Secondary Logo

Journal Logo

Supplement Article

Injection-Related Risk Behavior and Engagement in Outreach, Intervention and Prevention Services Across 20 US Cities

Mackesy-Amiti, Mary E. PhD*; Boodram, Basmattee PhD*; Spiller, Michael W. PhD; Paz-Bailey, Gabriela MD, PhD; Prachand, Nikhil MPH; Broz, Dita PhD; for the NHBS Study Group

Author Information
JAIDS Journal of Acquired Immune Deficiency Syndromes: July 1, 2017 - Volume 75 - Issue - p S316-S324
doi: 10.1097/QAI.0000000000001406



Of the estimated 16 million persons who inject drugs (PWID) worldwide, 1.3 million reside in the United States.1 Injection drug use is a major risk factor for acquiring HIV and hepatitis C virus (HCV) infections through syringe and other paraphernalia sharing.2 Although lifetime PWID comprise an estimated 2.6% of the US population aged 13 years or older, they account for 22% of all persons living with HIV infection3 and about 70% of the estimated 30,000 new HCV infections per year.4,5 Prevention of these infections among PWID is critical to reduce ongoing transmission, morbidity, and mortality. Moreover, given the wide geographic variability in HIV4 and HCV6 in US regions, it is imperative to identify groups at heightened risk and to inform targeted interventions. Specifically, the estimated number of HIV diagnoses attributed to injection drug use is highest in the South compared to other regions,7 yet this region continues to have the highest numbers of people without health insurance and other social and policy factors that may limit access to much-needed health and prevention services.8,9 Understanding regional differences in risk behaviors and access to prevention services among PWID is a key to meet the goals of the US National HIV/AIDS Strategy10 to strengthen HIV prevention efforts in communities where HIV is most heavily concentrated and to expand a combination of effective, evidence-based, and scalable approaches.

To effectively respond to HIV and HCV among PWID, a comprehensive, multilevel prevention strategy is needed that includes access to sterile injection equipment, substance use disorder treatment and medication-assisted treatment, HIV and HCV testing and counseling, and sexual and injection risk reduction.11,12 In addition, identifying specific subgroups of PWID with distinct patterns of high-risk behaviors will inform targeting of HIV/HCV prevention interventions.

Latent class analysis (LCA) is a method for empirically identifying distinct patterns or subtypes, based on a set of observed categorical variables.13,14 LCA has been applied in public health research to identify subgroups, or “classes,” defined by distinct response patterns of risk behaviors. Further analysis can be conducted to identify correlates or predictors of these classes. LCA has been used to evaluate patterns of substance use,15,16 sexual behavior,17,18 injection risk behavior,18,19 and other behaviors. In this study, we used LCA to identify distinct patterns of injection-related risk behavior in a large, multisite sample of PWID and to examine how participation in prevention and intervention programs was related to risk behavior. Additionally, we examined these relationships by geographic region because of regional differences in public health policy and the availability of prevention services.


Sampling Method

We used data from the 2012 National HIV Behavioral Surveillance (NHBS) surveys of PWID in 20 US cities (n = 10,171). Methods for NHBS among PWID are described in detail elsewhere.20 Briefly, NHBS was conducted in 20 large US cities within metropolitan statistical areas with >500,000 residents. NHBS was approved by the institutional review boards at each of the participating sites. No personally identifying information was collected during enrollment, interview, or HIV testing.

Participants were recruited using respondent-driven sampling (RDS).20,21 Eligible seeds (ie, initial recruits identified during the formative research phase)22 recruited up to 5 PWID to complete the survey, who in turn recruited others. Guidance for recruiting initial seeds was provided to all NHBS cities to ensure standardization. The recruitment process was tracked using a coded coupon system and continued until the sample size was reached or the sampling period ended. The RDS sampling period ranged between 2 and 5 months. Participants who agreed to recruit others received an incentive of $10 for each recruit who completed the interview. Persons were eligible to participate and recruit if they injected drugs in the past 12 months and were aged ≥18 years, current residents of the city, able to complete the survey in either English or Spanish, and able to provide informed consent. Drug injection in the past 12 months was confirmed by observing physical evidence of recent injection (eg, track marks) and by assessing knowledge of injection practices.

