“Todo se trata de a quién conoces”: Social Networks and Drug Use among Female Sex Workers Living with HIV in the Dominican Republic

Abstract

The purpose of this study was to characterize the social networks of female sex workers (FSWs) living with HIV in the Dominican Republic (DR) and to examine the association between daily drug use and network risk profile. The study employed a micro-longitudinal observational design using a 7-day mobile health (mHealth) daily diary to collect daily substance use behaviors and social network data was collected at study enrollment. A series of crude and adjusted modified log-Poisson repeated measures regression models with generalized estimating equations (GEE), clustering by individual with a compound symmetry working correlation structure were fit to estimate the relative risks and 95% confidence intervals. Controlling for individual level factors, findings revealed that FSWs with more network members who were drug users (≥3) and more network members who were sexual partners and also drugs users (≥2) were 8.89 (95% CI: 2.62, 30.22) and 6.08 (95% CI: 1.20, 30.92) times more likely to engage in daily drug use compared to women with small drug and sex and drug networks. Study findings demonstrate the role high risk networks have on risk behaviors. Results may be used to inform interventions that focus on modifying negative social ties, creating and/or improving existing positive support relationships, and integrating drug use harm reduction promotion within HIV treatment programs.

Introduction

Although engagement and retention in Human Immunodeficiency Virus (HIV) care and treatment is a critical issue across populations, it is of particular importance among female sex workers (FSWs) as they are disproportionately affected by HIV due to large sexual networks with the potential to bridge different risk pools (1, 2). A systematic review on the burden of HIV among FSWs from 50 low-and middle-income (LMIC) countries reported that the odds ratio for living with HIV is 13.5 times higher for FSWs compared to all women of reproductive age (1, 3), suggesting that sex work is an important factor contributing to the ongoing HIV epidemic. Despite their increased risk of HIV acquisition and transmission (4), FSWs are less likely to access care, have lower uptake of antiretroviral therapy (ART), and are disproportionately lost at each stage of the HIV care cascade (3, 5, 6). While FSWs living with HIV face many of the same barriers to care and treatment as other people living with HIV (PLWH), they often experience more pervasive forms of social stigma and violence which makes it harder for them to access HIV treatment services (7–9). Pooled estimates from a meta-analysis reported that only 38% (95% CI: 29%–48%) of FSWs living with HIV were active ART users (95% CI: 29%–48%) compared to 46% (95 CI: 43–50%) among PLWH globally (3, 10); and of FSWs on treatment, only 57% were virally suppressed (3). Ever ART use among FSWs living with HIV was estimated at 54% (95% CI: 21%–84%), and ART initiation rates varied across settings, ranging from 19% in Kenya to 48% in Rwanda (3). While ART adherence among FSWs was similar to global estimates for all PLWH, 76% and 62% respectively, pooled estimates for FSWs were based on 4 studies compared to 84 among the general population (3, 11).

While multiple factors may hinder successful HIV care and treatment (e.g., difficulty managing treatment due to side effects, forgetting to take medication, disrespectful and abusive treatment by service providers, fear of rejection by partner, health care policies, discrimination, food insecurity, mental illness etc.), substance use is one of the most consistent and salient barriers reported across settings and populations (12, 13). Substance use has been consistently linked to ART non-adherence, increased sexual risk behaviors, decreased health care utilization, and poorer immunologic and virologic outcomes (14, 15). Among a sample of alcohol-using PLWH for example, ART adherence ranged between 25% and 57% compared to 56% and 76% in nondrinkers (14). In their study, Hinkin et al. (16) reported that substance-using individuals living with HIV were four times more likely to be non-adherent to ART compared to those not using illegal substances. Similarly, a study on drug use among PLWH found significant differences in medication adherence between active cocaine users and nonusers, 27% vs. 68% respectively (17). Drug and alcohol consumption, needle sharing, and injection drug use (IDU) increase HIV susceptibility directly through unsafe injection practices and indirectly through mechanisms that effect decision making ability, impair perception and weaken the ability to process social cues (18, 19). Altered decision making ability can in turn reduce sexual inhibition and ART adherence.

For many FSWs, substance use is a common part of their daily lives (8, 20–22). For some women, drug dependence may have triggered their entry into sex work, others may use substances to cope with the stigmatizing and challenging life of sex work, and many are requested or required to use substances by their clients. According to a review of FSWs studies, 81.2%–100% of FSWs reported ever using alcohol and 73.3%–74.8% reported current use (23). Prevalence of injection drug use (IDU) ranged from 32% in Hanoi, Vietnam, 56% in Vancouver, Canada, 14% and 22% in Ciudad Juarez and Tijuana, to 82% in Amsterdam (24).

The majority of research on drug use and FSW populations have examined drug use as a risk factor for primary HIV infection (17, 25). FSWs who use illicit drugs have an increased risk of acquiring HIV compared to FSWs who are non-drug users. A study among FSWs along the US-Mexico border, for example, reported a two-fold increase in HIV prevalence among FSW-IDUs compared to non-IDU FSWs (24). According to the study authors, FSWs who injected drugs had significantly different and higher risk profiles of socio-demographic and sexual risk behaviors relative to non-IDU FSWs, which may be one explanation for the markedly different HIV prevalence.

There has been little research on drug use (especially non-injecting drug use) among FSWs living with HIV. Among the few studies, findings indicate a correlation between drug use and poor treatment adherence. In their study among FSWs living with HIV in the Dominican Republic (DR), Donastorg et al. (26) found that FSWs who used drugs were more likely to experience treatment interruption and had worse HIV clinical outcomes compared to non-drug users. Another study in Canada reported that FSWs living with HIV who used drugs had a reduced odds of viral suppression compared to their non-drug using counterparts (27). These findings support the evidence that drug use is an important determinant of HIV care and treatment behaviors and outcomes.

