Background
HIV risk reduction education is often provided in substance abuse treatment programs because of the documented association between substance abuse and HIV risk behaviors [
1‐
5]. For example, survey studies of clinics participating in the National Institute on Drug Abuse Clinical Trials Network (NIDA CTN) have indicated that most provide HIV risk reduction education [
6,
7]. Typically, these consist of single 30- to 90-minute sessions delivered in group or individual formats and are limited to providing information rather than improving motivation and teaching skills (e.g., role plays, etc.).
A number of studies have examined the effectiveness of psychosocial interventions for reducing injection and sexual risks for HIV in drug users. A meta-analysis of 35 such studies concluded that, in general, there are minimal differences identified between multi-session psychosocial interventions and standard educational interventions for both drug risks (injection) and sexual risks for HIV, though both types of interventions typically result in relatively large pre-post changes in risk behaviors [
8]. However, some evidence for single-gender groups was found in this review.
Two such gender-based studies of HIV sexual risk reduction were conducted in the context of the NIDA CTN [
9,
10]. Specifically, NIDA CTN investigators were interested in seeing if a multi-session motivational and skills training program would improve the results of the typical single-session HIV sexual risk reduction education sessions and to that end. Separate randomized controlled trials for men (CTN0018) and women (CTN0019) were conducted because the skills training components of these interventions were gender-specific, very detailed and personal.
In CTN0018, men in methadone maintenance or outpatient psychosocial treatment were randomly assigned to attend either “Real Men Are Safe” (REMAS); five sessions containing information, motivational exercises, and skills training (e.g. understanding and managing the interplay between substance use and sexual performance), or HIV education (HIV-Ed; one session containing HIV prevention information). The main outcome results of CTN0018 revealed that REMAS participants engaged in significantly fewer unprotected vaginal and anal sexual intercourse occasions (USO) during the 90 days prior to the 3- and 6-month follow-ups than HIV-Ed participants [
9]. For those who completed the REMAS program, the results were even stronger, with completers reducing their number of USO by 21% from baseline to 6-month follow-up. In contrast, HIV-Ed completers increased the number of USO by 2%.
The CTN0019 study of women also yielded positive results for the 5-session safe sex skills building (SSB) intervention (e.g. use of safer sex negotiation and risky sex refusal skills) compared to the 1-session HIV education intervention [
10]. A significant difference between the intervention groups in mean USOs was found over time, with both groups decreasing USO at 3 months but the SSB group maintaining this improvement at 6 months and the single-session HIV group returning to baseline USO levels. Women in SSB had 29% fewer USOs than those in the single-session HIV education group.
Although statistically significant results were evident in both trials, the overall effects were not large. In CTN0018, at 3 months the effect size (Cohen’s
d) was 0.10 for all subjects and 0.21 for completers. In CTN0019, the intervention group difference was not significant at 3 months, while the effect size was 0.42 at 6 months. Because these were effectiveness studies conducted in community-based clinics, there was substantial sample heterogeneity that may have reduced the overall between-group effects. For example, the samples consisted of patients in both methadone and drug-free outpatient clinics; primary drugs of abuse included cocaine, heroin, alcohol, and methamphetamine; severity of drug use varied, as did the tendency to engage in sex under the influence of drugs/alcohol; patients differed on age (about half under 40 and half over 40 years of age), race/ethnicity (about 60% European American; 40% minority), education (about 1/3 over 12 years), and monogamy status (about half currently monogamous). In fact, the effectiveness of REMAS in the CTN0018 study was found to differ significantly based on treatment setting (drug-free vs. methadone maintenance) [
11]. However, no attention has been directed at determining whether the heterogeneity of outcomes in the CTN0018 and CTN0019 studies can be attributed to patient characteristics.