Data Collection

Trained interviewers obtained informed consent and conducted face-to-face interviews, which took approximately 40 minutes and consisted of questions about participants' demographic characteristics, HIV testing history, sexual and drug-use behaviors, and use of HIV prevention/intervention services and programs. All participants were offered an anonymous HIV test and, for those who consented, testing was performed by collecting blood or oral specimens for either rapid testing in the field or laboratory-based testing. A nonreactive rapid or laboratory-based screening test result was considered HIV-negative; a reactive test result was considered HIV-positive if confirmed by Western blot or indirect immunofluorescence assay. Incentive amounts were determined locally and were in line with similar study types and designs and commensurate with the amount of time required of participants. Participants received $20–$30 for the interview and an additional $10–$25 for HIV testing.


Sociodemographic, Drug Use, Health Status, and Network Variables

Sociodemographic measures included region, gender, age, race/ethnicity, self-reported sexual orientation, and homelessness (past year and current). Other variables included self-reported HIV and HCV status, types of drugs injected in the past 12 months, male–male sex, and participation in substance use treatment. Network size was assessed by asking how many people the participant knew that injected drugs and whom they had seen within the past 30 days.

Injection Risk Behaviors

Injection risk behaviors were measured using responses to 5 questions, including 2 questions about receptive syringe sharing and 3 questions about sharing of other equipment. Participants were asked “In the past 12 months, with how many people did you use a needle after they injected with it?” Responses were recoded into 3 categories: 0, 1, and >1. Frequency of receptive syringe sharing was assessed by asking, “In the past 12 months, how often did you use needles that someone else had already injected with?” Responses were on a 5-point Likert scale (never, rarely, about half the time, most of the time, or always). Similar items measured the frequency of using shared cookers, cotton filters, and rinse water. Responses on these items were recoded into 3 categories: never, rarely, and about half the time or more.

Prevention and Intervention Participation

In 2 separate questions, participants were asked if they had obtained needles from a needle exchange program in the past 12 months and if they had received free sterile needles from any source in the past 12 months. These items were combined to form a measure of participation in syringe services programs (SSPs). Participation in intervention programs was measured with the following questions: (1) “In the past 12 months, have you had a one-on-one conversation with an outreach worker, counselor, or prevention program worker about ways to prevent HIV? Don't count the times when you had a conversation as part of an HIV test;” (2) “During those one-on-one conversation(s), did you: Discuss safe drug-injecting practices? Practice how to prepare for safe injections?” (3) “In the past 12 months, not including discussions with friends, have you been a participant in any organized session(s) involving a small group of people to discuss ways to prevent HIV?” and (4) “During these organized group session(s), did you: Discuss safe drug-injecting practices? Practice how to prepare for safe injections?” Based on these items, we created 4 variables as indicators of (1) individual didactic intervention (discussed safe injecting practices one-on-one), (2) individual practical skills intervention (practiced preparing for safe injection one-on-one), (3) group didactic intervention (discussed safe injecting practices in a group), and (4) group practical skills intervention (practiced how to prepare for safe injections in a group). Practical skills interventions always included discussion.

Data Analysis

Regional Comparisons

We compared the proportions of respondents reporting prevention/intervention participation among regions by computing corrected χ2 statistics using -svy- command in Stata (v. 13).23

RDS Analysis

Weighting of RDS samples is used to adjust for oversampling of individuals with larger networks (ie, higher probability of selection) and for differential recruitment.24,25 We computed individualized (dual component) RDS weights25 for the 5 dependent variables using the -rds- command in Stata.26 The weights were highly correlated (r = 0.99), and we used the average of the 5 weights as sampling weights in the LCA described below.

Latent Class Analysis

We conducted a LCA in Mplus (version 7)27 using the 5 measures of injection risk behavior, with complex sampling specifications including strata (20 MSAs), cluster (181 RDS chains), and sampling weights. LCA is a method for building a measurement model, similar in some ways to latent factor analysis, but the latent variable is categorical rather than continuous. The number of classes was decided based on Bayes Information Criterion and examination of bivariate residuals. Next, we tested whether the measurement model was equivalent (invariant) across regions. If the class structure varied across regions, then we would conduct analyses separately for those regions that differed. We tested measurement invariance by including direct effects of region dummy variables on latent class indicator variables,28,29 setting all equal to zero, and requesting modification indices. The modification index is the expected change in χ2 if the parameter is not constrained to be zero. Considering the sample size and the number of parameters in the model (ie, 10 thresholds per class), we set a significance cutoff of P < 0.001 (corresponding χ2 value 10.83). For each modification index above the cutoff, we tested the effect of freeing that parameter on the model fit with a Wald test and inspected the item probabilities.