Although individual-level characteristics likely explain a substantial portion of variability in drug use behavior, accumulating evidence has demonstrated that differences in health behaviors and outcomes cannot be fully explained by individual-level factors alone (28–30). Previous explorations of substance using populations have identified social networks as a key factor in perpetuating alcohol and drug use and associated risk behaviors (19, 29, 31–33). Network characteristics have been associated with heavy episodic drinking, drug use (e.g., sharing of needles), sexual risk (e.g., multiple sex partners), HIV testing, and HIV medical care. Having more network members who use drugs, for example, has been associated with increased sexual risk, unsafe injection practices, and frequency of drug use (34, 35). A network study among women found that having more network members who used heroin or cocaine was associated with increased odds of engaging in a risky sexual partnership and having more than 2 sexual partners (36). In their study among substance using women, Davey’s et al. found that women who had a higher number of network members who they drank alcohol with were more likely to be heavy episodic drinkers (37).

Identifying network properties that create or sustain health risks is essential for epidemiologic surveillance and developing interventions that modify the social environments that shape risk behavior. Information on network structure and function has been used effectively to reduce HIV-related risk behaviors by restructuring network dynamics and altering group norms (38, 39). For example, a network-based intervention study among IDUs reported that breaking social ties with network members who engaged in IDU was associated with long-term abstinence among index participants (18). Similarly, interventions focused on improving social support networks for PLWH reported decreased substance use as well as long-term engagement in HIV care and treatment (40).

While there is a large body of research documenting the association between social network properties and HIV risk and other behaviors among key populations like MSM and IDUs, there have been few network studies among FSWs living with HIV. A systematic review from 2016 reported 19 studies that investigated interpersonal relationships among FSWs and all but 4 focused on the role of social support in relation to condom use (41). Furthermore, most used a cross-sectional approach with questions about social support and did not collect social network data using traditional social network methods (i.e., which includes generating a list of network members and then soliciting information about each member’s attributes and relationships with other network members).

Barrington et al. (42) conducted a personal network study among male partners of FSWs living with HIV in the DR and examined associations between network norms, characteristics and consistent condom use, but FSW network dynamics were not explored. A study examining peer influences on injection drug use among FSWs found that FSW-IDUs were more likely to have relationships that perpetuated drug use; yet, the question of which social networks (i.e., types of networks) and what social network characteristics (e.g., size, composition, density, etc.) were associated with drug use was not explored (22).

The aim of this study was to describe the first order social network characteristics of FSWs living with HIV in the DR and assess the relationship between risk network composition and daily drug use. As there is limited empirical information on FSWs living with HIV who use drugs, delineating their network profiles may be useful for strategies that seek to change social environments and mitigate drug use. This in turn, may lead to better HIV clinical care outcomes and reduced risk of transmission to sexual partners.

Methods

Setting

The DR is one of the largest sex tourism destinations in the Caribbean with an estimated 100,000 women involved in the sex industry (43). Sex work is not explicitly illegal in the country for people over the age of 18. Historically, the majority of sex work was establishment-based, but recent estimates suggest that more than 60% of FSWs independently solicit clients from streets, parks, beaches or other public places. FSWs who are establishment-based tend to work in brothels, bars, discos, liquor stores, or “car washes” that function as outdoor dance halls. Even though sex work is not illegal in the country, harassment by police and other law enforcement officials is common (44, 45).

Even though the country has made significant gains in HIV risk reduction among the general population, HIV prevalence among the most vulnerable populations remain disproportionately high. HIV prevalence estimates for FSWs range between 1.7–6.3% compared to 0.7% among the general population (46, 47). The DR began large-scale provision of antiretroviral therapy (ART) in 2003, but national estimates of HIV clinical care and treatment outcomes stratified by key populations are not available. However, findings from a cohort study among FSWs living with HIV in Santo Domingo, reported that less than half of the women in the study sample were virally suppressed despite being on treatment, and the odds of having a detectable viral load were 3.09 times (95% CI: 1.44 to 6.59) higher in women who reported ART interruption (48). Additionally, drug use was a significant predictor of ART interruption and detectable viral load in adjusted models.

The DR in particular, has been recognized as a central player in the transnational transport of drugs (49). According to the National Drug Control Directorate, cocaine confiscations increased by approximately 50% between 2010 and 2011, suggesting an increase in local drug availability (50). Furthermore, the drug-scene in the country is amplified by the large sex tourism economy. Nonetheless, research on illicit drug use and substance use health services in the country are extremely limited. The main illicit drugs consumed in the country are powder and crack cocaine and marijuana, and to a much lesser extent heroin (51). Under the current Dominican drug law (Law 50–88), anyone found in possession of illegal drugs may be classified as a narcotrafficker and can be sentenced to years in prison (52). Government investment in drug treatment and harm reduction programs is low, with most rehabilitation programs provided by non-governmental organizations (NGO) and focused on abstinence. The country has no formal drug harm reduction policy and all spectrums of drug use are heavily stigmatized and criminalized. Despite the evidence that drug use contributes to the concentrated HIV epidemic among key populations, including sex workers, integration of drug prevention and treatment services within HIV policies are largely non-existent.

Study design

The present study employed a micro-longitudinal observational study design and was nested within an ongoing five-year (2016–2021) NIH-funded parent study (5R01MH110158) in the DR (53). The parent study is a prospective longitudinal observational cohort study with 200 FSWs living with HIV in Santo Domingo. Participants were recruited for the parent study using a hybrid sampling approach through peer referrals and FSW peer navigators associated with the in-county community partner, MODEMU (Movimiento de Mujeres Unidas), a community-driven organization and network of sex workers. Further details on the parent study can be found elsewhere (see Kerrigan et al. 2020) (54).

