Sex with Men (YBMSM) (N=384)¹: PrEP Chicago study, Chicago, IL, USA; 2016–2019.

Characteristics	N(%)
Demographics
Age, mean (SD; min, max)	25.71 (4.15; 18,35)
Educational Attainment
Less than high school	30 (7.8)
High school diploma or equivalent	249 (64.8)
More than high school diploma	95 (24.7)
Employment Status
Employed	192 (50.0)
Unemployed	161 (41.9)
Disabled	10 (2.6)
Sexual Orientation
Gay/Lesbian	240 (62.5)
Straight	11 (2.9)
Bisexual	98 (25.5)
Other	20 (5.2)
Experiential Traits
Living with HIV	167 (43.5)
PrEP aware	275 (71.6)
PrEP use (current or past)	38 (9.9)
Psychographics
Leadership, mean (SD; min, max)	3.2 (0.5; 1, 4)
Innovativeness, mean (SD; min, max)	2.7 (0.5; 1, 4)
Network Characteristics
Degree Centrality, mean (SD; min, max)2	15.87 (15.37; 1,74)
Eigenvector Centrality, mean (SD; min, max)3	0.23 (0.22; 0, 1)
EV Brokerage, mean (SD; min, max)4	45.99 (41.75; 0, 234)
Engagement and Effectiveness Measures
Peers recruited, mean (SD; min, max)	0.89 (1.35; 0,6)
Boosters completed, mean (SD; min, max)	1.99 (2.07; 0,8)
Peers linked to PrEP, mean (SD; min, max)	1.33 (1.85; 0,10)

¹384 of 423 participants successfully provided Facebook data at Baseline and were therefore included in the analysis.

²Degree centrality represents the number of YBMSM PCAs to whom an individual is connected to on Facebook.

³Eigenvector centrality represents the degree to which an individual is connected to other well-connected PCAs on Facebook.

⁴EV brokerage represents the degree to which an individual connects otherwise disconnected sub-communities in the Facebook friendship network among PCAs

At Baseline, PCAs were on average 26 years old and majorities had a high school diploma or equivalent (65%), were employed (50%), and identified as gay (62%). With respect to their HIV-related knowledge and experiences, approximately 43% were HIV positive, 72% had heard of PrEP prior to enrolling in the study, and 10% were currently using PrEP or had used PrEP in the past. Regarding personality traits, PCAs scored on average 3.2 and 2.7 (out of a maximum of 4) on the opinion leadership and innovativeness scales, respectively. With respect to their network positions, PCAs were on average connected to approximately 16 other study participants on Facebook, were only moderately connected to other well-connected network peers (Mean eigenvector centrality = 0.23), and had on average a normalized brokerage score of 46 (range of 0 to 234). Finally, in terms of their study engagement and intervention impact, PCAs recruited on average 0.89 peers (range of 0 to 6), completed 1.99 boosters (range of 0 to 8), and had 1.33 network peers linked to PrEP care (range of 0 to 10).

Multivariable analyses of factors associated with PCA engagement and impact

Three separate multivariable negative binomial regression analyses were performed for each outcome, the results of which are shown in Table 2.

Table 2.

Multivariable negative binomial regression models of the number of peers recruited into the study, the number of booster calls completed, and the number of Facebook friends who were linked to PrEP care.

