An International Systematic Review and Meta-analysis of Multisession Psychosocial Interventions Compared with Educational or Minimal Interventions on the HIV Sex Risk Behaviors of People Who Use Drugs
- First Online:
- Cite this article as:
- Meader, N., Semaan, S., Halton, M. et al. AIDS Behav (2013) 17: 1963. doi:10.1007/s10461-012-0403-y
- 460 Views
This systematic review and meta-analysis examines the effectiveness of multisession psychosocial interventions compared with educational interventions and minimal interventions in reducing sexual risk in people who use drugs (51 studies; 19,209 participants). We conducted comprehensive searches (MEDLINE, EMBASE, CINAHL, Cochrane Central Register of Controlled Trials and PsychINFO 1998–2012). Outcomes (unprotected sex, condom use, or a composite outcome) were extracted by two authors and synthesised using meta-analysis. Subgroup analyses and meta-regression were conducted to explore heterogeneity. Multisession psychosocial interventions had modest additional benefits compared to educational interventions (K = 46; OR 0.86; 95 % CI 0.77, 0.96), and large positive effects compared to minimal interventions (K = 7; OR 0.60; 95 % CI 0.46, 0.78). Comparison with previous meta-analyses suggested limited progress in recent years in developing more effective interventions. Multisession psychosocial and educational interventions provided similar modest sexual risk reduction justifying offering educational interventions in settings with limited exposure to sexual risk reduction interventions, messages, and resources.
KeywordsHIV/AIDS preventionPeople who use drugsHarm reductionMeta-analysisSexual risk behavior
Sharing of drug injection equipment is a very efficient method for transmitting HIV. Persons who inject drugs (PWID) are also at risk of acquiring and transmitting HIV through unsafe sexual behavior. Sexual transmission of HIV among PWID is of increasing importance for several reasons including (but not limited to) the following factors: HIV infection among PWID has been reported in more than 120 countries, with a 20–40 % prevalence in at least 4 countries and over 40 % in nine countries . High seroprevalence epidemics among PWID may transition to generalized heterosexual epidemics [2, 3]. In areas where large-scale syringe exchange programs have been implemented, sexual transmission of HIV in some settings is now more common than injecting-related transmission among PWID [4, 5].
Additionally, non-injecting use of drugs (e.g., heroin, crack cocaine, methamphetamine) by inhaling, smoking, or ingesting, is associated with high rates of unsafe sexual behavior and HIV transmission in persons who use drugs (PWUD) [6–9]. In New York City, for example, HIV prevalence among non-injecting drug users is similar to HIV prevalence among PWID .
Drug use may lead to unsafe sexual behavior and HIV transmission through multiple pathways including exchange of sex for drugs or for money to purchase drugs; impaired judgement due to intoxication; or direct effects of certain drugs (e.g., crack, methamphetamine) on increasing the sense of sexual pleasure and on unsafe sexual practices [11–13].
The multiple relationships that exist between drug use and unsafe sex, in combination with potential difficulties in negotiating safer sex and technical difficulties in correct condom use have led to the development of multisession psychosocial interventions to reduce sexual risk behavior among PWUD (for example, cognitive-behavioral skills training). These interventions have been part of the intervention portfolio for many years, partly because multi-session interventions might address complex behavioral changes affecting personal and intimate behaviours . It is expected that multi-session psychosocial interventions would be more effective compared to brief educational interventions, although they might be more expensive and resource-intensive.
Several factors may moderate sexual risk reduction, including participant characteristics, intervention level, country, content, and delivery. For example, the psychological factors associated with sexual risk reduction to avoid becoming infected with HIV may be quite different from the psychological factors associated with reducing risk behavior to avoid transmitting HIV to others . HIV prevalence in a population may influence estimates of effectiveness . HIV positive persons may be more motivated to reduce their sexual risk behaviour to reduce potential harm to others. Other moderators of effectiveness of sexual risk reduction interventions include factors related to a person’s drug use. In addition, persons receiving drug treatment may make changes in sexual risk behaviour . The country in which the intervention is conducted may also be a moderator of effectiveness through contextual factors, such as HIV prevalence, stigma, and awareness . Further, as interventions have focused on targeting behavior change at different levels such as individuals, social networks, couples, and communities it is important to assess if these different approaches differ in effectiveness. While the content differs between these different levels of intervention (and therefore subgroup analyses may pick up on some of these differences) we did not comprehensively examine the specific components of interventions as moderators of effectiveness. Although some important factors will be explored in the analysis (see “Methods” section for further details).