Data collection activities for the present study included a mobile health (mHealth) daily behavior diary collected for 7 days, an interviewer-administered social network survey at study enrollment, and secondary data from the parent study. Data collection instruments and measures were piloted with 5 FSWs, translated to Spanish, and adapted to the Dominican context. Written informed consent was obtained from all participants. Study enrollment was held at the Instituto Dermatológico Dominicano y Cirugía de Piel (IDCP) in Santo Domingo where the parent study is located. Ethics approval from the Internal Review Boards (IRB) at Tulane University and IDCP was obtained.

Study participants and recruitment

Women were eligible to participate in the study if they met all the parent study’s inclusion criteria (which included being at least 18 years of age, having a confirmed HIV positive diagnosis (determined by a single rapid test), and having exchanged sex for money in the month prior to study enrollment), and had used drugs in the past 6 months [required for half the sample] and were willing and able to answer electronic daily behavior diary questions for 7 days. Participants were recruited from the parent study using selective/purposive sampling based on drug use. Data from the parent study were analyzed to determine drug users and non-drug users, with drug use defined as having consumed drugs in the 6 months prior to data collection. Two lists were created for recruitment purposes, drug users and non-drug users.

Among the 200 participants in the parent study at baseline, 37% had ever used drugs and 16% were current drug users. For the present study, drug users were randomly sampled from the 16% of current drug users in the parent study. Non-drug users were randomly selected based on viral load detectability (>400mL). Non-drug users were categorized as viral load detectable and non-detectable, and every 5^th participant from each group was selected as a potential participant. Based on power and sample size calculations, a minimum sample size of 50 (equivalent to 350 observations in a repeated measures study design) was determined to have sufficient power at 80% and an alpha of 0.05 to detect significant differences between drug users and non-drug users.

FSW peer navigators contacted, confirmed eligibility and recruited participants. Participants received $10 USD for participating in the study and an additional $1.5 per day for completion of the online daily diary over the 7-day data collection period (1 diary a day x 7 days= $10.5). Transportation to and from the study site was covered for 2 trips. At the time of enrollment, cellphones were loaded with a pre-paid 7-day data package to cover Internet costs for the daily diary survey.

Data collection procedures

Social network data

Social network data was collected as ego networks. An ego network analysis collects information from the perspective of the participant, referred to as the ‘ego’, about each member, referred to as the ‘alter’, in the network, and the ‘ego-alter’ relationship. Egos are also asked to report on whether relationships exist between each alter to characterize the wider network. Ego network data was solicited from each participant based on a modified version of the Personal Network Inventory which collects information about each member of the participant’s social network (32). This inventory has been shown to have good predictive validity and high internal consistency (32). Each participant was asked to list by first name and first letter of last name individuals who play different roles and functions in their lives such as social support or sex partner. Name generating questions included: ‘During the last 6 months, whom could you talk to about things that were personal?’; ‘If you needed to borrow $25, who are the people you know who would give you $25?’; ‘Who are the people you have used drugs with in the past 6 months?’; ‘Who are the people with whom you have had sexual relations in the past 6 months?’; ‘Who are the people you have sexual relations with who are also drug users?’; ‘Who are the people you have used alcohol with in the past 6 months?’ Categories were not mutually exclusive. The maximum number of alters that could be listed was 15. Participants were asked to provide basic demographic or attribute information for each alter listed including age, gender, occupation, education, relationship type (e.g., friend, client, family, etc.), perceived HIV status, drug use, alcohol use, and whether the member is a sex partner of the participant. Participants were also asked to rate the quality (i.e., trust/closeness) of their relationship with each alter using a Likert scale ranging from 1 to 3, where ‘1= ‘very close’, ‘2=sort of close’, ‘3=not close’. Finally, participants were asked to indicate which network members know each other on a scale of 1 to 4, where ‘1= very close/know very well’, ‘2 =sort of know each other’, ‘3=acquaintance’, ‘4=don’t know each other at all’. For each participant, network and attribute data were collected using a 2 by 2 matrix where columns represented alter attributes and network data and rows represented alters.

Daily behavior diary

Daily behaviors were captured using a mobile web-based daily diary survey. The daily diary survey collected information on daily drug use (type, location of consumption, and with whom), daily alcohol use (quantity, location of consumption, and with whom), daily affect, daily violence experience, sex exchange, drugs in the environment, social support, and daily ART adherence. The survey was generated using Qualtrics Mobile Survey Platform (https://www.qualtrics.com), which provides secure, encrypted mobile technology messaging services. Prior to data collection, diary items were reviewed with participants to ascertain their comprehension of the items and to ensure consistency in the interpretation of potentially ambiguous items. Participants were sent the link to the web-based survey via SMS text message once a day in the morning with participants reporting on the previous day’s behaviors. Morning was chosen as the optimal time given that many participants work at night and allowing them to reflect on the entire evening prior to the morning survey. Participants had 24 hours to complete the survey before it became invalid. Diaries could be completed in 2–3 minutes. Daily diary completion was monitored and any participant who did not complete the diary questions was sent a reminder message via WhatsApp. Results from the online diary were automatically stored for each participant according to their unique study ID in a Qualtrics password-protected database.