Latent Class Regression Models

After establishing the measurement model, we looked at correlates of latent class using latent class regression analyses, simultaneously estimating class membership and a multinomial logistic regression of the categorical latent variable on covariates (1-step analysis).30,31 Covariates included in the first latent class regression model were demographics (region, age, gender, race/ethnicity), sexual orientation, male–male sex (for male participants only), homelessness, self-reported HIV and HCV infection status, injection network size, types of drugs injected in the past 12 months, and substance use treatment. Significant predictors (P < 0.05) were retained in the model. Individuals identified as transgender (n = 51) were excluded in the final model as sexual behavior data (ie, male–male sex) were not collected for this group. Next, in separate models adjusting for significant covariates identified in the first analysis, we examined (1) participation in SSPs and (2) participation in individual and group interventions with or without practical skills training. For each model, we computed Wald tests for the overall multinomial effect and then conducted a region-stratified analysis and computed a Wald test of equality of coefficients to test whether associations were consistent across regions. Wald tests were computed on the effects of individual and group interventions separately and in combination. Covariate associations were held equal across regions in the stratified models.



The distribution of the sample by region, demographic characteristics, and intervention participation is shown in Table 1. The South and Midwest samples were predominantly black and older (aged older than 50 years), whereas the Northeast and West samples included greater proportions of white and younger (younger than 50 years) PWID. The San Juan sample included the greatest number of PWID younger than 50 years. The percentage of men reporting male–male sex in the Midwest sample was less than 6% compared to 16%–25% in other mainland regions and 34% in San Juan. The types of substances injected also varied considerably by region, with heroin dominating in the Midwest, speedball (ie, heroin and cocaine injected together) in San Juan, and methamphetamine injection being rare except in the West. In addition, the median injection network size was smallest in the Midwest (15 persons) compared to other mainland regions (19–26 persons) and San Juan (50 persons).

Characteristics of People Who Inject Drugs in the NHBS System, United States 2012 (n = 10,171)

More than half (57%) of participants reported getting needles from a SSP. SSP use varied significantly across region (corrected χ2 = 12.07, P < 0.001); participants in the South were least likely to use an SSP. Intervention participation was similar across regions, with only slight variation in reported participation in practical skills interventions (corrected χ2 = 3.15, P < 0.05). Overall, receptive syringe sharing was reported by 33% of participants, whereas larger proportions reported sharing cookers (55%) and cottons and rinse water (46%).

Latent Class Analysis

The sample for the LCA included 10,170 PWID who completed the survey and responded to questions about injection risk behavior. (One person was excluded for missing data on all injection risk items.) The Bayes Information Criterion indicated that a 6-class solution had the best fit. The test of measurement invariance across region found that only 1 Wald test was significant (Wald = 17.60, P < 0.001); therefore, we concluded that the latent class structure was invariant across regions.

Table 2 reports on the risk behavior profiles of 6 classes identified in the analysis with 1 being the lowest risk class and 6 the highest risk class: (1) low risk with very low probability of any risk behavior; (2) low equipment with low-frequency non-syringe equipment sharing; (3) high equipment with high-frequency non-syringe equipment sharing; (4) syringe that includes low-frequency syringe sharing with low probability of other equipment sharing; (5) moderate risk syringe sharing with low-frequency equipment sharing; and (6) high risk syringe sharing with high-frequency other equipment sharing. Figure 1 reports on adjusted regional prevalence estimates and shows that, after adjusting for age, gender, and race/ethnicity, the Midwest region had the lowest prevalence of high-risk behavior, whereas the highest prevalence was observed in the South.

Posterior Probabilities of Risk Indicators by Latent Class
Regional adjusted estimates of risk class prevalence (n = 10,152). aNortheast: Boston, Nassau, New York City, Newark, and Philadelphia; South: Atlanta, Baltimore, Dallas, Houston, Miami, New Orleans, and Washington (DC); Midwest: Chicago and Detroit; West: Denver, Los Angeles, San Diego, San Francisco, and Seattle. bInjection risk classes are as follows: class 1, Low Risk: very low probability of any risk behavior (reference category); class 2, Low Equip: low-frequency non-syringe equipment sharing; class 3, High Equip: high-frequency non-syringe equipment sharing; class 4, Syringe: syringe sharing with low probability of other equipment sharing; class 5, Mod. Risk: low-frequency syringe and other equipment sharing; class 6, High Risk: syringe sharing with high-frequency other equipment sharing. Estimates adjusted for gender, age, and race/ethnicity. 18 cases missing on covariates.