This analysis updates a previous systematic review and meta-analysis that included studies published between 1991 and early 1999 conducted in the United States . It is important to also evaluate studies developed outside of the United States to examine more than a decade later the progress in effectiveness literature on multisession psychosocial interventions and to reflect on how to target future research. Therefore the present analysis includes worldwide studies published between 1991 and August 2012. Effectiveness of multisession psychosocial interventions, compared with educational interventions or minimal interventions (i.e., waitlist control, self-help booklet, or treatment as usual) for reducing sexual risk behaviors in PWUD was assessed in studies that used RCTs or quasi-experimental designs. The paper is consistent with the PRISMA reporting guidelines .
Prior to conducting the systematic review and meta-analyses, a review protocol (available on request) was developed outlining eligibility criteria, search strategy, quality assessment, data extraction, and statistical analyses. Studies of multisession psychosocial interventions compared with either educational interventions or minimal interventions for PWUD with a minimum eligibility age of 18 years were eligible. In addition, studies were required to provide data in a format that could be extracted and included in the meta-analysis.
Multisession psychosocial interventions consisted of more than two sessions that targeted psychological or social factors associated with illicit drug use and sexual risk behavior. These interventions were further sub-divided into four categories by level: individually focused (delivered on an individual basis or in a group), couples-focused, social network, and structural interventions. Individually focused interventions aimed primarily at changing individual behavior or personal factors that affect individual behaviors (e.g., skills training in reducing injection and sexual risk behaviors, communication skills, assertiveness and self esteem). Couples focused interventions aimed at changing both individual and couples’ behaviour. Social network interventions also included components of individually focused interventions and aimed to seek change at the social group level through changing social norms and behaviors. Structural interventions delivered as community-level interventions included different components, such as syringe exchange programs, condom distribution, HIV voluntary counselling and testing, and peer-delivered education and aimed to seek change at the community level .
We compared effectiveness of multisession psychosocial interventions delivered to the experimental group to (1) minimal interventions where participants in the control group had limited contact with intervention staff and limited intervention content or duration (i.e., waitlist control, self-help booklet, or treatment as usual); and (2) educational interventions where participants in the control group had a psychosocial or educational intervention delivered in one or two sessions (that might have included HIV testing and counselling), or provided information over a similar number of sessions as the intervention but without psychosocial input.
Outcomes included unprotected sex, condom use, a validated scale that provided a composite measure of sexual risk, and number of sex partners.
Study designs included randomized controlled trials (RCTs) and quasi-experimental studies (i.e., experimental studies comparing at least two groups with non-random assignment to the intervention). Quasi-experimental studies were included if method of assignment had a low risk of bias, there was no statistically significant evidence of baseline differences, or if any such differences were accounted for in the statistical analysis.
A search was conducted using five electronic bibliographic databases (Cochrane Central Register of Controlled Trials, CINAHL, EMBASE, MEDLINE, PsycINFO) and search terms that combined subject headings and text words. Search terms aimed to capture psychosocial interventions (e.g. ‘psychotherapy’, ‘cognitive therapy’, ‘behavior therapy’), drug dependence (e.g. ‘substance related disorders’) and sexually transmitted infections (e.g. ‘sexually transmitted diseases’, ‘HIV’, ‘sexual behavior’, ‘condoms’). No language restrictions were applied in the search.
Each database was searched on August 16, 2012 for reports published during January 1998 through August 2012. Studies conducted between 1991 and 1998 were identified by utilising the search strategy from a systematic review of US studies  and a worldwide Cochrane review . In addition, papers were found by searching references of retrieved articles, tables of contents of relevant journals, and other previous systematic reviews.
Eligible studies were each assessed individually for methodological quality using the National Institute of Health and Clinical Excellence (NICE) checklist for RCTs for selection, performance, and attrition bias . Limitations of individual studies identified in the quality assessment are provided in Table 1 in Appendix F (Supplementary Material). In addition, we created grading of recommendations, assessment, development and evaluation (GRADE) profiles  and classified the overall quality of the studies (high, moderate, low, very low), which considered study design (randomized or quasi-experimental design), quality of individual studies (as assessed in the NICE checklist), consistency of results (I2 ≥ 50 %, Q-statistic p < 0.1 and visual examination of forest plots used to judge inconsistency/heterogeneity), imprecision (whether confidence intervals included both a clinically meaningful effect (OR ≤ 0.75) and no benefit (OR ≥ 1.00); or sparse data (study sample size <100)), directness of the evidence (i.e. similarity of participants to PWUD who engage in sexual risk behaviour), and publication bias (substantial asymmetry identified by funnel plot).