Key measures

The primary outcome was daily drug use. Daily drug use was examined as a binary variable (1= ‘Yes’, 0= ‘No’) from the daily diary responses. The primary exposure variables were risk network variables operationalized as: Drug network, the number/proportion of network members who were drug users in the past 6 months, sex and drug network, the number/proportion of sex partners (paying clients and/or permanent partners) in the past 6 months who were also drug users, alcohol network, the number/proportion of network members with whom the participant drank alcohol in the past 6 months, and sex network, the number/proportion of network members with whom the participant had sexual relations in the past 6 months. Sex partners included any person with whom the ego had sexual relations in the past 6 months including paying clients and/or permanent partners. Risk network variables were examined as continuous variables and then due to non-normal distributions as binary variables cut at the 75^th percentile. Other network variables that were explored included total network size (calculated as the total number of network members named by the participant), network density (calculated based on proportion of ties that exist in a network relative to the total number of possible ties), network homophily (measured by the E-I statistic which represents the ego’s propensity to have ties with alters in the same group or class as the ego with a range of −1 to 1, where scores closer to −1 represent greater homophily on the grouping characteristic and scores closer to 1 represent greater heterogeneity on the grouping characteristic), emotional support network (calculated as the number/proportion of network members who the participant talked to about personal or private topics or sought advice from in the past 6 months), and financial support network (calculated as the number/proportion of network members who lent or gave money to the participant in the past 6 months). Network attributes that were explored included length of time known, relationship strength, number of family members, friends, sex workers, males, and HIV positive network members. Sociodemographic characteristics came from the parent study (except for age diagnosed with HIV and age started sex work). The PLWH perceived stigma scale came from the parent study and consisted of an aggregated 5-item measure with a 4-point Likert scale (54). The sex work discrimination measure also came from the parent study and was assessed as an aggregated measure with a score ranging from 0–12 with higher numbers reflecting higher self-perceived sex work discrimination (54). Daily behavior measures (e.g. daily alcohol use, daily exposure to violence, daily sex exchange, etc.) came from the daily diary responses.

Data management and statistical analysis

Data were collected and managed in different platforms. SAS 9.4 (SAS Institute Inc. Cary, NC) was used for final statistical analysis. Each participant had a .csv file with 7 days of behavior data from the daily diary. The data files were imported and appended in SAS to create one dataset with repeated measures for each participant. Network data was analyzed in E-Net (55), a social network analysis platform developed for ego network analysis. The E-Net file with network calculations was then imported to SAS and merged with daily diary and parent study data using the participant unique ID as the common linking variable.

Descriptive statistics (frequencies and means) were used to describe the population’s socio-demographics. Univariate analysis (means, standard deviations, range, and proportion of total network) was performed to characterize daily diary measures and network variables and bivariate analysis (two-sample t-test and chi-square) to explore correlations between daily diary measures and drug use and network measures and drug use. A series of crude and adjusted modified log-Poisson repeated measures regression models with generalized estimating equations (GEE) clustering by individual and a compound symmetry working correlation structure were fit to estimate the relative risk of daily drug use associated with network risk structure (56, 57). A benefit to using repeated measures is that it allows for greater statistical power because it does not rely on averaging or aggregating data for each individual and it accounts for the dependence between multiple observations per individual. Ignoring correlations between repeated measures may lead to biased parameter estimates and underestimated standard errors due to the inter-dependency of within-subject measures. Furthermore, daily measures of drug use far exceed the validity of retrospective recall where participants give a cumulative response about drug use. Control variables were modeled as fixed effects. Variables that created 10% or greater difference between the unadjusted and adjusted effects and that were not considered mediators were considered confounders and controlled for in the final model. Additionally, covariates with significant statistical associations to both the outcome and exposure in bivariate analyses were included in the regression models, in addition to variables that were not statistically significant but that are important theoretical confounders to drug use among PLWH. Standard errors, 95% confidence intervals and unless otherwise stated, a p-value <0.05 was used to define statistically significant associations. The final analytic sample was N=51 with 326 observations. Six individuals did not have a smartphone or had difficulty accessing the survey link on their phone model so they were called by the PI each morning and answered the questions via telephone with the PI completing the online survey for the participant. Sensitivity analysis was conducted to determine if key findings differed from the primary analysis. The results of the sensitivity analysis were consistent with the findings of the adjusted models.

Results

Sample demographic characteristics are presented in Table I. The mean age of participants was approximately 40 years [range 24–59]. The average age participants were diagnosed with HIV was 30 [range 16–53] and began sex work at age 20 [range 10–45]. The mean number of children was 3 [range 0–9]. Slightly more than 30% of women were in steady partnerships, 35% were separated, and 28% were single. Participants had an average monthly income of 11484.30 [range 3000–30000] Dominican pesos, equivalent to $229.69 US dollars of which 63% came from sex work. Drug users had a slightly higher monthly income than non-drug users ($246 vs. $222) but a higher percentage of drug users’ monthly income came from sex work, 71% vs. 30%. All participants reported having some education and significant differences in level of education achieved were detected by drug use status (t-value= 6.54, p<0.01). Eighty-one percent of drug users had some primary education, 13% some secondary education, and 6% some university. Forty-three percent of non-drug users had some primary education, 40% some secondary education, and 17% some university. Viral load was non-detectable in 74% of the sample. There was no significant difference in viral load detectability by drug use. Over 90% of participants were alcohol users with 100% of drug users reporting alcohol use in the past 6 months compared to 86% of non-drug users. Approximately 61% of the sample consumed 6 or more drinks per week according to the Alcohol Use Disorder Test (AUDIT) scale. On average, participants reported engaging with 6 [range 0–42] clients per week and the majority of participants, 82%, were street-base/self-employed. The mean sex work discrimination score was 3.78 [range 0–11]. Mean perceived HIV stigma score was 12.61 [range 5–18].