	Model 1: Peers Recruited		Model 2: Boosters Completed		Model 3: Peer PrEP Linkage
	IRR	95% CI	IRR	95% CI	IRR	95% CI
Experiential Factors
Living with HIV	1.17	[0.81, 1.67]	0.97	[0.74, 1.27]	0.92	[0.74, 1.15]
PrEP Awareness	1.37	[0.88, 2.12]	1.55*	[1.11, 2.17]	1.16	[0.86, 1.56]
Prep Use (current or past)	0.66	[0.35, 1.25]	0.75	[0.48, 1.17]	1.06	[0.79, 1.42]
Psychographic Factors
Leadership	1.13	[0.80, 1.60]	1.09	[0.84, 1.42]	0.93	[0.75, 1.15]
Innovativeness	1.02	[0.70, 1.50]	1.02	[0.78, 1.35]	1.14	[0.92, 1.41]
Network Factors
Degree Centralitya	1.00	[0.99, 1.02]	1.00	[0.98, 1.01]	1.02***	[1.01, 1.02]
Eigenvector Centralityb	2.15	[0.68, 6.74]	0.86	[0.32, 2.29]	2.63**	[1.40, 4.94]
EV Brokeragec	1.01*	[1.00, 1.01]	1.00	[0.99, 1.00]	0.99**	[0.983, 0.995]
Controls
Age	1.03	[0.98, 1.08]	1.02	[0.98, 1.05]	1.02	[0.99, 1.05]
Educational Attainment
Less than high school	1.67	[0.95, 2.95]	0.83	[0.50, 1.37]	1.10	[0.75, 1.62]
High School or GED	ref	ref	ref	ref	ref	ref
More than High School	0.87	[0.56, 1.36]	1.35	[0.98, 1.87]	1.06	[0.83, 1.36]
Employment Status
Employed	ref	ref	ref	ref	ref	ref
Unemployed	0.82	[0.57, 1.18]	0.90	[0.68, 1.19]	0.79*	[0.64, 0.99]
Disabled	1.26	[0.46, 3.42]	0.87	[0.36, 2.11]	1.25	[0.64, 2.43]
Sexual Orientation
Gay	ref	ref	ref	ref	ref	ref
Straight	1.32	[0.33, 5.29]	1.13	[0.37, 3.46]	0.38	[0.09, 1.56]
Bisexual	0.90	[0.59, 1.]38	1.02	[0.74, 1.40]	0.79	[0.59, 1.07]
Other	1.15	[0.52, 2.56]	0.70	[0.37, 1.34]	0.99	[0.60, 1.62]
Date of Enrollment	1.00**	[0.995, 0.999]	1.00	[0.998, 1.00]	0.99***	[0.993, 0.995]

^*p<.0.05;

^**p<0.01;

^***p<0.001

^aDegree centrality represents the number of YBMSM PCAs to whom an individual is connected to on Facebook

^bEigenvector centrality represents the degree to which an individual is connected to other well-connected PCAs

^cEV brokerage represents the degree to which an individual connects otherwise disconnected sub- communities Facebook friendship network among PCAs

First, analysis of factors associated with peer recruitment (Model 1) reveal that a PCA’s network brokerage — i.e., the ability to connect otherwise disconnected subcommunities in network — is positively related to their peer recruitment (IRR = 1.01; 95% CI: 1.00–1.01]. Specifically, a one unit increase in network brokerage increases peer recruitment by a factor of 1.01 holding all other variables constant.

Second, completing booster calls with study staff (Model 2) was positively related to a PCA’s awareness of PrEP at baseline (IRR = 1.55; 95% CI: 1.11–2.17]. Those who had heard of PrEP prior to being activated as a peer leader had a rate of booster completion 1.55 times greater than those who had not heard of PrEP.

Finally, the analysis of a PCA’s impact on linking peers to PrEP care (Model 3) highlighted the importance of accounting for their structural position vis-à-vis other PCAs. Having more PCA Facebook friends [IRR = 1.02; 95% CI: 1.01–1.02] and being more connected on Facebook to other well-connected PCAs (i.e., eigenvector centrality) [IRR = 2.63; 95% CI: 1.40–4.94] were positively related to being connected to more non-participant peers who received initial PrEP care linkage. Network brokerage was negatively associated with peer linkage [IRR = 0.99; 95% CI: 0.983–0.995]. Results also show that PCAs who were unemployed had a lower peer linkage rate than employed PCAs [IRR = 0.79; 95% CI: 0.64–0.99].