Assessment of study eligibility, quality, and data extraction were independently completed by at least two authors, with disagreements resolved by discussion with other authors and re-review of the original articles. For each study included in the meta-analysis, we extracted data on one sexual risk outcome in the following order of priority: unprotected sex (all participants included), condom use (only sexually active participants included), a validated scale that provided a composite measure of sexual risk, or number of sex partners .
Where studies reported multiple follow up periods and data collection window periods, data at end of the intervention (end of treatment, or first post-intervention) and the longer window risk behaviour period (e.g. 3 months or 6 months) were extracted to maximise consistency across studies included in the meta-analysis (end of treatment or first post-intervention) and were used in assessing effectiveness of interventions. When data were not reported we estimated measures of variability (such as standard deviations or standard errors) using standard formulae from other related statistics such as p values, F ratios, or t values . When only the summary effect size, such as standardised mean difference or odds ratio was reported (e.g. without reporting data on number of events in each arm), we extracted and used this data instead of means or proportions for each group.
Studies were also coded for age, gender, drugs used and route of drug administration, intervention and recruitment setting (e.g., substance abuse treatment, outreach), content of the intervention, country, sexually- and parenterally-transmitted infections (i.e., HIV, STDs, hepatitis B/C), intervention duration, sample size, follow up period, and participants’ completion of the intervention.
Random effects meta-analysis was conducted using Review Manager 5 software . Standardized mean differences (SMD) were calculated using Hedges g (for further details, please see [21, 23, 24]) from continuous (change from baseline) data with negative effect sizes favouring the multi-session psychosocial intervention.
Odds ratios (OR) were calculated from dichotomous data with values less than 1 favouring the multi-session psychosocial intervention (for further details on synthesis of dichotomous data, please see [21, 24, 25]).
SMDs calculated from continuous data and ORs from dichotomous data were converted into log ORs using standard formulae  and then entered into Review Manager along with their standard error.
Heterogeneity was assessed using the I2 statistic  and the Q-statistic (p < 0.10: statistically significant). For further details on assessment of heterogeneity please see . Publication bias was examined by visual inspection of the symmetry of funnel plots and, formally, using Egger’s test (p < 0.05: statistically significant) .
To better assess effectiveness and sources of heterogeneity, subgroup analyses (using Review Manager 5) were examined. In addition, mixed-effects meta-regression (Stata 10 ) was conducted using the metareg command. This method uses an extension of variance weighted least squares by estimating an additional additive component of variance (tau2). There were 8 planned covariates: study-provided HIV testing (yes or no), HIV prevalence (0–20 % or 21–100 %), administration of drugs (PWID or PWID and also use crack), intervention level (individual, couple, social network, or structural), recruitment (outreach or substance abuse treatment), intervention setting (methadone maintenance (reference group), opioid detoxification, psychosocial treatment), location (United States or not), publication date (1991–1999 or 2000–2012). In addition, we examined three covariates relating to the content of the intervention and control groups. We examined the impact of including condom skills and motivational enhancement techniques in multisession interventions as these have been extensively used as components of multisession psychosocial interventions in primary studies and investigated in previous meta-analyses of sexual risk reduction as covariates [30, 31]. Further, we examined the impact of including psychosocial components in a control group as this analysis of inclusion of psychosocial components may potentially contribute to our understanding of the heterogeneity.
We did not explore all potential moderators of effectiveness as there is an important balance between exploring heterogeneity extensively and the risk of identifying false positive results which increases when more covariates are examined in the meta-regression and subgroup analyses . There is no complete consensus on how best to address this balance (e.g., the number of covariates) but was based on the judgement of the authors.
We did not undertake a comprehensive assessment of all potential intervention components as predictors of effectiveness. Such analyses, to minimise potential bias, require the use of validated taxonomies of behaviour change techniques  which can include up to 40 different components and the assessment of the theoretical underpinnings of these behavior change techniques . Poor reporting in many studies may lead to bias in the analyses and therefore, minimizing this risk requires obtaining full treatment manuals used in the primary studies. A seperate paper is required to comprehensively address how the content of interventions may moderate effectiveness in a transparent and unbiased manner.