Table I.

Socio-demographic characteristics of FSWs living with HIV (N=51)

	Drug Users (N=16)			Non-Drug Users (N=35)			Total			Test Statistic	df
	Mean / %	SD	Range	Mean / %	SD	Range	Mean / %	SD	Range
Average age (years)	39.09	7.30	[24–59]	38.06	7.38	[24–52]	39.54	7.32	[24–59]	0.67	49
Average age diagnosed with HIV (years)	29.44	7.28	[16–43]	30.83	8.23	[18–53]	30.39	7.90	[16–53]	0.58	49
Average age started sex work (years)	18.06	4.73	[10–29]	21.20	7.47	[12–45]	20.22	6.84	[10–45]	1.54	49
Average number of children	3.36	1.95	[1–9]	2.79	1.34	[0–6]	2.96	1.54	[0–9]	−1.15	49
Current relationship status										1.12	3
Partnered	37.50%			31.43%			33.33%
Separated	37.50%			34.39%			35.29%
Single	25.00%			28.57%			27.45%
Other	0.00%			5.71%			2.00%
Average monthly income (Dominican peso)	12300.00	7814.00	[3000–30000]	11111.43	5765.91	[3000–25000]	11484.30	6421.63	[3000–30000]	−0.61	49
Average monthly income from sex work only (Dominican peso)	8718.75	7694.43	[300–30000]	6505.71	3295.00	[700–15000]	7200.00	5120.51	[300–30000]	−1.45	49
Education										6.54*	2
Primary	81.25%			42.86%			54.90%
Secondary	12.50%			40.00%			31.37%
University	6.25%			17.14%			13.73%
Viral Load (mL)										1.79	1
≤400 mL	86.67%			68.57%			74.00%
>400 mL	13.33%			31.43%			26.00%
Current alcohol use (past 6 months)	100.00%			85.72%			90.20%			2.53	1
Alcohol Use Disorder Test (AUDIT)										3.22	2
Abstainer	0.00%			14.29%			9.80%
1–6 drinks per week	25.00%			31.43%			29.41%
>= 6 drinks per week	75.00%			54.29%			60.78%
Average number of clients per week	7.38	7.35	[1–25]	5.54	7.50	[0–42]	6.12	7.43	[0–42]	−0.81	49
Location of Sex Work										0.22	1
Street-based	81.25%			82.86%			82.35%
Venue-based	18.75%			17.14%			17.65%
Sex work discriminationa	3.69	3.94	[0–11]	3.83	3.36	[0–11]	3.78	3.32	[0–11]	0.14	49
PLWH perceived stigmab	12.06	2.46	[5–15]	12.86	2.75	[5–18]	12.61	2.66	[5–18]	0.99	49

^aScored 0–12 with a higher score indicating higher sex work discrimination

^bScored 1–20 with a higher score indicating higher perceived HIV stigma.

^*p<0.01

Notes. FSW=Female sex workers; PLWH=People living with HIV; df= degrees of freedom; Mean / % based on non-missing data (<10% missing on any variable).

Table II presents means and frequency analysis from the daily behavior diary stratified by current drug use status. The total number of observations was 326 of 357 possible with a 91% response rate. The minimum number of diary days completed was 2, the maximum 7, and an average of 6 days. During the 7-day diary, 22% of the sample used drugs of which 18% reported consuming cocaine, 6% marijuana, 2% crack, and 3% other drugs (not shown in table). Among drug users, 69% engaged in drug use during the week of data collection and drugs were consumed an average of 3 days per week [range 0–7]. Cocaine was consumed on 32% of response days, marijuana 13% of response days, and crack on 4% of response days. Drug consumption occurred most frequently with friends (36.17%), clients (34.04%), and partners (12.77%) (not shown in table). Most frequent locations of drug use included household (54.34%), bar/disco (19.57%), street (10.87%) (not shown in table). Drug users were statistically more likely to have been exposed to drug environments during the week compared to non-drug users, (64%vs.37%, t-value= −2.46, p<0.001). The weekly negative affect (mood) score was 2.4 [range of 0–13]. The weekly experience with violence was low with a mean of 0.51 [range of 0–4], with drug users having a slightly higher mean than non-drug users, 0.56 [range 0–3] vs. 0.49 [range 0–4]. Sex exchange occurred an average of 3 days [range 0–7] per week. The average number of drinks per day was 2 [range 0–9]. Non-drug users consumed a higher number of drinks per week than drug users, 14.89 [range 0–40] vs.11.50 [range 0–37].

Table II.

Daily behavior diary results among FSWs living with HIV (N=51, 326 observations over 7 days)

