Introduction

The global HIV epidemic remains a public health challenge. More than 34 million people are currently living with HIV/AIDS worldwide with sub-Saharan Africa bearing the heaviest burden, accounting for 69 % of all people living with HIV [1]. The global HIV epidemic remains stable with an estimated 2.5 million people newly infected with HIV per year [1]. Unprotected sex continues to account for the largest proportion (80 %) of all new HIV infections globally [1, 2]. Because sexual transmission of HIV originates from individuals infected with HIV, national AIDS campaigns in the U. S. and elsewhere increasingly emphasize secondary prevention as critical to the control of HIV [3].

Most PLWHA remain sexually active after diagnosis, and many report engaging in unprotected sex [4–6]. Reductions in sexual risk behavior often occur immediately following an HIV diagnosis but for some PLWHA, changes in sexual risk behavior are difficult to maintain [7]. A meta-analytic review of U.S. adults unaware they are infected with HIV found that participants reported a 37 % reduction in unprotected sex after receiving notification of their HIV-positive status [6]. In another meta-analytic review, 26 % of HIV-positive men who have sex with men (MSM), living with HIV for a median of 70 months, reported engaging in unprotected anal sex [5]. Unprotected sex among PLWHA increases both the risk of HIV transmission to an uninfected partner and the possibility of acquiring drug-resistant strains of the virus from an infected partner [8, 9]. Thus, understanding the factors associated with sexual risk-taking among PLWHA is critical to the development of secondary prevention strategies.

Alcohol use has been implicated as a prominent risk factor in the transmission of HIV [10]. Alcohol consumption may increase an individual’s risk by impairing sexual decision-making and by weakening the immune system [11, 12•]. Baliunas, Rehm, Irving, and Shuper [13] found the incidence of HIV was associated with overall level of alcohol consumption. Drinkers were at 77 % higher risk of HIV infection than non-drinkers. Moreover, risk of HIV infection among heavy episodic drinkers was double that of non-heavy episodic drinkers. Another meta-analysis examining the association between alcohol use and HIV from studies conducted in Africa [14] found drinker status to be associated with HIV infection such that drinkers were 70 % more likely to be HIV-positive than non-drinkers. Again, risk of HIV was notably higher among problem drinkers vs. non-problem drinkers. Thus, alcohol use appears to increase the likelihood of becoming infected with HIV in dose-response fashion.

Alcohol use among people already infected with HIV is also associated with sexual risk behaviors. A meta-analysis of 27 studies examining the association between alcohol (any alcohol consumption, problematic drinking, and alcohol use before/during sex) and unprotected sex among people living with HIV/AIDS found a positive association between alcohol use and unprotected sex [15]. These findings suggest that alcohol use is a significant risk factor for transmission among PLWHA.

The purpose of this systematic review and meta-analysis was to determine the current state-of-the-science concerning the (a) prevalence of alcohol use, (b) prevalence of sexual risk behaviors, and (c) strength of the association between alcohol and sexual risk behavior among individuals infected with HIV. Prior meta-analyses of the alcohol-sexual risk-HIV literature [13–15] focused on the overall alcohol-risky sex and alcohol-HIV associations within studies published through 2008. To determine the current state-of-the-science, we restricted our meta-analysis to the most recently published studies (i.e., January 1, 2012 through April 1, 2013) that assessed alcohol use and sexual risk behaviors among PLWHA. (We restricted the time-frame of our search because the methods used and quality of reporting tends to be stronger in more recent literature [16].) We also evaluated whether characteristics of the study (e.g., geographical region) and sample (e.g., sex of participant, medication adherence) influenced the prevalence of alcohol use or sexual risk behaviors in this population.

Methods

Search Strategy, Inclusion Criteria, and Study Selection

We searched multiple electronic reference databases (PubMed, EMBASE, Global Health, PsycINFO, CINAHL, ERIC, SocIndex, The Cochrane Library, and Web of Science [social sciences and science citation indices]) using a Boolean search strategy: (((binge* OR drinking OR "drinking behavior"[mesh:noexp]) AND (liquor[all fields] OR alcohol OR ethanol OR alcoholic beverages)) OR binge drinking OR alcoholic beverages OR alcohol drinking OR alcohol abuse OR alcoholic OR alcohol OR "alcohol-related disorders" [MeSH] OR alcoholism OR intoxicat* OR drunk* OR liquor[all fields]) AND ((dangerous behavior OR risk taking OR risk factors AND sexu*[tiab]) OR sexual behavior OR sexual partners OR unsafe sex OR behavior disinhibition OR condoms OR "men who have sex with men" OR "men having sex with men" OR MSM) AND (AIDS[sb]). Our search strategy was developed with the assistance of an expert health sciences librarian in the Alpert Medical School of Brown University. Because many electronic databases have specific search methods (e.g., Medical Subject Heading [MeSH] terms used in PubMED are not available in other databases such as PsycINFO), our basic search strategy was modified based on the specific search requirements for each electronic database. All final electronic reference database searches were conducted on April 1, 2013.