The exponentiated form of the log odds ratio comparing multisession psychosocial interventions with educational interventions or with minimal interventions was the outcome variable in the meta-regression analyses. The permutation test (10,000 replications) was used to assess statistical significance to reduce the probability of a false positive result . Overall effect by publication date was examined using cumulative meta-analysis (see Appendix D in Supplementary Material).
Sensitivity analyses examined whether using observed cases or worse case scenario assumptions for loss to follow-up (drop outs were assumed to have a bad outcome) affected results. Corrections based on worse-case scenario assumptions were limited to studies after 1999 as sufficient data were often not available in earlier studies.
Further, sensitivity analyses assessed the impact of including only RCTs in the meta-analysis, and also of excluding studies for which it was not possible to adjust for clustering because of insufficient reported data.
The previous  US-based analysis included 33 studies of individual or social network interventions on 13,847 participants. One study of adolescent drug users  was excluded from this current meta-analysis as the eligibility criteria was limited to studies of adults (see Appendix B for list of excluded studies, Supplementary Material).
Of these 19 additional studies (18 RCTs and 1 quasi-experimental study with a total of 5,186 participants), 12 were US-based; 6 were conducted either in China (Ref. 45, Appendix A), Thailand (Ref. 25, Appendix A), Russia (Ref. 24, Appendix A), Kazakhstan (Ref. 50, Appendix A) or Australia (Refs. 3 and 22, Appendix A); 1 study was conducted in the United States and Thailand (Ref. 16, Appendix A) and was analyzed as international in subgroup analysis. Of the 19 studies, 13 studies evaluated individually focused interventions (Refs. 2, 3, 6, 10, 12, 13, 18, 19, 22, 24, 45, 47, 49; Appendix A), four evaluated social network interventions (Refs. 15, 16, 25, 51; Appendix A), one evaluated a couples-based intervention (Ref. 50, Appendix A) and one evaluated a structural intervention (Ref. 48, Appendix A).
We included 51 studies with a total of 19,209 participants in the meta-analyses (for study characteristics: Appendix F and G, Table 1, Supplementary Material). Of these 51 studies, 46 studies compared multisession psychosocial interventions to educational interventions and seven studies compared multisession psychosocial interventions to minimal interventions. Two studies had three intervention groups (multisession psychosocial, educational, and minimal) and were included accordingly in both meta-analyses (Refs. 3 and 47, Appendix A). Comparisons with the previous meta-analysis  appear in Appendix C (Supplementary Material).
Effectiveness of Multi-session Psychosocial Interventions
Summary characteristics of studies included in the meta-analysis on interventions (k = 51) for reducing sexual risk behavior in people who use drugs
% of total K
Cross country (US + Thailand)
Not reported or insufficient data
Mean age (in years)
K = 45, 35 (4.32)
Not reported or insufficient data
K = 6
Needle and syringe exchange
Type of drug use
Dominant % cocaine use
Dominant % opioid use
Dominant % use of both or equal % of both above
Unclear which type of drug use dominant in sample
Other drugs: methamphetamine
HIV testing and HIV status
No HIV testing and No HIV status reported
HIV testing provided: 0–20 % sero-positive
HIV testing provided: 21–100 % sero-positive
No testing but HIV status self reported
Time to first follow-up data
Total number of study participants
Median (Inter-quartile range)
Multisession psychosocial interventions
Evidence summary of multisession psychosocial interventions for reducing sexual risk in drug users
OR (95 % CI)
Multisession vs. educational
Completer (k = 46)
OR = 0.83 (0.72, 0.95)
Worse case scenario (k = 46)
OR = 0.86 (0.77, 0.96)
RCTs only and worse case scenario (K = 26)
OR = 0.81 (0.68, 0.97)
Multisession vs. minimal control
All studies (K = 7)
OR = 0.60 (0.46, 0.78)
RCTs only (K = 6)
OR = 0.58 (0.41, 0.80)
The GRADE quality assessment indicated a low rating for the overall quality of evidence of the 46 interventions (Table 2). A sensitivity analysis with only RCTs found a slightly larger effect size and moderate quality evidence according to GRADE criteria (K = 26; OR = 0.81; 95 % CI 0.68, 0.97; I2 = 64 %; Q = 67.57, p < 0.0001). However there was still high and statistically significant heterogeneity.
