The Association Between Homophily on Illicit Drug Use and PrEP Conversations Among Latino Men Who Have Sex with Men Friends: A Dyadic Network and Spatially Explicit Study

Abstract

Despite the wide availability of PrEP, Latino men who have sex with men (LMSM) continue to experience access barriers. Novel HIV prevention research strategies to increase PrEP uptake and adherence among the high incidence populations, such as LMSM who misuse drugs, include social network analyses. This study identified the associations of drug use homophily within LMSM friendship networks and PrEP promotion conversations and described the physical overlap between geographic drug risk areas with conversations of PrEP promotion. Respondent-driven sampling was used to recruit 10 sociocentric networks. Quadratic assignment procedure (QAP) correlations and multiple regression QAPs were used to identify influences of drug use homophily, and geocoding and visualizations to describe drug use polygons and PrEP conversations. Friendship relationships in which both friends used cocaine or marijuana were more likely to report PrEP-related conversations in the past six months. The likelihood of talking about PrEP in the next six months was higher among dyads with cocaine use homophily and ecstasy use homophily, while lower among dyads with marijuana use homophily. Participants reported using marijuana and cocaine throughout Miami-Dade County while ecstasy polygons were mostly in urban areas. The majority of drug polygons associated with PrEP conversations were located in north and central Miami. Future interventions can consider enrolling entire sociocentric friendship groups, configuring friendship networks to connect those without PrEP information to those with information, and incorporating peer leaders.

Introduction

Latino men who have sex with men (LMSM) are disproportionately vulnerable to HIV (Centers for Disease Control & Prevention, 2019). Despite new infections decreasing among non-Latino Black and non-Latino White MSM, new diagnoses increased by 30% among LMSM from 2010 to 2016 (Centers for Disease Control & Prevention, 2019). If current rates persist, one in five LMSM will be diagnosed with HIV by age 50 (Hess et al., 2017). LMSM living in Miami-Dade County, Florida (Miami) are at even higher risk than those living elsewhere in the USA as this county continues to experience an HIV incidence rate four times the national average (Centers for Disease Control & Prevention, 2020). Of new HIV diagnoses from 2015 to 2017 in Miami, half were LMSM, indicating the need for additional HIV prevention efforts in this population (HIV/AIDS Section, Division of Disease Control and Health Protection, & Florida Department of Health, 2018). Despite Pre-Exposure Prophylaxis (PrEP) reducing HIV by over 90%, LMSM encounter barriers to its access (Smith et al., 2011). LMSM are aware of PrEP and express high levels of interest in using it; however, PrEP use among LMSM varies widely, with uptake ranging between 3 and 30% (Holloway et al., 2017; Hoots et al., 2016; Kuhns et al., 2017; Mantell et al., 2014; Strauss et al., 2017). It is imperative that HIV prevention research concentrates on understanding new strategies, such as social network analyses, to increase PrEP uptake and adherence among those groups at highest risk, such as LMSM living in Miami (Kanamori et al., 2017; M. Kanamori et al., 2019a, 2019b; Liu et al., 2014).

Social networks are important transmitters of norms and behaviors. Substances such as cocaine, ecstasy, and crackcocaine are often used in social settings with friends and/or sexual partners (Finlayson et al., 2011; Hammoud et al., 2018; Patterson et al., 2005; Robles, 2018). Prior research has shown that homophily, defined as the interaction with other “individuals similar to themselves in respect to a variety of qualities and characteristics,” does not only occur in the context of socio-demographic factors, but also in contextual factors such as drug use (Latkin et al., 1995; McPherson et al., 2001; Monge et al., 2003; Valente et al., 2004. Furthermore, hearing about PrEP from a friend often leads to becoming a self-referral for PrEP (Algarin et al., 2019; Ezennia, et al., 2019; Fuchs, 2015). For this reason, it is critical to delineate the important role social networks play in disseminating PrEP-related information and encouragement to enroll in PrEP programs.

According to the Syndemic Theory, “two or more afflictions, interacting synergistically, contribut[e] to excess burden of disease in a population” (Wilson et al., 2014). Among MSM, those who use substances are at the greatest risk for HIV infection (Margolis et al., 2014; Pines et al., 2014; Plankey et al., 2007). Nearly a third of new HIV infections among MSM are associated with non-injection drug use (Mansergh et al., 2008; Van Tieu & Koblin, 2009). There is a resurgent drug use epidemic that is especially notable in Miami-Dade County. In this county, the 12-month prevalence of drug use is among the highest nationally for MSM, with a high prevalence of marijuana and party drugs such as stimulants (e.g., cocaine, ecstasy) relative to other drugs (Finlayson et al., 2011; Robles, 2018). The syndemic of drug use and HIV can be explained by the increase in sexual risk behaviors when using these party drugs (Hammoud et al., 2018; Patterson et al., 2005).