DISCUSSION

Our analysis revealed several indicators of peer leadership potential that stand out as significant predictors of a PCA’s study engagement and impact. First, with respect to peer recruitment, we learned that being a broker in the Facebook friendship network among other study participants was a significant and positive predictor of a PCA’s recruitment productivity. Although the magnitude of this effect was small, it nonetheless suggests that PCAs who do a better job of recruiting are those who are embedded in more structurally diverse networks — i.e., their peers have different sets of friends than they do. Having access to more varied groups of peers, as brokers do, gives PCAs a larger and more heterogenous pool of people to choose from when recruiting peers into the study. Conversely, among PCAs who are embedded in more insular networks, there may be a real or perceived social cost to recruiting multiple peers for a PrEP for prevention intervention from within a network where everyone knows everyone else. Given that peer referral recruitment strategies like respondent-driven sampling (RDS) rely on adequate propagation of referral chains, recruiting RDS “seeds” based on their network brokerage may help optimize that process [12].

Second, with respect to booster completion, we learned that PCAs who came into the study with prior awareness of PrEP had an increased rate of booster completion relative to those who only became aware of PrEP for the first time through our intervention. Given that we also account for potential confounders like PrEP use and living with HIV, an experience that we know fosters awareness of PrEP, we consider the effect of PrEP awareness to be a unique one. As such, early exposure to PrEP through advertisements, campaigns, or personal conversations is an important indicator of one’s willingness to remain engaged in a study meant to advocate for PrEP adoption within their community. The fact that those who were unaware of PrEP entering the intervention were less likely to engage with study staff throughout the duration of their enrollment warrants further reflection about how the intervention’s training and support components might be adapted to better engage the PrEP unaware.

Finally, with respect to PrEP linkage, our analysis underscored the importance of considering a PCA’s network centrality. We learned that PCAs who were more central in the Facebook network among other PCAs — either by virtue of how many PCAs they were connected to (i.e., their degree centrality) or by virtue of how many well-connected PCAs they were connected to (i.e., their eigenvector centrality) — had more non-participant Facebook friends linked to initial PrEP care. At first glance, this would seem to reflect a proxy effect — i.e., that YBMSM study participants who are well-connected to other YBMSM study participants are also more likely to be well-connected to other non-participant YBMSM from the same community, and, therefore, are more likely to have more non-participant YBMSM Facebook friends who initiated PrEP linkage.

However, we also speculate a possible second-order effect suggested by this finding. By design, it is commonplace for Facebook users to be exposed to information being shared with and among their friends, even without contributing to these exchanges firsthand. So, given that PCAs were instructed throughout the study to use Facebook to engage with one another and to publicize information about PrEP, it seems plausible that PCAs with more Facebook ties to other PCAs will inevitably expose their mutual non-participant Facebook friends to a greater critical mass of information about the study and about PrEP, thereby increasing the chances that their non-participant friends will eventually initiate PrEP themselves. Research to verify this exposure thesis through an analysis of PCAs’ Facebook timeline communication is currently underway.

Also, worth mentioning are two noticeably absent effects. Contrary to prior research that found a link between living with HIV and productive peer recruitment [23, 24], being HIV positive was not associated with any form of study engagement or impact. We had hypothesized that living with HIV may create extra motivation for PCAs to engage in “intravention” within their networks, translating to increased engagement in a PrEP intervention [47]. Instead, having previous knowledge of PrEP, which many HIV positive PCAs had, was a more important indicator of a willingness to engage. In our minds, leveraging the hindsight and perspectives that people living with HIV possess toward persuading their HIV negative counterparts of the benefits of PrEP remains a potentially powerful strategy. That said, living with HIV may make some feel alienated from something like PrEP, which they can no longer benefit from personally. Moving forward, adapting PCA training and support components to speak more directly to the experiences and feelings of peer leaders who are living with HIV may help integrate them more effectively in the mission of encouraging PrEP uptake in their community.