There were too few studies to assess publication bias for multisession psychosocial interventions versus minimal interventions alone therefore we included studies that compared multisession psychosocial interventions with either control group (educational or minimal) in the funnel plot. Egger’s (bias = −0.89, p = 0.09) and Begg’s tests (z = 1.80; p = 0.073) did not show statistically significant evidence of publication bias although there was a trend which suggests this cannot be ruled out. However, the trim-and-fill analysis suggested that no adjustment for publication bias was required. Further, visual inspection of the funnel plot does not suggest strong asymmetry.
Predictors of effectiveness for multisession psychosocial interventions compared with educational interventions (K = 46)
We examined the impact of publication date (1991–1999 vs. 2000–2011), location (US based vs. non-US based), recruitment (substance abuse treatment vs. outreach) and treatment setting (MMT vs. detoxification, MMT vs. outpatient psychosocial substance misuse treatment), HIV prevalence (0–20 % vs. over 20 %), HIV testing (yes vs. no), use of condom skills training (yes vs. no) or motivational enhancement techniques (yes vs. no) in multisession psychosocial interventions, and the inclusion of a psychosocial component in the control group (yes vs. no).
Subgroup analyses on the predictors of effectiveness for multisession interventions
OR (95 % CI)
GRADE (overall quality of evidence)
1991–1999 (K = 31)
OR = 0.87 (0.76, 0.99)
2000–2011 (K = 15)
OR = 0.84 (0.68, 1.04)
Type of interventione
Individual (k = 43)
OR = 0.84 (0.74, 0.95)
Social network (k = 5)
OR = 0.70 (0.47, 1.05)
Structural (k = 2)
OR = 0.75 (0.60, 0.94)
Couples (k = 1)
OR = 0.34 (0.15, 0.77)
US studies (k = 39)
OR = 0.86 (0.75, 0.97)
Non-US studies (k = 7)
OR = 0.89 (0.72, 1.11)
In treatment (K = 12)
OR = 0.73 (0.55, 0.96)
Not in treatment (K = 32)
OR = 0.93 (0.83, 1.04)
MMT (K = 5)
OR = 0.57 (0.46, 0.71)
Detoxification (K = 4)
OR = 0.96 (0.71, 1.29)
Psychosocial interventions (K = 3)
OR = 0.51 (0.18, 1.44)
Yes (K = 16)
OR = 0.91 (0.80, 1.03)
No (K = 30)
OR = 0.76 (0.60, 0.95)
<20 % (K = 25)
OR = 0.85 (0.74, 0.98)
>20 % (K = 10)
OR = 1.15 (0.98, 1.36)
Injection drug and crack use
PWID alone (K = 30)
OR = 0.84 (0.73, 0.95)
PWID and crack use (K = 12)
OR = 0.86 (0.67, 1.12)
Studies conducted outside of the United States (K = 7) comprised three studies using individually focused interventions (Refs. 3, 22, 24; Appendix A), two studies using social network interventions (Refs. 16, 25; Appendix A), one couples-based intervention (Ref. 50, Appendix A) and one structural intervention study (Ref. 48, Appendix A). Both U.S. and non-U.S.-based studies had a small effect (Table 3), with no statistically significant difference identified between them (exp β = 0.99; p = 0.97).
Recruitment and Treatment Setting
There were differences for recruitment setting (substance abuse treatment vs. outreach, exp β = 0.80, p = 0.07; explaining 16.9 % of between study variance) approaching statistical significance suggesting people recruited from substance abuse treatment (K = 12; OR 0.73; 95 % CI 0.55, 0.96) were more likely to reduce sexual risk behaviour compared with those not currently receiving drug treatment (K = 32; OR 0.93; 95 % CI 0.83, 1.04). The differences between interventions delivered in methadone maintenance, opioid detoxification (exp β = 1.69, p = 0.18), and outpatient psychosocial substance misuse treatment settings (exp β = 1.13, p = 0.73) were not statistically significant.