Social networks, substance use, and HIV infection clusters are also bound by geography (Des Jarlais et al., 2018, 2019; Gafos et al., 2019; Gelaw et al., 2019; Reid, 2009). Geospatial analysis can explain the syndemic of HIV and drug use by identifying overlapping geospatial clusters of HIV (Gelaw et al., 2019) and drug use (Des Jarlais et al., 2018; Reid, 2009). Previous research found that different sexually transmitted infections were related to specific geographic areas (Gafos et al., 2019). These high prevalence and incidence areas are traditionally called “hotspots” (Des Jarlais et al., 2019). However, geographic patterns of HIV infection have changed, in part due to geosocial networking phone apps that allow people to find casual sex partners almost everywhere (Francisco Luz Nunes Queiroz et al., 2017).

Because there is limited information on LMSM social networks’ characteristics, there is a need to understand how drug use patterns within social networks can influence LMSM PrEP awareness and enrollment. The characterization of how social network structures and geospatial contexts facilitate or hinder PrEP communication can help HIV prevention initiatives develop innovative approaches to increase PrEP uptake in the LMSM community. This intersection may be the missing key to explaining how PrEP programs for drug users can capitalize on the multidimensionality of social network dynamics and geographic-based socializations. This paper aims to: 1) identify drug use homophily inside LMSM friendship networks and their association with PrEP conversation and promotion, and 2) describe the geographic overlap between drug risk areas with PrEP conversation and promotion.

Methods

Eligibility Criteria, Study Sample, and Recruitment

Participant inclusion criteria were: 1) cis-male identity, 2) HIV-negative status, 3) sex with a man in the past six months, 4) Hispanic, Latino, or Latinx identification, and 5) qualification for PrEP prescription and candidacy in accordance with CDC PrEP Clinical Practice Guidelines (Centers for Disease Control & Prevention, 2018). Data were collected from October 2018 to August 2019. To ensure that our study methods, needs, and materials were culturally appropriate, we partnered with a community-based HIV prevention organization (CBO) that prioritized the population of LMSM living in South Florida.

This study included 10 sociocentric networks, each consisting of a constrained group of 13 LMSM friends. These sociocentric networks were created using respondent-driven sampling (n = 130 participants). The project coordinator recruited a seed for each network (n = 10) from two of our CBO partner’s non-clinical locations. Recruitment materials were adapted for Latin cultural values such as simpatía (cultural script characterizing Latinos as agreeable, friendly, sympathetic, and polite) and personalismo (importance that Latinos place on personal character and inner qualities such as respectful listening and caring interactions) (Ayon & Aisenberg, 2010; Marin, 1989; Ramírez-Esparza et al., 2008). For instance, our recruitment material included a picture portraying a friendship network of LMSM hugging each other in front of a car parked on a southern Miami-Dade County road, thus encouraging potential participants to join the study. Another picture portrayed a friendship network of LMSM enjoying a day in South Beach, one of Miami’s iconic gay beaches. Part of the text included in the recruitment material stated, “Special invitation: we would like to invite you to PrEParados that has specially being created for Latino gay and bisexual men 18–39 years old,” with the gay ribbon included as a banner. Participants interested in being part of the study were encouraged to contact two coordinators, one for LMSM who preferred to speak English, and another for those who preferred to talk in Spanish. This information was provided next to a picture of a smiling LMSM coordinator from the same age range. To address potential network overlaps, seeds were randomly drawn from the CBO partner clientele by randomly selecting day/time intervals for recruitment at one of two sites. To have an even distribution of PrEP use among seeds, we recruited five seeds who reported using PrEP and five who reported not using PrEP, ensuring an adequate comparison. Each of the ten seeds invited three friends (referred to as first-order friends). These first-order friends then each invited three friends (second-order friends). If any friends declined to participate, the seed or first/second-order friend was asked to invite another friend. If a participant was able to recruit only one or two friends, seeds and/or other first-order friends were asked to recruit a fourth friend, who recruited additional friends using this respondent-driven sampling approach until 13 LMSM were enrolled into one sociocentric network. Participants could only be a part of one sociocentric network. Each participant received a $50 gift card as compensation for their time, and a $10 gift card for each friend referred and enrolled in the study.