Covariate Analysis

We now describe the results of multinomial regression analyses of risk class on selected variables; low-risk class (class 1) was used as the reference category in these models (see Table S1, online supplement High-risk (class 6) injection behavior was less likely among black participants than non-Hispanic white participants and more likely among women, especially women who were identified as bisexual. Also, those who were younger than 40 years (vs. 50+), homeless in the past 12 months, or reported male–male sex but did not identify as gay had higher (P < 0.01) relative odds of high-risk behavior class. Speedball and methamphetamine injection, daily injection, and larger network size also were associated with high-risk class membership. Participants who self-reported HIV infection were somewhat less likely to be in the high-risk class (P = 0.016), whereas those who reported HCV infection were more likely to be in 1 of the 3 classes that include syringe sharing. Substance use disorder treatment had no significant association (P > 0.05) with class membership and was dropped from the model. Membership in the syringe class was associated with methamphetamine injection and white race/ethnicity. In these adjusted models, participants in the South were more likely than those in other regions to belong to the high-risk class and the high equipment sharing class. Adjusted high-risk class odds for San Juan and the West and Midwest were not significantly different from those in the Northeast.

SSP Participation

Table 3 reports the effects of SSP participation on the probability of injection risk class membership by region, with the Wald test of the overall multinomial effect, and odds ratios (ORs) and 95% confidence intervals (CIs) for each contrast. Overall, participants who reported SSP use had significantly lower odds of being in any syringe-sharing class (high-risk, moderate risk, or syringe) vs. low-risk class membership. However, the effect was not uniform across regions [test of equality of coefficients between regions, Wald(20) = 33.86, P = 0.02]. The effect was significant in all regions except the Midwest, with high-risk vs. low-risk ORs ranging from 0.58 in the South to 0.32 in San Juan. Syringe program access did not have a consistent significant relationship with being in low or high equipment sharing classes vs. low-risk class membership, although the overall effect indicated higher odds of low equipment sharing (P = 0.048).

Regional and Overall Adjusted* Association Between Syringe Services Program Usage and Probability of Injection Risk Class Membership (n = 10,051)

Intervention Participation

Across all regions, there was a significant effect of participating in individual interventions with skills training [Wald(5) = 17.15, P = 0.004, see Table S2, online supplement]. Individual skills training was associated with a significantly lower likelihood of high-risk class membership (vs. low-risk, OR = 0.49, 95% CI: 0.34 to 0.71). The overall effect of group intervention with skills training was not significant [Wald(5) = 7.99, P = 0.16], although it was associated with a lower likelihood of high equipment sharing behavior (vs. low-risk, OR = 0.57, 95% CI: 0.33 to 0.99). Interventions that included only information and discussion had no significant relationship with risk class membership.

Table 4 shows the effects of intervention participation on injection class membership by region. Again, the relationships were not uniform across regions [for individual skills-based intervention, Wald(20) = 57.10, P < 0.0001; for group intervention, Wald(19) = 47.50, P = 0.0003]. Individual skills-based intervention showed significant effects in the Midwest, West, and San Juan, with exposed individuals having a relatively lower likelihood of high-risk compared to low-risk class membership. In the South, the overall test was not significant, although individual skills-based intervention was associated with lower odds of high-risk class membership. Group skills-based intervention showed significant effects in the West and San Juan. In the West region, group skills intervention participants had a relatively lower likelihood of high-risk or high equipment sharing class compared to low-risk membership. In San Juan, group skills intervention participation was associated with a greater likelihood of low-risk behavior compared to all other classes. The combined effect of individual and/or group skills-based intervention was significant in all regions except the Northeast.