For HIV serostatus (using self report or study-delivered HIV testing results), only five studies recruited exclusively HIV positive (K = 2; OR = 0.91; 95 % CI 0.48, 1.73) or HIV negative populations (K = 4; OR 0.63; 95 % CI 0.36, 1.11). The other studies (K = 26) reported very low proportions of HIV-positive participants and did not include results stratified by HIV status. To enable a sufficient number of studies to examine the impact of prevalence of HIV-positive participants on effectiveness estimates, studies were categorized as having 0–20 % of HIV-positive participants versus >20 % of HIV-positive participants. The impact of HIV prevalence approached statistical significance (lower vs. higher HIV prevalence among participants, exp β = 1.29, p = 0.056; explaining 54.91 % of between study variance), suggesting studies with lower HIV prevalence (K = 25; OR 0.85; 95 % CI 0.74, 0.98) had more risk reduction than studies with higher HIV prevalence (K = 10; OR 1.15; 95 % CI 0.98, 1.36). However, when including only studies that provided HIV testing, there was no difference (K = 32; exp β = 1.10, p = 0.65). An examination of pre-post data for sexual risk suggests that in the 35 studies, participants in both intervention and control groups significantly reduced their sexual risk behavior.
HIV Testing and Drug Use
No statistically significant differences in outcome were found between studies (k = 46) providing HIV testing or not (exp β = 1.21, p = 0.12), or for studies (k = 42) including either PWID alone or PWID and crack users (exp β = 1.05, p = 0.71).
Condom Skills Training and Motivational Enhancement
Neither condom skills training (K = 29; exp β = 1.08, p = 0.71) nor motivational enhancement (K = 29; exp β = 1.15, p = 0.49) components were found to predict effectiveness of multisession psychosocial interventions compared with educational interventions. There was substantial missing data for these covariates as many included studies included limited details concerning the content of interventions.
Psychosocial Components Offered to Control Groups
There was also no evidence that the inclusion of psychosocial components in the control group impacted on effectiveness estimates (K = 46; exp β = 1.07, p = 0.65).
In summary, studies with lower HIV prevalence appeared to show greater risk reduction compared with those with higher HIV prevalence. Studies that recruited participants receiving substance misuse treatment appeared to show greater effectiveness than studies of participants who were not in substance misuse treatment. There was no evidence that publication date, location, receiving HIV testing, type of drug use, inclusion of condom skills training, or inclusion of motivational enhancement, or the inclusion of psychosocial components offered to the control group impacted effectiveness of multisession psychosocial interventions.
Comparing the effectiveness of individual-focused, social network, couples-focused, and structural interventions with educational and minimal interventions
As multisession psychosocial interventions differ in their intervention level (i.e. individually focused, social network, couples, or structural) we assessed whether intervention level impacted the effectiveness of these interventions. There were a small number of studies (K = 8) examining social network, couples and structural interventions. A sensitivity analysis removing these interventions from the comparison of multisession psychosocial interventions and educational interventions resulted in only minor changes to the pooled effect estimate and confidence interval but the effect size was no longer statistically significant (K = 38; OR 0.89; 95 % CI 0.79, 1.00).
To examine further the effect of different intervention levels and to enable sufficient studies for comparison we compared multisession psychosocial interventions with educational and minimal interventions. Almost all studies (K = 43) evaluated individually focused interventions, and had a slightly smaller effect size (OR 0.84; 95 % CI 0.74, 0.95) than that of other interventions compared with control groups. Structural interventions (Refs. 7 and 48, Appendix A) had a moderate and statistically significant reduction in risk behaviour compared to the control group (K = 2; OR 0.75; 95 % CI 0.60, 0.94). Although social network interventions (Refs. 15, 16, 25, 46, 51 Appendix A) were associated with a moderate effect size (K = 5; OR 0.70; 95 % CI 0.47, 1.05), the 95 % confidence interval was relatively wide and not statistically significant. One small trial of a couples-focused intervention found a large benefit but was based on a small number of participants (K = 1; OR 0.34; 95 % CI 0.15, 0.77).
This meta-analysis comprised 51 studies conducted in the United States or in 5 other countries, over 50 % more than the 33 U.S.-based studies of the previous meta-analysis . Similar to the previous US-based meta-analysis, the overall effect size for multisession psychosocial interventions was modest  and is consistent with other systematic reviews examining sexual risk reduction in PWUD [15, 36].