Investigating moderators of intervention effectiveness has important clinical and financial implications. Although in both CTN0018 and CTN0019 the 5-session skills building intervention was more effective than a single educational session, administering a 5-session HIV intervention to all patients would raise issues of cost and patient interest/compliance. Identifying moderators of intervention effectiveness would potentially allow for matching strategy where the 5-session REMAS and SSB interventions could be targeted at those most likely to benefit. To date, there has been limited exploration of moderator variables in the CTN0018 and CTN0019 studies, largely because statistical power was limited for investigating such relationships within each study. However, preliminary analyses within the CTN0018 study data has suggested that European Americans benefitted more from the REMAS intervention than African Americans in their rates of condom use, and that none of the Hispanic men who attended the REMAS intervention were frequently using condoms with their casual sex partners [
12]. These results suggested a potential differential intervention effect for European Americans, compared to African American and Hispanic men and has led to a new version of the REMAS intervention with adaptations for using the intervention with African American and Hispanic men.
In addition to the examination of moderators, it is also useful to identify potential predictors of outcome, i.e., patient variables that are associated with outcome across both intervention groups. Information on predictors might be useful in the re-design of both types of interventions used in the CTN0018/CTN0019 studies. Alternatively, information on predictors of outcome might suggest that HIV risk reduction for those not expected to benefit from either intervention should be addressed via other means (e.g., individual counseling; referral to psychotherapist). The goal of the current study was to examine potential predictors (main effects) and moderators (interactions between predictors and intervention group) of outcomes within a pooled database of CTN0018 and CTN0019 study data. The pooled database allowed for greater statistical power for testing predictor/moderator variables and also allowed for examination of whether any moderator effects varied by gender. Two potential primary predictor/moderator variables were identified a priori: 1) severity of drug use, and 2) if the patient recently engaged in sex under the influence of drugs/alcohol. Heavy alcohol/drug use [
3,
13‐
17] and engaging in sex under the influence of alcohol or drugs [
4,
5,
18‐
21] are viewed as important risk factors for HIV. We hypothesized that individuals with a greater severity of drug use, and those who engage in sex under the influence, would have relatively worse outcomes in both intervention conditions because high levels of drug severity and sex under the influence will continue to drive risky sex behaviors following the intervention. However, we also hypothesize that these variables will be moderators.
The more intensive intervention, with a focus on sex under the influence and more time to repeatedly address patient issues, has greater potential to modify drug severity and sex under the influence, leading to better outcomes in the 5-session intervention for those with these risk factors compared to the 1-session intervention. Testing for these moderators will provide information that is useful for dissemination of the study intervention.
Secondary predictor/moderator variables also of interest included: 1) whether or not the patient was currently in a monogamous relationship (this variable was used as a covariate in the primary outcome reports [
9,
10]); 2) duration of use of the primary substance; 3) age; 4) gender; and 5) specific racial/ethnicity groups. Patients hypothesized to be at greatest risk and therefore more likely to benefit from the study intervention included those not in a monogamous relationship, long-term users, and younger patients. Gender was explored as a potential moderator due to existing research, which indicates that gender-specific HIV-prevention interventions are more successful [
22,
23]. Existing research also laid the foundation for examining race/ethnicity as a predictor and moderator. Studies [
24] have indicated that minorities have relatively worse outcomes following HIV-prevention interventions. Furthermore, preliminary examination (without statistical testing) of the CTN0018 data [
12] suggested that the more intensive 5-session intervention was relatively less effective for African Americans and Hispanics compared to European Americans, but no such difference was evident with the 1-session intervention (i.e., a moderator effect).
Discussion
The results of the analyses presented here suggest that the relative degree of reduction in unprotected sexual occasions following both a single HIV education group session and a 5-session gender-specific skills training group were dependent on certain patient characteristics. As expected, those in monogamous relationships had reduced use of condoms. This has been found in previous research [
35‐
37]. It has been suggested that having unprotected sex with a committed relationship partner who has not been tested for HIV may be a major and unrecognized source of HIV risk [
37], a situation that may be particularly true in a high-risk population such as substance abusers. HIV sex risk interventions may need to be altered to place greater emphasis on risks even in monogamous relationships. One qualification of this conclusion, however, is that the effect found here for being in a monogamous relationship was significant only for the logistic portion of the ZINB analysis.