Studies (or portions of studies) were included if they (1) sampled PLWHA (i.e., 100 % of the sample was HIV-positive or outcomes were separated by HIV-status); (2) evaluated alcohol use and any sexual risk behavior (e.g., condom use, number of partners); (3) provided sufficient information to calculate effect sizes; and (4) were published (including electronic publications) between January 1, 2012, and April 1, 2013. Reference sections of relevant manuscripts (including published reviews obtained through the electronic reference database search) were also reviewed. Studies that fulfilled the inclusion criteria were included. When authors reported details and/or outcomes of the intervention in multiple studies, the studies were linked in the database and represented as a single study. The manuscript reporting the main trial outcomes was selected as the primary study; the publication date from the primary study was used to determine eligibility. The final sample included 17 studies reporting 19 reports (Fig. 1) [17–33].

Fig. 1
figure 1

Study selection process for the meta-analysis. Note. Two studies overlapped and were used as both a primary manuscript and a supplemental manuscript (pilot data) for another final study

Full size image

Coding and Reliability

Two independent coders rated the study information, sample characteristics (e.g., gender), design (e.g., recruitment strategy) and measurement specifics, and length and content of control and intervention condition (e.g., number of sessions). If data from the same sample appeared in multiple publications, the most comprehensive report was used while supplementing missing data from the other report(s). Study methodological quality was assessed using 17 items (e.g., random assignment) from validated measures [34–36]; total possible quality score is 25. We examined inter-rater reliability for all study descriptors. For the categorical variables, raters agreed on 92 % of the judgments (mean Cohen’s κ = .82). Reliability for the continuous variables yielded an average intraclass correlation coefficient (ρ) of 0.85 across categories (median = 1.00). Disagreements between coders were resolved through discussion.

Study Outcomes and Effect Sizes

Effect sizes were calculated to assess both the prevalence and association between alcohol use and sexual risk behavior. Prevalence outcomes included any alcohol consumption, alcohol use before sex, condom use, condom use at last sexual event, and multiple sexual partners. Prevalence estimates were calculated as the proportion of HIV-positive persons who had engaged in alcohol use or sexual risk behavior divided by the total number of HIV-positive persons in the sample. (For sexual risk behaviors, we included only HIV-positive persons who were sexually active.) Consistent with standard meta-analytic methods [37], prevalence estimates (and the standard error and inverse variance weight) were converted to logits and used in all analyses. Results of the analyses (using the logits) were converted back into prevalence estimates (and corresponding confidence intervals) for ease of interpretation of study findings. The association between alcohol use and unprotected sex was determined from the analyses of binary data (i.e., 2 [alcohol use: yes, no] × 2 [unprotected sex: yes, no]). Because the raw cell frequencies of the 2 × 2 contingency tables were not reported in any of the studies, other statistical information (i.e., odds ratio, chi-square, correlation coefficients) was used to estimate the effect sizes (corrected for sample size bias) using standard procedures [37, 38]. Effect size estimates for the alcohol-risky sex association are expressed as a standardized mean difference (d) where the sign of d is positive when alcohol use was associated with unprotected sexual behavior.

From the 17 studies that met the inclusion criteria, 19 reports were analyzed. (Two studies provided outcomes separate for men and women [28] and treatment and control groups [33]). All of the studies reported at least one prevalence estimate (15 alcohol use, nine alcohol use before sex, 12 condom use, five condom use at last sex event, and 11 multiple sexual partners) or association (five alcohol and unprotected sex). Studies that specifically recruited PLWHA who engaged in alcohol or sexual risk behaviors (e.g., heavy drinkers [26, 27]; unprotected sex [26, 27, 31]) or recruited from high-risk samples (e.g., sex workers [18]) were excluded from corresponding analyses to avoid overestimating the prevalence of alcohol use (any) or sexual risk behaviors (condom use, condom use at last sex, and multiple sexual partners). Multiple effect size estimates were calculated from individual studies when they reported more than one outcome variable (e.g., condom use separated by steady and casual partners).

Statistical Analyses

Data analyses were conducted with Stata 12 [39] using published macros [37, 40]. Weighted mean logit effect sizes (prevalence estimates) and weighted standardized mean effect sizes (association) were calculated using random-effects procedures [37]. The 95 % confidence intervals (CIs) surrounding a weighted mean logit effect size were calculated; CIs indicate the degree of precision as well as the significance of the mean logit effect size [37]. The homogeneity statistic, Q, was calculated; a significant Q indicates a lack of homogeneity and an inference of heterogeneity. To assess the extent to which outcomes were consistent across studies, the I 2 index and its corresponding 95 % CIs were calculated [41, 42]. I 2 varies between 0 (homogeneous) and 100 % (heterogeneous) [43]. If the CIs around I 2 include a zero, the set of effect sizes is considered homogeneous.