Regional Adjusted* Association Between Intervention Participation and Probability of Injection Risk Class Membership (n = 10,051)


LCA identified 6 distinct classes of injection risk behavior among PWID. Adjusting for age, gender, and race/ethnicity, high-risk class membership was more prevalent in the South, whereas low-risk class was more prevalent in the Midwest region. Some of the variance in risk class prevalence between regions was attributable to differences in population characteristics, such as prevalence of male–male sex, type of drug injected, and network size. Some of the differences in proportions of high-risk and low-risk class membership between regions were lessened after adjusting for these additional variables. However, covariate adjustment only heightened the risk discrepancy between the South and other regions, indicating that elevated high-risk behavior in this region is not explained by these known factors. Moreover, use of SSPs (38%) was lowest in the South compared to other regions (52%–85%), indicating the need for more targeted interventions in this region. These findings are consistent with data reported from a survey of SSPs conducted in 201332 and the legal status of needle exchange in these states.9,33 As of 2012, there were no legal provisions for SSPs in Texas, Louisiana, Florida, or Georgia. Interventions in this region should also address the specific needs of the predominantly older, African American population who primarily inject heroin (Table 1). Minority PWID may be especially vulnerable where the legality of syringe possession is unclear or laws are enforced unequally.33

In multinomial regressions on latent class, SSP participation was significantly associated overall with lower odds of all syringe-sharing classes (high-risk, moderate-risk, and syringe only). Region-specific effects were significant for high-risk class membership in all except the Midwest region, where high-risk behavior was already least prevalent. There were no consistent significant associations between SSP participation and the probabilities of non-syringe equipment sharing classes. Sharing non-syringe equipment, such as cookers, has been previously shown to be an important risk factor for HCV transmission.34–36 However, it is likely that PWID are less aware of the risks associated with non-syringe equipment sharing, and social norms against equipment sharing are less prevalent than those against syringe sharing.37,38 Interventions, particularly those directed toward reducing HCV transmission, should address the low adherence to harm reduction practices among these classes.

Participation in intervention programs that included a skills training component was also associated with less risky behavior in all regions except the Northeast. However, the effects in the South reached significance only when individual and group intervention effects were combined. Participation in interventions consisting of discussion only had no effect on risk class. In a similar vein, a meta-analysis of 37 randomized controlled trials39 found that interventions were more likely to reduce injection drug use if they included an interpersonal skills training component for safer injection. However, another review and meta-analysis40 found little evidence of any added benefit of multisession psychosocial interventions compared to standard educational interventions for reducing injection risk behaviors. The lack of intervention effects in the Northeast and the relatively weak effects in the South compared to the strong effects in the West and in San Juan warrant further investigation. In addition, research is needed on whether interventions are effective with high-risk versus low-risk participants.19


There are some limitations of the NHBS sample. NHBS recruitment is conducted in cities with high AIDS burden, thus our findings may not reflect risk behavior patterns in lower prevalence areas. Furthermore, young suburban PWID may be underrepresented because they tend to have smaller, more isolated networks,41,42 thus our findings may not be representative of all PWID in the areas where NHBS recruited. Although RDS weights were used to adjust for oversampling of PWID with larger networks, future surveys should include enhanced efforts to recruit younger PWID.

For consistency with other reports based on NHBS data, we grouped cities into standard regions. Consequently, there may be differences between cities that are not detected in this analysis. Some regional differences might be related to data not collected in the 2012 cycle of NHBS. For example, the substance use treatment question did not ascertain type of treatment program (ie, behavior/education counseling, medication-assisted treatment). Given that several recent reviews have highlighted significant differences in treatment type effectiveness in reducing HIV and HCV risk behaviors,43,44 regional differences that might exist in the NHBS sample were not found. Furthermore, this analysis is based on cross-sectional data; intervention participation and risk behavior were both assessed for the past 12 months. Associations between intervention participation and risk behavior are not necessarily causal.


LCA can be used to identify distinct classes of injection risk behavior, providing a more nuanced and useful measure of behavior than unidimensional measures. The results of this analysis suggest that increasing access to sterile syringes should be a priority in the South, where risky injection practices are prevalent, syringe access is low, and HIV diagnosis rates are the highest in the country.45 More attention should also be given to providing skills-based interventions rather than didactic interventions for reducing injection risk behavior.


The authors thank Chicago Department of Public Health for facilitating this data agreement and the NHBS participants for their participation.