Data Collection Procedures

After providing written and verbal consent to participants (in either Spanish, English, or both), an interview was administered in a private office at one of our community partner’s sites. Interviews lasted approximately two hours. The interview was first developed in English, then translated into Spanish, then back-translated into English by a certified translator, and verified by a trained bilingual study author, as is commonly used in cross-cultural research (Brislin, 1970). Interviews were pilot tested to achieve clarity in language and appropriateness of the content. Trained LMSM interviewers questioned participants in either English, Spanish, or both. Interviewers used iPads to administer the interviews, and responses were recorded using REDCap (Research Electronic Data Capture) hosted at the University of Miami (12, 13). Sociocentric network data was analyzed using R (R Core Team, 2019). Drug use data was collected using the ESRI Web GIS online software (ESRI, 2014).

Measures

Participants were asked if they personally knew or interacted with all other members from their sociocentric network. Study staff did not reveal the full names of participants to other network members. Instead, they referred to participants in a network by their first name, and traced the relationship to each respective participant (e.g., “Do you know A.A.? He is friends with B.B. who referred you to the study”). Then, participants were asked to rate each member of the network they knew across the following PrEP-related communication items:

“Frequency in which participants talked to their friends (within the network) about PrEP” (3-point scale: 1 = never, 2 = sometimes, and 3 = frequently);

“Likelihood of participants talking with their friends (within the network) about PrEP within the next six months” (4-point scale: 1 = not likely, 2 = somewhat not likely, 3 = somewhat likely, 4 = very likely);

“Likelihood of participants encouraging each friend (within the network) to begin using PrEP within the next six months” (4-point scale: 1 = not likely, 2 = not very likely, 3 = a little likely, 4 = very likely).

Drug use polygons.

To collect spatially explicit data, interviewers displayed a South Florida map (centered on Miami-Dade County) on a computer screen. Participants were asked to draw a polygon of areas in which they currently use, obtain, or use and obtain ten types of drugs. Interviewers selected the corners of these drug use polygons to create shapes. Once a polygon was created, participants were asked about their drug-related activity (use, obtain, or both) within polygons. Three types of drugs (cocaine, ecstasy, and marijuana) were identified as most common among the sample and were included for analyses as binary variables (1 = use, obtain, or use and obtain; 0 = do not use).

Dyad-level variables were created to measure drug use similarity. Dyads were scored 1 for homophily (e.g., both members either used or did not use X drug) and 0 for heterophily (e.g., only one member used X drug). In addition to these similarity measures for each drug type, a continuous variable of the number of drugs each participant used was constructed. This variable was then compared to each friendship tie by calculating the absolute difference: lower scores indicated higher homophily.

Data Analyses

To analyze the association between drug use homophily and heterophily with PrEP-related communication, we used quadratic assignment procedure correlation (QAP) and multiple quadratic assignment procedure (MRQAP) tests. QAPs and MRQAPs serve as multiple regression analyses for networked or dyadic data which is structured in sociocentric matrices (Dekker et al., 2007; Krackhardt, 1987, 1988). MRQAPs are homologous to “their nonnetwork counterparts [multiple regressions] with respect to parameter estimates,” with the exception that they rely on randomized matrix permutation to produce estimates and test for significance (Borgatti & Cross, 2003, p. 438). Each of the MRQAP models is based on the aggregated results from 1,000 permutations. Since our data were organized by network cohorts, 130 × 130 matrices were constructed, which included structural missing data across different cohorts. A tie across networks could not exist.

All MRQAP tests were run using the sna package (Butts, 2019) for R (R Core Team, 2019). We adapted the netlm function to permute the 130 × 130 matrices for every 13 × 13 section for a total of 10-nested network matrices. Correlation distributions were then aggregated to produce estimates based on all network cohorts as a function of the complete sample. Additionally, visone software was used to visualize our networks (Brandes & Wagner, 2004). Three adjacency matrices, which contained the scores for our outcome variables, were extracted for each of the ten networks. These were then imported to visone, and graphs were created (see Figs. 1 and 2). Since cocaine and/or ecstasy use homophily was of importance for PrEP-related communication, network visualizations focused on these two drugs.

Fig. 1.