Also contrary to extant literature, we found that the psychographic characteristics of PCAs, namely their perceptions of their own opinion leadership and innovativeness, were unrelated to their study engagement and impact. Instead, the sociometric analogs to these personality features — i.e., network centrality and network brokerage — proved to be more satisfactory indicators of how well PCAs performed. From analysis not presented, we know that the psychographic and sociometric measures of the leadership and innovativeness constructs are uncorrelated. As others have warned [21], this suggests that respondents may be intentionally or unintentionally inflating their responses in self-reports or that the psychographic and objective measures of each construct are in fact measuring different things. In either case, these results suggest that screening individuals for psychographic measures like perceived leadership may not be an effective means for identifying the most promising peer leader candidates. Further, without the need to screen for these traits, the peer leadership role is opened to a larger and more heterogenous pool of candidates.

Taken together, our findings underscore the importance of considering the multiple roles and responsibilities expected of a peer leader when thinking about which constellation of leadership indicators should motivate their selection. Although others have made similar calls to select an identification method that matches the type of advocacy that a peer leader is expected to perform [21], we highlight the need to also think about benchmarks that are more reflective of a peer leader’s commitment to the mission of the study. The success of a behavioral intervention that draws on the influence of opinion leaders rests on more than just their ability to affect change in their peers’ behaviors; it also depends on their willingness to engage in the growth of the study through responsibilities like peer recruitment and their willingness to engage with and learn from staff mentors who are there to support their efforts. Typically, the dissection of study engagement challenges takes place during or after the implementation of the intervention. Alternatively, interventionists may be able to preempt some of these engagement challenges by thinking more carefully about the level of engagement that one needs from a peer leader and the characteristics that may reflect that level of commitment when identifying individuals who will be best-suited for the role.

With the awareness that peer leaders in any given study have multiple obligations to fulfill to make the intervention effective, it is hoped that interventionists, when possible, will contemplate using multiple and diverse methods for identifying individuals for these roles. Some methods like selecting people on the basis of their position in the network may identify people who are seen as influential by others, while other methods like drawing on personal experience with the behavior may help identify leaders who are more willing to talk about the behavior or who are more committed to the mission of the study. Although different methods of peer leader selection might not identify the same set of individuals, all of these individuals ultimately have a role they can play in ensuring that an intervention is effective.

Our study should also be considered in light of its limitations. First, we cannot definitively conclude that PrEPline referrals or clinic visitors who appear in the Facebook friend list of PCAs were directly influenced by PCAs to initiate PrEP linkage. That said, information obtained by asking either party directly would likely have been incomplete and subject to reporting biases. For this reason, we developed a consolidated surrogate outcome designed to capture the more distal effects of being exposed to the online activity and interpersonal conversations of PCAs on a peer’s decision to initiate PrEP linkage, which may be a more reasonable expectation for how social influence works in naturalized settings. Second, our measure of peer PrEP linkage also does not allow us to connect the caller or clinic visitor to a single PCA. In some cases, a non-participant PrEPline caller may have been connected to more than one PCA who had completed their training at the time of their call. Third, our measure of peer PrEP linkage is not an actual measure of PrEP initiation. Nonetheless, we believe that initiating the linkage process (e.g., calling into a PrEP warmline or making an initial clinic appointment) is an important element in early stages of PrEP adoption that is often overlooked in PrEP demonstration projects [34]. And finally, we acknowledge that, by design, our eligibility criteria excluded YBMSM who did not have a presence on Facebook. As such, our study participants may be biased in other ways. That said, analyses involving the same cohort of YBMSM [48] revealed little to no differences within the cohort between Facebook users and non-users.

Despite these limitations, results of this study provide a retrospective account of the constellation of individual and social network characteristics associated with effective peer leadership both in terms of their engagement in study procedures and their overall impact on intervention outcomes. Our findings highlight the importance of considering network characteristics like centrality and brokerage as well as prior knowledge about the outcome of interest when selecting individuals to fill peer leadership roles. Although our findings are in part specific to the nature of the intervention and the target population, the overall retrospective analytic approach performed here can be used to inform future PCA selection and to identify subpopulations who may require additional supports to excel as PCAs.