For the full list of NHBS Study Group participants, please see Supplemental Digital Content,


1. Mathers BM, Degenhardt L, Phillips B, et al. Global epidemiology of injecting drug use and HIV among people who inject drugs: a systematic review. Lancet. 2008;372:1733–1745.
2. Alter MJ. Epidemiology of hepatitis C virus infection. World J Gastroenterol. 2007;13:2436–2441.
3. Lansky A, Finlayson T, Johnson C, et al. Estimating the number of persons who inject drugs in the United States by meta-analysis to calculate national rates of HIV and hepatitis C virus infections. PLoS One. 2014;9:e97596.
4. Centers for Disease Control and Prevention. Surveillance for Viral Hepatitis—United States, 2014. Atlanta, GA: Division of Viral Hepatitis, CDC; 2016.
5. Suryaprasad AG, White JZ, Xu F, et al. Emerging epidemic of hepatitis C virus infections among young non-urban persons who inject drugs in the United States, 2006–2012. Clin Infect Dis. 2014;59:1411–1419.
6. Centers for Disease Control and Prevention. Viral Hepatitis Surveillance—United States, 2013. Division of Viral Hepatitis, CDC; 2013. Available at: Accessed July 20, 2015.
7. Centers for Disease Control and Prevention. HIV Surveillance Report, 2014. 2015. Available at: Accessed April 27, 2016.
8. Kaiser Family Foundation. The Kaiser Commission on Medicaid and the Uninsured. Key Facts about the Uninsured Population. Available at: Accessed June 15, 2016.
9. Coleman A. Needle Exchange Legality by State. 2015.Available at: Accessed December 13, 2016.
10. White House Office of National AIDS Policy. Natl HIV/AIDS Strategy United States Updated 2020. 2015. Available at: Accessed June 25, 2016.
11. Des Jarlais DC, Kerr T, Carrieri P, et al. HIV infection among persons who inject drugs: ending old epidemics and addressing new outbreaks. AIDS. 2016;30:815–826.
12. Centers for Disease Control and Prevention. Integrated prevention services for HIV infection, viral hepatitis, sexually transmitted diseases, and tuberculosis for persons who use drugs illicitly: summary of guidance from CDC and the U.S. Department of Health and Human Services. MMWR Recomm Rep. 2012;61:1–43.
13. McCutcheon AL. Latent class analysis. Vol 64. Newbury Park, CA: Sage; 1987.
14. Goodman LA. Latent class analysis: the empirical study of latent types, latent variables, and latent structures. In: Hagenaars JA, McCutcheon AL, eds. Applied Latent Class Analysis. New York, NY: Cambridge University Press; 2002.
15. Agrawal A, Lynskey MT, Madden PAF, et al. A latent class analysis of illicit drug abuse/dependence: results from the National Epidemiological Survey on Alcohol and Related Conditions. Addiction. 2007;102:94–104.
16. Harrell PT, Mancha BE, Petras H, et al. Latent classes of heroin and cocaine users predict unique HIV/HCV risk factors. Drug Alcohol Depend. 2012;122:220–227.
17. Mackesy-Amiti ME, Ouellet LJ, Finnegan L, et al. Transitions in latent classes of sexual risk behavior among young injection drug users following HIV prevention intervention. AIDS Behav. 2014;18:464–472.
18. Noor SB, Ross M, Lai D, et al. Use of latent class analysis approach to describe drug and sexual HIV risk patterns among injection drug users in Houston, Texas. AIDS Behav. 2014;18:276–283.
19. Mackesy-Amiti ME, Finnegan L, Ouellet L, et al. Peer-education intervention to reduce injection risk behaviors benefits high-risk young injection drug users: a latent transition analysis of the CIDUS 3/DUIT study. AIDS Behav. 2013;17:2075–2083.
20. Lansky A, Abdul-Quader AS, Cribbin M, et al. Developing an HIV behavioral surveillance system for injecting drug users: the National HIV Behavioral Surveillance System. Public Health Rep. 2007;122:48–55.
21. Heckathorn DD. Respondent-driven sampling: a new approach to the study of hidden populations. Soc Probl. 1997;44:174–199.
22. Allen DR, Finlayson T, Abdul-Quader A, et al. The role of formative research in the national HIV behavioral surveillance system. Public Health Rep. 2009;124:26–33.
23. Stata Statistical Software: Release 13 [computer Program]. College Station, TX: StataCorp LP; 2013.