	Drug Users (N=16)			Non-Drug Users (N=35)			Total			Test Statistic	df
	Mean / %	SD	Range	Mean / %	SD	Range	Mean / %	SD	Range
Number of diary days completed (days)	6.13	1.26	[4–7]	6.51	1.01	[2–7]	6.37	1.09	[2–7]	1.18	49
Weekly negative affect score	2.81	3.08	[0–13]	2.26	2.42	[0–10]	2.43	2.63	[0–13]	−0.70	49
Weekly experience with violence	0.56	0.96	[0–3]	0.49	1.01	[0–4]	0.51	0.99	[0–4]	−0.26	49
Average number of days sex exchanged	2.50	2.25	[0–7]	3.03	1.98	[0–7]	2.86	2.06	[0–7]	0.85	49
% of response days sex exchanged	40.79	33.30	[0–100]	46.34	29.04	[0–100]	44.74	30.23	[0–100]	0.60	49
Number of drinks per week	11.50	10.00	[0–37]	14.89	11.51	[0–40]	13.82	11.08	[0–40]	1.01	49
Number of drinks per day	2.02	2.96	[0–8]	2.38	3.19	[0–8]	2.27	3.12	[0–9]	0.92	308
Average number of days drugs consumed	2.75	2.44	[0–7]	--	--	--	0.86	1.85	[0–7]	--	--
% of response days drugs consumed	47.00	37.18	[0–100]	--	--	--	15.07	29.92	[0–100]	--	--
% of response days marijuana consumed	12.50	29.91	[0–100]	--	--	--	3.92	17.38	[0–100]	--	--
% of response days cocaine consumed	32.34	36.50	[0–85.71]	--	--	--	8.97	23.07	[0–85.71]	--	--
% of response days crack consumed	3.75	15.00	[0–60]	--	--	--	1.18	8.40	[0–60]	--	--
% of response days other drugs consumed	3.39	10.39	[0–20]	--	--	--	0.67	3.41	[0–20]	--	--
% of response days exposed to drug environment	63.54	35.08	[0–100]	36.72	36.50	[0–100]	45.14	37.85	[0–100]	−2.46**	49

Notes. FSW=Female sex workers; df= degrees of freedom; Mean / % based on non-missing data (<10% missing on any variable).

^**p <0.001.

Table III displays social network characteristics of study participants by drug use status. A total of 579 alters were listed by participants, 177 by drug users and 402 by non-drug users. The homophily E-I statistic at the sample-level (N=51) was −0.511 [SD: 0.52] on a range of −1 to 1 on the grouping variable of drug use. Average network size among drug users was 11.06 [range 7–15] and 11.49 [range 5–15] among non-drug users. Network density was slightly lower among non-drug users than drug users (0.50 vs. 0.57). Both groups had a high proportion of males in their networks (0.47 and 0.52), followed by friends (0.36 and 0.31) and then family (0.23 and 0.29). Participants who were drug users had a significantly higher proportion of network members who used drugs compared to non-drug users (0.46 vs. 0.12, chi-square=−5.90, p<0.05). Of all members listed by drug users, 73% were also drug users. Among drug network members 62% were male, 41% clients, and 38% friends. The proportion of network members who were drinking buddies was similar both groups (0.81 vs. 0.84). Participants who used drugs had a smaller proportion of network members who provided social (0.39 vs. 0.47) and economic (0.18 vs. 0.27, chi-square=2.65, p<0.05) support than non-drug users. Drug users had a significantly higher proportion of network members who were sex partners and drug users than non-users (0.21 vs. 0.07, chi-square= −3.59, p<0.05). Average length of time egos reported knowing alters was 12.52 years [range 1–46] for drug users and 14.30 years [range 1–56] for non-drug users. Average strength of relationship to alters was 1.88 for drug users and 1.77 for non-drug users on a scale of 1 to 3 with ‘1’ extremely close and ‘3’ not close at all.

Table III.

Social network characteristics of FSWs living with HIV (N=51)¹

	Drug Users (N=16)				Non-Drug users (N=35)				Test Statistic	df
	Mean	SD	Range	Average proportion of total	Mean	SD	Range	Average proportion of total
Network degreea	11.06	2.74	7–15	1.00	11.49	3.33	5–15	1.00	0.44	49
Network densityb	0.50	0.32	0–1	--	0.57	0.24	0–1	--	0.91	49
Network E-I statisticc (drug)	0.08	0.32	−0.5–0.8	--	−0.77	−0.68	−1–0.07	--	−6.13**	49
Drug network	5.13	2.47	1–9	0.46	1.37	1.93	0–8	0.12	−5.90**	49
Use drugs with network	2.50	1.90	0–7	0.23	--	--	--	--	--	--
Alcohol network	9.00	2.97	5–15	0.81	9.66	3.69	2–15	0.84	0.53	49
Social support network	4.31	2.18	1–8	0.39	5.37	2.91	1–13	0.47	1.30	49
Economic network	1.94	1.57	0–6	0.18	3.11	1.43	0–6	0.27	2.65*	49
Sex network	3.94	2.08	1–9	0.36	4.00	1.83	1–8	0.35	0.11	49
Sex and drug network	2.38	1.88	0–6	0.21	0.80	1.23	0–6	0.07	−3.59**	49
No. HIV-positive in network	1.69	2.15	0–8	0.15	2.11	2.39	0–9	0.18	0.61	49
No. family in network	2.56	2.28	0–9	0.23	3.31	2.07	0–9	0.29	1.17	49
No. friends in network	3.94	2.41	0–10	0.36	3.51	2.28	0–9	0.31	−0.60	49
No. sex workers in network	2.81	2.17	0–8	0.25	1.97	2.08	0–7	0.17	−1.32	49
No. neighbors in network	0.56	0.96	0–3	0.05	1.09	2.29	0–12	0.09	1.15	49
No. male in network	5.19	2.61	2–11	0.47	6.02	2.55	2–11	0.52	1.08	49
Length of time known (years)	12.52	12.53	1–46	--	14.30	13.53	1–56	--	1.30	49
Relationship strength to altersd	1.88	0.81	1–3	--	1.77	0.78	1–3	--	−0.98	49

¹Categories of relationships are not mutually exclusive;

^aDegree, total network size

^bDensity, proportion of ties that exist in a network relative to the total number of possible ties

^cE-I statistic, ego’s propensity to have ties with alters in the same group or class as self with scores closer to −1 representing greater homophily on the grouping characteristic and scores of 1 representing greater heterogeneity on the grouping characteristic

^dScore range from 1 to 3 with ‘1=very close’ ‘2=sort of close’ ‘3=not close’.