Relatively poorer intervention outcomes were evident for younger participants and men, controlling for other predictor variables. The effect sizes for these main effect predictor findings were all moderate in size [
38], and potentially provide some guidance to clinicians about what types of individuals are at greatest risk for unprotected sex. Such individuals might be candidates for extra attention in group sessions, review of such issues in individual counseling sessions, or referral for further interventions. There were complex findings about the possible association between African American race and unprotected sex occasions. African Americans (compared to non-African Americans) were more likely, following either intervention, to engage in unprotected sex at least once. However, among those with at least one unprotected sex occasions, African Americans had fewer unprotected sex occasions (compared to non-African Americans). Since African American heterosexual women (and African American men who have sex with men), represent disproportionately large proportions of people with HIV in these risk groups [
39], future research is essential to better understand the reason for this, and to tailor interventions to decrease this disparity.
The moderator results reported here provide information on the relative benefits of the 1-session versus 5-session interventions also depend on certain patient characteristics. Most notably, Hispanic individuals did relatively more poorly in the 5-session intervention than in the 1-session intervention. This finding is consistent with the preliminary (not tested statistically) suggestion from within the CTN0018 study of less effectiveness for Hispanic individuals [
12] and raises the possibility that the 5-session intervention may not have addressed culturally specific issues, or used the most culturally relevant examples in the didactic elements of the intervention. These findings therefore support the development of culturally adapted versions of the 5-session intervention [
40]. The primary modifications of the intervention were to add modules that addressed a stronger focus on understanding how each man’s cultural and socialization experiences about sex contribute to his past and current sexual behavior. However, further research designed to understand the exact mechanism through which Hispanic ethnicity is associated with relatively poorer outcomes of the 5-session skills building intervention may be needed so that any further intervention development steps are properly targeted to the relevant causal variables. Such further research is particularly indicated given that the moderator effects found here were only in the small to moderate range using the descriptors provided by Rosenthal [
38].
An additional moderator effect occurred for recent sex under the influence. For those who engaged in sex under the influence, outcomes were relatively poor for both intervention groups. Among those who did not engage in sex under the influence, the 5-session intervention group had slightly more positive outcomes than the 1-session group. This finding is surprising given the attention paid to the topic of sex under the influence in the 5-session intervention. It should be noted, however, that in this sample, no causative link between sex under the influence of drugs or alcohol and sexual risk behavior was evident [
41]. The reasons for the lack of causative link are unclear, but it may be that such a link is difficult to detect within a sample of individuals currently receiving treatment in substance abuse treatment facilities. Thus, at least in this high risk sample, the need to reduce sex under the influence in order to increase safe sex was not apparent. However, it is possible that in other types of samples it would be important to achieve greater reduction in occasions of sex under the influence. Further intervention development work might be indicated to achieve this goal.
The fact that the 5-session intervention was particularly effective, relative to the 1-session intervention, for those individuals who had a long duration of primary drug use is notable. This finding provides some justification for the added expense and effort of clinical implementation of the 5-session intervention, at least for a subgroup of individuals receiving drug abuse treatment services in community agencies.
The current study found that age, severity of drug use, gender, monogamous relationship status, and race predicted degree of reduction in risky sexual behaviors in substance users. Other studies have obtained similar results for this population (treatment-seeking substance users) with regard to substance use outcomes, finding that age [
42‐
45]; race [
46‐
48]; and severity of use [
49,
50] were each predictors of substance use outcome. Contrary to risky sex behavior outcomes; studies that have examined gender [
51,
52] and relationship status [
53] as predictors of substance use outcomes yield inconsistent and non-significant findings, respectively. Thus, the finding for gender and relationship status found in the current study may be specific to the interventions and outcomes used herein. However, our findings for age, race, and severity of use may reflect a broader tendency for substance users to be non-compliant and/or non-responsive to a range of treatments on a range of outcomes. Further research is needed to clarify the extent to which the predictors and moderators found here are specific to the interventions and outcomes examined here. Regardless of their specificity, the present findings contribute to the potential clinical usefulness of such predictors/moderators and alerts investigators to their relevance for research designs in regard to choice of covariates when examining these interventions.