Network visualization of friendship ties with cocaine and ecstasy use

Fig. 2.

Network visualization of past or future PrEP-related conversations with cocaine and ecstasy use

Results

Table 1 shows the socio-demographic characteristics of the participants. About a third of participants reported using PrEP at the time of the study. Almost all participants self-identified as gay, with a small minority identifying as bisexual. Participants were predominantly Latino White, with some college, employed full-time, single or never married, and born in the USA. Of those born outside of the USA, more than one-third were born in Cuba or another Latin American country. Approximately three-fourths of the participants reported using marijuana (76%), more than half said they used cocaine (55%), and about a quarter mentioned they used ecstasy (23%). These three drugs were the most common among our sample. Other drug types considered were ketamine (9%) and non-prescribed medications (7%).

Table 1.

Participants’ socio-demographic characteristics (N = 130)

Characteristic	Percentage
Mean age (SD)	28.31 (4.2)
Mean years lived in South Florida (SD)	19.4 (11)
Mean years lived in the US (SD)a	15.43 (10)
Mean age when immigrated to the US (SD) a	12 (8)
Currently uses PrEP (Yes)	30%
Racial Identity
White	72%
Black/African American	4%
Multi-racial	19%
Other	6%
Education
High school or trade school	5%
Some college	52%
Bachelor’s degree	31%
Post-Graduate	3%
Employment Status
Full-time	84%
Part-time	9%
Other (e.g., student, unemployed)	7%
Marital Status
Single or never married	87%
Married	8%
Have a domestic partner	6%
Household Income
$24,999 or less	13%
$25,000–$34,999	43%
$35,000–$49,999	28%
$50,000 or more	17%
Country of birth
USA	56%
Latin American country	44%
Country of birth for Foreign Born1
Cuba	39%
Nicaragua	9%
Dominican Republic	9%
Honduras	7%
Puerto Rico	5%
Perú	5%
El Salvador	2%
Other Latin American country	25%

^aInformation from foreign-born participants

Frequency of PrEP-Related Conversations

Participants reported a total 813 outgoing friendship ties (dyads). Eleven percent of these dyads were pair of friends who had frequent conversations, 35% who talked sometimes, and 54% who never talked with their friend about PrEP.

Homophily

As shown in Table 2, dyads in which both friends used cocaine were more likely to have had PrEP-related conversations in the past six months (B = 0.36; p-value < 0.001). Homophily on marijuana use was associated with a lower likelihood of having PrEP-related conversations in the past six months (B = −0.23; p-value ≤ 0.05).

Table 2.

Likelihood of PrEP Conversations and Encouragement for Use among LMSM, stratified by Drug Use Homophily and Heterophily

	Drug use homophilya			Drug use heterophilyb
	B	p-value		B	p-value
Freq of PrEP conversation
Cocaine	0.36	0.00	***	−0.30	0.00	**
Ecstasy	0.11	0.32		0.02	0.78
Marijuana	−0.23	0.03	*	0.09	0.46
Non-pres Meds	−0.40	0.20		0.20	0.03	*
Ketamine	−0.18	0.48		0.11	0.33
Number of drugs diff	0.03	0.40		0.03	0.69
Talk PrEP next six months
Cocaine	0.30	0.00	***	−0.55	0.00	***
Ecstasy	0.33	0.02	*	−0.18	0.05
Marijuana	−0.41	0.00	**	0.06	0.82
Non-Pres Meds	−0.35	0.37		0.26	0.02	*
Ketamine	−0.12	0.67		0.03	0.82
Number of Drugs Diff	0.04	0.32		0.18	0.03	*
Convince to use PrEP
Cocaine	0.32	0.00	***	−0.58	0.00	***
Ecstasy	0.49	0.00	***	−0.16	0.11
Marijuana	−0.44	0.00	**	0.12	0.65
Non-pres meds	−0.24	0.55		0.23	0.06
Ketamine	0.09	0.75		0.08	0.59
Number of drugs diff	0.06	0.11		0.19	0.02	*

^aHomophily: Both members of a dyad report either using/obtaining/using and obtaining each drug or not using/obtaining/using and obtaining each drug

^bHeterophily: One member of a dyad reports using/obtaining/using and obtaining each drug, and the other members does not report using/obtaining/using and obtaining each drug

^*p ≤ 0.05

^**p ≤ 0.01

^***p ≤ 0.001.