^*p<0.01

^**p<0.001.

Notes. FSW=Female sex workers; df= degrees of freedom; Means based on non-missing data (<10% missing on any variable).

Crude relative risks and 95% confidence intervals from the repeated bivariate measures analysis of risk network variables regressed on daily drug use are presented in Table IV. Participants with a higher number of drug users in their network were 10.35 (95% CI: 2.22, 48.30) times more likely to consume drugs compared to those with fewer drug using members in their network. Similarly, participants with a higher number of sexual partners who were also drug users were 4.30 (95% CI: 1.12,16.47) times more likely to consume drugs compared to those with fewer drug using sex partners. The other risk network variables were not statistically associated with daily drug use.

Table IV.

Crude relative risks and 95% confidence intervals (CI) between daily drug use and network risk variables (N=51, 326 observations)

	Relative Risk	95% CI	P-Value
Sex network (75th percentile)a	1.62	0.42, 3.70	0.379
Alcohol network (75th percentile)b	0.78	0.26, 2.32	0.653
Drug network (75th percentile)c	10.35	2.22, 48.30	0.003
Sex and drug network (75th percentile)d	4.30	1.12, 16.47	0.033

^aFive or fewer vs. six or more

^bFive or fewer vs. six or more;

^cTwo or fewer vs. three or more

^dOne or fewer vs. two or more.

Table V and VI present results from fully adjusted multivariable models. After controlling for age, age diagnosed with HIV, age started sex work, income, education, perceived PLWH stigma, sex work discrimination, daily alcohol use, daily negative affect, daily violence, and daily sex exchange, participants with more network members who were sexual partners and also drug users were 6.08 (95 CI%: 1.20, 30.92) times more likely to engage in daily drug use compared to those with networks with 1 or fewer sexual partners who were also drug users. Drug network size was also significantly associated with increased risk of daily drug use in fully adjusted models. Participants with larger drug networks were 8.89 (95% CI: 2.62, 30.22) times more likely to engage in daily drug use than those with smaller drug using networks (2 or fewer drug using members).

Table V.

Generalized Estimated Equations (GEE) Repeated Measures Analysis: Adjusted relative risks and 95% confidence intervals (CI) between daily drug use and sex and drug network size (N=51, 326 observations)

	Relative Risk adj	95% CI	P-Value
Sex and drug networka	6.08	1.20, 30.92	0.030
Age (years)	0.95	0.88, 1.03	0.188
Age diagnosed HIV (years)	0.99	0.90, 1.08	0.814
Age started sex work (years)	0.98	0.93, 1.03	0.381
Income (Dominican peso)	1.00	1.00, 1.00	0.020
Education	0.21	0.08, 0.51	0.001
Perceived PLWH stigma	0.36	0.08, 1.58	0.173
Sex work discrimination	0.82	0.38, 1.76	0.605
Daily alcohol use	2.68	1.30, 5.53	0.008
Daily negative affect	1.17	0.46, 2.95	0.745
Daily violence	1.13	0.83, 1.54	0.445
Daily sex exchange	1.23	0.75, 2.00	0.404

^aOne or fewer vs. two or more.

Notes. Adj= adjusted, controlling for potential confounders.

Table VI.

Generalized Estimated Equations (GEE) Repeated Measures Analysis: Adjusted relative risks and 95% confidence intervals (CI) between daily drug use and drug network size (N=51, 326 observations)

	Relative Risk adj	95% CI	P-Value
Drug networka	8.89	2.62, 30.22	0.005
Age (years)	0.96	0.86, 1.06	0.377
Age diagnosed HIV (years)	1.02	0.96, 1.09	0.545
Age started sex work (years)	1.00	0.98, 1.03	0.914
Income (Dominican peso)	1.00	1.00, 1.00	0.061
Education	0.29	0.14, 0.60	0.001
Perceived PLWH stigma	0.46	0.09, 2.33	0.345
Sex work discrimination	0.52	0.23, 1.19	0.123
Daily alcohol use	2.53	1.26, 5.09	0.009
Daily negative affect	0.59	0.26, 1.32	0.199
Daily violence	1.12	0.86, 1.45	0.413
Daily sex exchange	1.40	0.91, 2.15	0.126

^aTwo or fewer vs. three or more.

Notes. Adj= adjusted, controlling for potential confounders.

Discussion

This study examined the ego network characteristics of FSWs living with HIV in the DR and assessed the role that network risk characteristics had on daily drug use. Analysis of structural network data suggests that drug using and non-drug using FSWs have similar network size and density. Results from two-sample difference of means t-tests revealed significant differences between drug users and non-drug users in network homophily on the grouping variable of drug use, drug network size, financial support network size, and sex and drug network size. Statistically significant associations in fully adjusted models between drug network size, sex and drug network size, and daily drug use were detected. Having more social network members who used drugs within the past 6 months and more sexual partners who also used drugs within the past 6 months was associated with an increased risk of daily drug use net of the effects of sociodemographic and behavioral characteristics. As the number of drug-using social network members increased-- 3 or more -- participants were 8 times more likely to engage in daily drug use. In addition, having 2 or more sexual partners who used drugs in the network increased the likelihood of drug use 6-fold.

These results corroborate other network studies that have demonstrated that network composition can be a risk factor for engaging in HIV-related risk behaviors including drug use, sharing needles, and exchanging sex for drugs (58, 59). Similar to our findings, in their study on predictors of illicit drug use among adults, Schroeder et al. (58) reported that having drug users in the social network was the most salient factor associated with continued drug use (OR: 5.70, P<0.0001). In their study on adult drug users, Williams and Latkin (59) similarly found that network attributes were important influences of drug outcomes, such that drug use was 8.5 times higher for individuals with more drug-using ties and connection to daily users than those with less drug influence networks.