The previous US-based meta-analysis of adolescent and adult studies that included only individually-focused and social network intervention studies, showed a small and similar favourable effect in comparison with educational interventions delivered to the control groups (K = 30; OR 0.91; 95 % CI = 0.81, 1.03) and a large effect in comparison with minimal interventions delivered to the control groups (K = 3; OR 0.60; 95 % C.I. = 0.43, 0.85) . The previous meta-analysis included one study with an adolescent population that was excluded from the current meta-analysis . Results of the current international meta-analysis on multisession psychosocial interventions compared to educational interventions (K = 46; OR 0.86; 95 % CI = 0.77, 0.96) suggests a similar reduction (14 % reduction in odds compared with 9 % in the previous review) in sexual risk and a slightly narrower confidence interval. Moderators of effectiveness of multisession psychosocial interventions were also similar in the previous and updated meta-analysis. For example, as with the previous meta-analysis, there were large positive effects for multisession psychosocial interventions (40 % reduction in odds) compared to minimal interventions (K = 7; OR 0.60; 95 % CI 0.46–0.78). In addition, studies which included people who were in treatment tended to find greater risk reduction. Further comparisons are provided in Appendix C (Supplementary Material).
A critical finding of this updated meta-analysis is the similar summary effect size compared to that of the previous meta-analysis. For the 15 studies published after 1999, the summary OR was 0.84 (95 % CI 0.68, 1.04). While historical comparisons must be interpreted with caution because of potential confounding (changes in drug use, drug market, risk behaviours) the similarities in the summary effects suggests that little progress has been made in developing more effective interventions to reduce sexual risk behaviors of PWUD.
The overall analyses suggest that multi-session psychosocial interventions that have a similar message to educational interventions may have diminishing returns or may not capture the needs of participants for sexual risk reduction. The small differences in effect between multisession psychosocial interventions and educational interventions suggest that educational interventions can be used to reduce sexual risk among PWUD with prior limited exposure to HIV sexual reduction interventions. In addition, a recent meta-analysis has shown the effectiveness of brief single session interventions in preventing STIs . Delivering these interventions to larger numbers of PWUD may be more beneficial and cost-effective than delivering longer interventions to smaller numbers of PWUD . Promoting sexual health and wellness and normalization of sexual communication and sexual risk reduction behaviours at national levels as well as at community levels may add to the success of sexual risk reduction interventions [39, 40].
There are several potential explanations for the continued moderate sexual risk-reduction effect in the updated meta-analysis. Firstly, most published interventions continue to focus on individual-level interventions. A sensitivity analysis in the present meta-analysis found that when only including individual-level interventions the effect size was reduced and no longer statistically significant. A greater number of non-individual level interventions (e.g. structural, couples-based interventions) in a future update may substantially impact estimates of effectiveness. A variety of other potential factors include insufficient exposure to the sexual risk component of the interventions, and simultaneous changes in both drug use and sexual risk behaviors are difficult [11, 41]. Because PWUD might be faced with reducing risk in two complex behaviours, there is a need for sexual risk reduction interventions to address the specific sexual needs of PWUD as well as the peer-level, community, and environmental factors that affect sexual risk [41, 42]. However, there was no difference in unprotected sex between the intervention and control groups (OR 1.25; 95 % CI 0.63, 2.48; based on our intention to treat analyses) in a recent small trial that trained PWUD as outreach workers to reduce their own and others’ sexual risk behavior .
In terms of moderator of effectiveness, the meta-regression results suggest that studies in which <20 % of participants were HIV-positive (by self-report or based on study test results) had greater reductions in sexual risk behaviors although this effect did not reach statistical significance. This effect was in the opposite expected direction but there are a number of limitations to this finding. The 20 % cut-off point was drawn arbitrarily to allow for comparison. The finding is also vulnerable to ecological bias, that is, although this suggests a potential effect for HIV status at the population level this summary data may misrepresent the effect for individual participants. Other meta-analyses have found HIV-positive PWUD report lower rates of sexual risk behavior than do HIV-negative PWUD . Furthermore, results of meta-regression can be vulnerable to confounding , for example, it is possible that HIV-positive participants in most studies knew their HIV status before study enrollment and have already reduced their sexual risk behaviors. Recent results of an RCT on the effectiveness of brief individual risk reduction counselling versus information (with and without HIV testing and the role of on-site HIV testing versus referral) showed an increase in receipt of test results of on-site rapid HIV testing in substance abuse treatment centers (versus referral) (45). However there was no 6-month effect of counselling on reducing sexual risk behaviours among HIV-negative participants who were sexually active at baseline and who reported they did have not have an HIV test in the last 12 months .
Our meta-analysis also shows that studies that recruited participants from substance abuse treatment had greater effects than studies with out-of-treatment drug users although this difference only approached statistical significance. This result may suggest that such participants were more motivated to change their sexual risk behaviour as they had already engaged in substance abuse treatment, the treatment setting provided an environment supportive of sexual risk reduction, or that combined or integrated prevention and treatment interventions provided stronger benefits [15, 46, 47]. Further research, such as qualitative or mixed methods studies, should be conducted to explore these findings further.