All significance based on 1,000 permutations

Heterophily

Cocaine use heterophily was associated with less likelihood of having had a PrEP-related conversation in the past six months (B = −0.30; p-value ≤ 0.01). Heterophily based on non-prescribed medication use was associated with a higher likelihood of having had a PrEP-related conversation in the past six months (B = 0.20; p-value ≤ 0.05).

Likelihood to Talk About PrEP in the Next Six Months

Twelve percent of the dyads included a friend reporting “very likely”, 15% “somewhat likely,” 25% “not very likely,” and 48% “not at all likely” to talk to their friend about PrEP in the next six months.

Homophily

Higher likelihood of talking about PrEP in the next six months was associated with cocaine use homophily (B = 0.30; p-value ≤ 0.01) and ecstasy use homophily (B = 0.33; p-value ≤ 0.05). Marijuana use homophily was associated with a lower likelihood of talking about PrEP in the following six months (B = −0.41; p-value ≤ 0.01).

Heterophily

Friendships in which only one person from the dyad used cocaine or ecstasy (marginal effect) had a lower likelihood to talk about PrEP in the next six months (B = −0.55; p-value ≤ 0.001; and B = −0.18; p-value = 0.05, respectively). Dyadic heterophily based on non-prescribed medication use was associated with a higher likelihood of talking about PrEP in the next six months (B = 0.26; p-value ≤ 0.05). A higher difference in the number of drugs used by each friendship dyad was associated with a higher likelihood to talk about PrEP in the next six months (B = 0.18; p-value ≤ 0.05).

Likelihood to Encourage Friend to Use PrEP

Sixteen percent of these friendship dyads included a friend who reported that he was “very likely”, 12% “somewhat likely,” 23% “not very likely,” and 49% “not at all likely” to talk to his friend about PrEP in the next six months.

Homophily

Participants reported a greater likelihood to encourage a friend to begin using PrEP in the next six months if their dyadic friendship included cocaine use homophily (B = 0.32; p-value ≤ 0.001) or ecstasy use homophily (B = 0.49; p-value ≤ 0.001). Homophily based on marijuana use was associated with a lower likelihood to encourage that friend to begin using PrEP in the next six months (B = −0.44; p-value ≤ 0.001).

Heterophily

Heterophily based on cocaine use was associated with a lower likelihood to encourage a friend to begin using PrEP in the next six months (B = −0.58; p-value ≤ 0.001). The higher the difference in the number of drugs used by each friendship dyad, the higher the likelihood to encourage a friend to begin using PrEP in the next six months (B = 0.19; p-value ≤ 0.05).

Network Visualizations of PrEP Outcomes and Drug Use

Figure 1 shows ten network visualizations of friendship ties. Four networks (networks 1.A, 1.D, 1.H, and 1.I) had densities above 0.50. All networks were composed of at least one member who either used ecstasy or cocaine. Eight networks include at least one member who used both ecstasy and cocaine. All networks had at least one member who did not use either ecstasy or cocaine.

Figure 2 includes ten network visualizations with connections representing previous PrEP conversation between a pair of friends and future potential for PrEP information dissemination in the next six months. More than half of the dyads within five of the networks had previously discussed PrEP (networks 2.B, 2.C, 2.E, 2.G, and 2.I). In four of the networks, about a quarter or more of existing ties that had not previously discussed PrEP reported future potential for sharing information about PrEP (networks 2.D, 2.F, 2.H, 2.I). The two networks with the greatest potential for sharing PrEP information in the future contained at least one ecstasy user and at least four cocaine users (networks 2.D and 2.H).

PrEP Conversation Areas and Drug Use

Participants identified areas in which they reported obtaining, using, or using and obtaining one of ten drugs. Figure 3 includes three maps displaying areas in which participant reported using marijuana (Fig. 3a; n = 140 areas), cocaine (Fig. 3b; n = 75 areas), or ecstasy (Fig. 3c; n = 44 areas). While marijuana and cocaine polygons were spread across Miami, ecstasy polygons were mostly in Downtown Miami, Miami Beach, and the predominantly Latino neighborhoods of Hialeah and Doral. The majority of drug polygons associated with PrEP conversations were located in north and central Miami, but not the southern Miami area of Homestead— where the seasonal farmworker community resides.

Fig. 3.