24. Heckathorn DD. Respondent-driven sampling II: deriving valid population estimates from chain-referral samples of hidden populations. Soc Probl. 2002;49:11–34.
25. Heckathorn DD. Extensions of respondent driven sampling: analyzing continuous variables and controlling for differential recruitment. Socio Meth. 2007;37:151–207.
26. Schonlau M, Liebau E. Respondent-driven sampling. Stata J. 2012;12:72–93.
27. MPlus [computer program]. Version 7. Los Angeles, CA: Muthén & Muthén; 2012.
28. Kankaras M, Moors G, Vermunt JK. Testing for measurement invariance with latent class analysis. In: Davidov E, Schmidt P, Billiet J, eds. Cross-cultural Analysis: Methods and Applications. New York, NY: Routledge; 2010:359–384.
29. Muthen BO. Latent variable mixture modeling. In: A.Marcoulides G, E.Schumacker R, eds. New Developments and Techniques in Structural Equation Modeling. Mahwah, NJ: Lawrence Erlbaum Associates; 2001.
30. Feingold A, Tiberio SS, Capaldi DM. New approaches for examining associations with latent categorical variables: applications to substance abuse and aggression. Psychol Addict Behav. 2014;28:257–267.
31. Bandeen-Roche K, Miglioretti DL, Zeger SL, et al. Latent variable regression for multiple discrete outcomes. J Am Stat Assoc. 1997;92:1375–1386.
32. Des Jarlais DC, Nugent A, Solberg A, et al. Syringe service programs for persons who inject drugs in urban, suburban, and rural areas—United States, 2013. MMWR Morb Mortal Wkly Rep. 11 2015;64:1337–1341.
33. Burris SC, Anderson ED, Craigg A, et al. Racial disparities in injection-related HIV: a case study of toxic law. Temple L Rev. 2011;82:1263–1307.
34. Thorpe LE, Ouellet LJ, Hershow R, et al. Risk of hepatitis C virus infection among young adult injection drug users who share injection equipment. Am J Epidemiol. 2002;155:645–653.
35. Hagan H, Pouget ER, Williams IT, et al. Attribution of hepatitis C virus seroconversion risk in young injection drug users in 5 US cities. J Infect Dis. 2010;201:378–385.
36. Pouget ER, Hagan H, Des Jarlais DC. Meta-analysis of hepatitis C seroconversion in relation to shared syringes and drug preparation equipment. Addiction. 2012;107:1057–1065.
37. Davey-Rothwell MA, Latkin CA, Tobin KE. Longitudinal analysis of the relationship between perceived norms and sharing injection paraphernalia. AIDS Behav. 2010;14:878–884.
38. Cox J, De P, Morissette C, et al. Low perceived benefits and self-efficacy are associated with hepatitis C virus (HCV) infection-related risk among injection drug users. Soc Sci Med. 2008;66:211–220.
39. Copenhaver MM, Johnson BT, Lee IC, et al. Behavioral HIV risk reduction among people who inject drugs: meta-analytic evidence of efficacy. J Subst Abuse Treat. 2006;31:163–171.
40. Meader N, Ryan L, Des Jarlais DC, et al. Psychosocial interventions for reducing injection and sexual risk behaviour for preventing HIV in drug users. Cochrane Database Syst Rev. 2010:CD007192.
41. Boodram B, Mackesy-Amiti ME, Latkin C. The role of social networks and geography on risky injection behaviors of young persons who inject drugs. Drug Alcohol Depend. 2015;154:229–235.
42. Akselrod H, Grau LE, Barbour R, et al. Seroprevalence of HIV, hepatitis B virus, and HCV among injection drug users in connecticut: understanding infection and coinfection risks in a nonurban population. Am J Public Health. 2014;104:1713–1721.
43. Coffin P, Rowe C, Santos GM. Novel interventions to prevent HIV and HCV among persons who inject drugs. Curr HIV/AIDS Rep. 2015;12:145–163.
44. Macarthur GJ, van Velzen E, Palmateer N, et al. Interventions to prevent HIV and hepatitis C in people who inject drugs: a review of reviews to assess evidence of effectiveness. Int. J Drug Policy. 2014;25:34–52.
45. Centers for Disease Control and Prevention. HIV in the Southern United States. CDC Issue Brief. 2016. Available at: Accessed June 15, 2016.

injection drug use; HIV risk behavior; HIV prevention

Supplemental Digital Content

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.