Although the data presented here provide some empirical guidance for making individual treatment prescriptions for the 5- and 1-session interventions evaluated in the CTN0018 and CTN0019 studies, additional research is needed to have a broader understanding of which HIV risk reductions interventions work best for distinct subgroups of substance users. Our analyses were restricted to a limited set of primary and secondary potential predictors/moderators. Other variables, not examined here, might also be important predictors/moderators. For example, some individuals who are less comfortable in a group setting may be relatively poor candidates for a 5-session group intervention. Interpersonal variables may be relevant to who participates the most during group sessions and therefore benefits the most from a group intervention. Aspects of individual’s sexual history and preferences may also be associated with degree of sex risk reduction evident following the 5- and 1-session interventions. Qualitative research strategies might be a particularly useful way to obtain some further insights about the range of variables that might determine who benefits the most from different HIV risk reduction interventions.
It is also important to put the findings from the current study into the context of different approaches to intervention science. As mentioned, the predictor and moderator effects found here might prompt investigators to develop new, adapted versions of their interventions. In fact, as mentioned, an adapted version of the 5-session intervention for men has already been developed [
40]. However, developing a large number of adapted versions (e.g., cultural adaptation; age-related adaptation; adaptations based on gender, sexual history, and severity of use) may not be feasible from an intervention development point of view. In addition, clinicians may find that the availability of so many different versions is confusing. Alternatively, a sequential strategy can be used in which non-responders to one form of intervention receive a second, different form of intervention. At this point in time, both approaches can be pursued until it is clearer which direction is yielding more useful data.
Several limitations of the original CTN0018 and CTN0019 studies are important to keep in mind when evaluating the results of the current predictor/moderator analyses. Both studies were conducted in a variety of settings and had few exclusion criteria, but there are limits on the generalizability of any results from the data based on a number of factors, including self-referral to the study, age, type of substance of abuse, psychiatric and substance abuse diagnosis, and sexual history (i.e., women who had not had sex with a man in the past 6 months were excluded from the CTN0019 study). Another limitation is that participating counselors received 30 hours of special training in the 5-session intervention. Less training, as would be common if this 5-session intervention was implemented clinically, might yield different results than found here. An important limitation in comparing the two interventions is the difference in duration of the interventions (1 vs. 5 sessions). It may be that five sessions of standard HIV counseling would achieve comparable results to the 5-session skills building intervention.
In addition to the above limitations of the study designs, there are limitations of the current predictor/moderator analyses. As mentioned, no correction for the number of predictors/moderators was implemented, though we place greater emphasis on results significant at a .01 alpha level and on the multivariate results. Furthermore, other variables might confound the relations of the predictors/moderators examined here with outcome. The potential existence of measured or unmeasured confounding variables highlights the need to remain cautious about any causal interpretations of the findings reported herein. Moreover, the relationships reported here might be affected by the quality of the implementation of the interventions. The mediating or moderating role of the process of treatment (i.e., adherence ratings) in understanding the relation of patient variables to the effectiveness of HIV risk reduction interventions is a topic that should be explored in additional studies. Despite these limitations, the findings reported here have important practical implications for clinical implementation of these treatment and design of research studies on HIV sexual risk reduction interventions. A particular strength of this study is the use of real-world clinics and clinicians, thereby increasing the external generalizability of the findings.
Acknowledgments
The original CTN0018 and CTN0019 studies were funded by National Institute on Drug Abuse (NIDA) Clinical Trials Network grants: U10 DA13714 (Dennis Donovan, PI), U10 DA13035 (Edward Nunes, PI), U10 DA15815 (James Sorensen, PI), U10 DA13043 (George Woody, PI), U10 DA13038 (Kathleen Carroll, PI), U10 DA13711 (Robert Hubbard, PI), U10 DA13732 (Eugene Somoza, PI), U10 DA13045 (Walter Ling, PI), U10 DA13727 (Kathleen Brady, PI), U10 DA15833 (William Miller, PI). The preparation of the current article was supported by NIDA grant R33 DA027521 (Wan Tang, PI).
We wish to thank Susan Tross, Ph.D., for her comments on an earlier draft of this article.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
PCC prepared the first draft of the manuscript. PCC, RG, DC, WT, XT, and GW designed the study (choice and definition of variables; data analytic plan). RG conducted all data analyses. All authors (RG, DC, JS, HM, WT, HH, XT, and GW) provided edits and revisions to subsequent manuscript drafts. All authors read and approved the final manuscript.