There is a large body of social network literature documenting that the behaviors of ego’s alters--whether that be excessive drinking, risky sexual partnerships including unprotected sex and number of sexual partners, and drug use-- are influential determinants of the ego’s own behaviors and attitudes (34, 60, 61). It is also likely that linkages among drug users exist because individuals seek out others who support their drug use habits. In an environment where norms promote or accept drug use, increased drug use may occur. Additionally, many sex workers who use drugs are requested or required to consume drugs by clients and sexual partners before engaging in a sex act (2, 62). As such, women who have more clients in their networks who are drug users are more likely to consume drugs. In this sample, participants who reported using drugs in their daily diary did so most frequently with clients and friends, which aligns with the sex and drug network and daily drug use finding.

The study findings should also be considered within the broader spectrum of HIV disease management. As demonstrated by the literature, drug use is common in the sex work industry and an established determinant of ART non-adherence; and the results of this study suggest that social network composition may be an important factor of drug use for FSWs living with HIV. As such, programs aimed at improving ART adherence and engagement and retention in HIV care among key populations should consider targeting the social microenvironment within the context of drug use harm reduction and HIV treatment. The findings are also informative for interventions targeting other key populations at risk for or living with HIV as they demonstrate the important role high-risk networks have on endorsing and perpetuating high-risk behaviors.

This study had several limitations. First, the small sample size likely influenced the large confidence intervals and precision of effect estimates. Second, because measures were self-report, the study is subject to information bias, particularly social desirability, which may threaten internal validity. Validated measurements and multiple modes of data collection were used to minimize misclassification bias. However, given the stigma associated with HIV, sex work, and drug use in the DR, it is likely that some measures (e.g., drug use, sex exchange, etc.) were underreported. Although mean completion rate of daily diaries was similar for both groups, it was slightly lower for drug users. It is possible that daily drug use was underreported by drug users due to recall issues from being under the influence, biasing the effect estimates towards the null. Third, a common critique of ego network data is that measures are based on the ego’s perception of their networks and it is difficult to know how well this reflects the actual behaviors of the network members. Nonetheless, risk perception studies have shown that an individual’s perception of their network’s behavior can be just as influential as the actual behavior of their network members (63). Density measures, however, are particularly prone to error given that they are calculated based on the ego’s knowledge of alter-alter ties. Additionally, although the diary data was quasi longitudinal, the reporting period (1 week) plus the cross-sectional network data is not sufficient for determining a causal relationship between network characteristics and drug use. Longitudinal network studies are needed to determine if social network composition predicts drug use, and to examine how changes in network composition may influence behavior change. Finally, generalizability of results (external validity) is limited considering that study participants were FSWs living with HIV receiving HIV treatment services and from the DR, and thus, not representative of all FSWs living with HIV.

Despite the limitations, this study has several strengths. The study adds to the limited body of network-based research among FSWs, in particular FSWs living with HIV in the context of a LMIC. Collecting and analyzing social network data of FSWs living with HIV provides a detailed picture of the social environments in their daily lives, and how such relationships and interactions may contribute to drug use behavior. Through social network analysis we captured a more in-depth understanding of what FSW social connections look like outside of just their work environment.

Study findings can be used to inform network-based interventions that promote harm reduction among network members. While interventions that focus on modifying social ties with drug users to facilitate reductions in illicit drug use have had some success (35), such interventions have yet to be employed among sex worker populations. Given that sex work environments and drug use environments overlap, interventions focused on breaking drug use ties may not be very effective among this population. Network-based interventions focused on diffusion of targeted information and resources through the use of peer educators may be a more effective strategy to decrease drug use among FSWs. Peer education interventions where the peer educator focuses on modeling positive health behaviors have been successful at changing risk behaviors among risk network members (38). Another approach is to focus on strengthening social support networks. While there was no association between social support networks and drug use in this study, other studies have shown that social support networks are important protective factors against risk behaviors and in some cases may significantly decrease the odds of drug use (64, 65).

Given the success of the mobile phone daily diary, the frequent mobility of sex workers--which can be a barrier to in-person interventions, and the social stigma of drug use, an mHealth intervention that provides online virtual support for substance users could be a promising option. The mobile application could include meditation exercises, educational content on substance use and HIV, and coping methods for dealing with daily stressors that may trigger substance use. The application could also include a map of ‘safe’ spaces where sex workers can go to talk openly about their addictions and bond with others who face similar struggles. By fostering new relationships and strengthening social connections with other sex workers, drug users may naturally modify their social ties and be supported to reduce drug use. Previous sex worker multilevel interventions demonstrate the power that social cohesion and group solidarity can have on changing health behaviors (66). Nonetheless, due to the social stigma and criminalization of drug use, sex workers who use drugs may not disclose their addictions for fear of further marginalization by non-users and service providers. Future research should explore the degree to which drug users participate in ongoing community mobilizing activities and consider how HIV interventions could be adapted to better integrate substance use testing and treatment services and increased social support related to drug use.

Conclusion

This study adds to the limited research on social networks and drug use among FSWs living with HIV in a LMIC context. Findings indicate that risk network profiles are significant predictors of daily drug use, highlighting the important role that social relationships have on risk behaviors in this population. In an environment where sex work and drug use overlap, network-based interventions that emphasize modifying social ties with network members who use drugs, especially if they are clients, may not be feasible. Rather, interventions that focus on substance use harm reduction promotion, as well as increasing HIV treatment programs, coping skills, and creating and/or improving existing positive support relationships are recommended.