Participants’ marijuana, cocaine, and ecstasy use by geographical area: a Marijuana use areas. n = 106 areas talked about PrEP; n = 34 areas did not talk about PrEP. b Cocaine use areas. n = 59 areas talked about PrEP; n = 15 areas did not talk about PrEP. c Ecstasy use areas. n = 30 areas talked about PrEP; n = 14 areas did not talk about PrEP

Discussion

This social network study aimed to describe the overlap between HIV, friendship drug use networks and drug use risk areas among LMSM living in Miami-Dade County, a group at highest risk of HIV and substance use disorder. We found that homophily on drug use was associated with conversations about PrEP—while homophily on cocaine and ecstasy use facilitated PrEP conversations, dyadic marijuana use served as a barrier to PrEP conversations. Our geospatial findings identified that marijuana risk areas were dispersed around Miami-Dade County, and cocaine and ecstasy risk areas were mainly concentrated in more urban and highly trafficked areas.

Our network findings suggest that interventions based on sociocentric friendship networks could address the Latino disparity in PrEP uptake. This suggests that PrEP information dissemination and promotion strategies should be tailored to specific drug use friendship networks. Previous research has shown the importance of social networks in various health-promoting behaviors. The best-known example is perhaps Alcoholics Anonymous (Groh et al., 2008; Kaskutas et al., 2002), a program built around social networks and social support. Social networks are the most significant mediator of change in this instance, even over family networks. These results play into the narrative that homophily can mediate powerful behavioral change (Groh et al., 2008; Kaskutas et al., 2002). Thus, if homophily serves to mediate protective behaviors, such as abstinence, homophily also likely plays a role in diffusing other health-promoting behaviors. In the context of HIV prevention, friendship networks could be used to diffuse correct information about PrEP and decrease PrEP stigma (Mehrotra et al., 2018; Phillips et al., 2019).

Our findings suggest that geographic regions in which people use each of these specific drugs differ, as did areas in which people conversed about PrEP. Our study’s areas of PrEP conversations align with maps from previous research conducted in Miami that found a high density of HIV prevention programs in these same areas (Kanamori et al., 2020). HIV prevention, PrEP promotion/navigation, and substance use prevention programs can consider expanding services to include areas of high drug use identified in this analysis and areas without accessibility, as identified in previous research (Kanamori et al., 2020). Syndemic theory posits that high drug use areas co-occur with high HIV risk areas, therefore, these findings offer an opportunity for the tailoring of interventions by drug and geographic location. Geographically unrestricted interventions, such as mobile vans and telehealth-delivered programs based in Miami-Dade County, can consider using our findings for service area provision.

Both the network and the geographic component of our findings show that there is an overlap of drug use areas and conversations surrounding PrEP. Our network component found that homophily on cocaine and ecstasy is positively associated with PrEP conversations. This implies that stimulant and party drug use networks have the potential to disseminate PrEP-related messages and information. Substance use prevention programs can be leveraged to disseminate PrEP-related information through the networks of people who use certain drugs (e.g., cocaine, and ecstasy). This supports the importance of community-based organizations with a “one-stop shop” model. Social network visualization approaches may be used by these programs to understand the structures that can help create or modify PrEP programs and drug use prevention/treatment services for LMSM. Network visualizations could also be used to guide the creation of policies that distribute PrEP among several venues providing drug use services to avoid bottlenecks or disruption in service provision (Kanamori et al., 2019a, 2019b).

There were limitations to this study. First, our findings may not be generalizable to the broader USA LMSM population for two reasons: 1) Miami is heterogenous for Latino nationalities, which is not seen throughout the rest of the nation, and 2) the study sample was identified using respondent-driven sampling, a nonrandom type of sampling method. Second, some participants may have been subject to recall and social desirability bias as behaviors were self-reported. To decrease this bias, we used forced-choice items and a hybrid of self-administered and interviewer assisted questionnaire. Third, the cross-sectional nature of the study cannot determine causality.

Conclusions

Future PrEP programs for drug users should consider the utility of a sociocentric friendship network component. We found that LMSM who share drug use patterns with their friends are more likely to share information about PrEP. Also, non-drug users could promote PrEP to their friends who use drugs. Current HIV prevention efforts should be strengthened, as evidenced by the increasing HIV infection rates among our priority population of LMSM, including those who use drugs. Instead of solely relying on social media or other mass communication efforts to increase PrEP-related information among this highly vulnerable group, future interventions can harness the power of friendship and drug use networks. This can include enrolling entire sociocentric friendship groups, configuring friendship networks to connect those with PrEP information to those without information, and incorporating peer leaders.