In terms of methodological considerations, there was substantial heterogeneity in all analyses, reflected by the forest plots, and the Q and I2 statistics, that was further explored by meta-regression and sub-group analyses. The high heterogeneity in part explains the low-to-moderate overall quality of evidence identified in the GRADE evidence profiles (other reasons include imprecision of confidence intervals and inclusion of quasi-experimental studies). A further limitation of the systematic review was the reliance on mainstream databases (e.g. Medline, Embase etc.). While these databases have a relatively wide international coverage, including journals from up to 70 countries, it is possible further studies may have been identified by searching region specific electronic databases. In addition, we only included data from end of treatment or first follow up period as most data was available for this time point. Therefore it is unclear if the reported effects are maintained in the long term. Furthermore, there was a trend that suggested it was not possible to rule out potential publication bias. However, the trim-and-fill analysis suggested there was no need to adjust for publication bias.
Our meta-analysis points to a number of gaps in the literature. Firstly, most studies evaluated individually focused interventions and therefore often did not attempt to modify the risk environment in which individuals interact. Secondly, many studies included a diverse range of participants who differed in the drugs they used and in the administration of these drugs (e.g., crack cocaine, injection of drugs, methamphetamine), HIV serostatus, and whether they engaged in commercial sex work. Further research is required to assess interventions in particular subgroups of drug users. For example, few studies evaluated interventions for HIV-positive PWUD. In addition, there is limited research examining whether subgroups of drug users have different prevention needs depending on prior exposure to sexual risk reduction interventions or messages. Thirdly, most studies were conducted in the United States, limiting the representativeness of our findings to other countries. Fourthly, most studies rely on self reported outcomes and do not report biological outcomes (for example, HIV seroconversion). While some recent studies report biological outcomes (e.g., Ref. 16, Appendix A) there are not yet sufficient data from well-powered intervention studies to conduct meta-analysis of biologic outcomes.
In terms of implications for programs, the multisession psychosocial interventions included various components such as providing information on HIV infection and risk reduction; access to condoms; HIV testing and counselling; interpersonal, personal, and technical risk reduction skills; and development of peer norms for HIV risk reduction behaviors. Our meta-regression analyses focused only on condom use skills and motivational enhancement and we did not find evidence that either of these components predicted effectiveness for multi-session psychosocial interventions. Motivation and condom skills training were not statistically significant predictors in other meta-analyses of sexual risk reduction [30, 31]. This is in contrast to a previous study based on a structural equation model of sexual risk reduction which suggested that both personal motivation and behavioural skills training had a positive impact on reducing risk behaviour . This difference between results of meta-analyses and primary studies may reflect the limitations of meta-regression analyses of summary data and the complex ways in which various components of multi-session psychosocial interventions impact sexual risk reduction. In addition, our subgroup analyses showed potential benefits of structural interventions and social network interventions. These studies focused on mobilizing community members to reinforce prevention messages, increase availability of condoms, diffuse HIV education, and advocate for changes in the risk environment. National strategies for sexual health, wellness, and communication would add to the role of sexual risk reduction interventions. Further research is needed to examine which components of psychosocial interventions are associated with greater effectiveness.
In summary, the updated meta-analysis found a modest effect for multi-session psychosocial interventions over one- or two-session educational interventions and a greater reduction compared to minimal interventions, and highlighted the potential promise of social network, couples-based, and structural interventions in sexual risk reduction. The most important finding, however, was that the effect size had not improved substantially since the previous meta-analysis, suggesting a lack of progress in sexual risk reduction interventions for PWUD. New research directions in developing more effective interventions to reduce unsafe sexual behaviour among PWUD including cost-effectiveness analyses are needed. Programmatic and financial priority can enhance efforts aimed at developing and evaluating promising interventions.
Although modest in effect, sexual risk-reduction interventions are still important for PWUD. Educational interventions may be particularly justified in resource limited settings because there are modest differences in risk reduction found for this approach compared with more intensive multisession psychosocial interventions and greater reductions as compared to minimal interventions.
The findings and conclusions in this paper are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention. Don Des Jarlais was supported in part by Grant R01 AI 083035 from the US National Institutes of Health. No conflicts of interest were declared by any authors.