To explain variability in the prevalence estimates, moderator analyses were conducted using a modified weighted regression analysis (following fixed-effects assumptions) with weights equivalent to the inverse of the variance for each effect size [37, 44]. Regression analyses examined a priori determined moderators. Study (geographical region based on the World Health Organization region groupings[45]) and sample (proportion women, proportion alcohol users, proportion MSM, proportion of participants with multiple partners, proportion currently on ART, proportion adhering to HIV medication) characteristics were examined as potential moderators of the prevalence estimates. Weighted regression analyses were conducted only for outcomes with sufficient studies (i.e., > 5 studies).

Results

Details of the 17 included studies are provided in Table 1. Studies were conducted in several geographical regions: 35 % Americas [19, 26, 28, 30–32], 29 % Africa [17, 18, 21, 23], 18 % South-East Asia [20, 29, 33], 12 % European [22, 27], and 6 % Western Pacific [24]. PLWHA were typically recruited through a clinic (65 %); 35 % were recruited through other venues (e.g., needle-exchange program, churches, research center). Studies were published in peer-reviewed journals in 2012 (88 %) and 2013 (12 %); data were collected between 2001 and 2011. Methodological quality (MQ) of the studies ranged from 10 to 21 (mean = 17 [68 %], SD = 3).

Table 1 Study and sample characteristics of the 17 studies included in the meta-analysis
Full size table

The study samples included 6257 PLWHA who consented to participate in the research. Women comprised nearly half (48 %) of the total sample; average age was 34 years (range = 21 to 45 years; SD = 9). The sample was ethnically diverse with 26 % Black African, 14 % White, 14 % Black non-African, 13 % Southeast Asian, 8 % East Asian, 7 % Hispanic, and the remaining 18 % of other ethnic/racial backgrounds (e.g., American Indian, mixed). Most of the PLWHA were currently sexually active (M = 87 %, SD = 0.20, Mdn = 100 %; n studies = 16). Slightly more than half were on antiretroviral medication (M = 51 %, SD = 0.33, n studies = 13) and 75 % (SD = 0.13, n studies = 4) were adherent to their medication, using definitions specific to individual studies (for details, see Table 1 note). Finally, 24 % (n studies = 13) of PLWHA reported using drugs and few of the studies (16 %, SD = 0.23, n studies = 3) reported IDU.

Prevalence of Alcohol Use and Sexual Behaviors among PLWHA

Results for studies reporting alcohol use and sexual risk behaviors are provided in Table 2. Because all effects were marked by substantial heterogeneity, analyses followed random-effects assumptions. Examination of I 2 confirmed high levels of heterogeneity.

Table 2 Prevalence estimates of alcohol use and sexual risk behaviors among individuals infected with HIV†
Full size table

Alcohol Use

The prevalence of any alcohol consumption among PLWHA is 37 % (95 % CI: 25, 51, Z = -1.88, p = .056). Alcohol use before sex was reported in 33 % of the sample (95 % CI: 18, 52, Z = -1.76, p = .079).

Sexual Risk Behavior

The prevalence of condom use and condom use during the last sexual occasion was 67 % (95 % CI: 56, 76, Z = 2.92, p = .003) and 76 % (95 % CI: 62, 86, Z = 3.42, p = .001), respectively. For sexual partners, the prevalence of having multiple sexual partners was 25 % (95 % CI: 11, 48, Z = -2.15, p = .03).

Moderators of Alcohol Use and Sexual Behaviors Among PLWHA

Moderator tests were conducted to examine whether a priori determined study (WHO region [45]) and sample (proportion women, proportion alcohol or drug users, proportion MSM, proportion of participants with multiple partners, proportion currently on ART, proportion adhering to HIV medication) characteristics related to the variability in prevalence estimates. Due to insufficient sample size (k ≤ 5), moderator tests were conducted only for alcohol use, alcohol use before sex, condom use, and multiple sex partners.

Alcohol Use

Alcohol use was more prevalent among studies conducted in the Americas (35 %; 95 % CI = 32, 38; k = 5) or South-East Asia (36 %; 95 % CI = 32, 40; k = 4) compared to Africa (27 %; 95 % CI = 25, 29; k = 5), Q B (2) = 28.29, p < .001. (Only a single study [22] reported alcohol use in the European region; it was excluded from analyses to avoid misrepresenting the prevalence of alcohol use in Europe.) The prevalence of alcohol consumption was higher when studies sampled fewer women (β = -0.69, p < .001, k = 15), more MSM (β = 0.71, p < .001, k = 14), drug users (β = 0.71, p < .001, k = 11), and participants reporting multiple sexual partners (β = 0.28, p < .001, k = 9). Alcohol use was less prevalent among PLWHA who were currently on ART (β = -0.49, p < .001, k = 12) and among those adherent to their HIV medication (β = -0.51, p = .05, k = 4).

Alcohol Use Before Sex

Alcohol use before sex was more prevalent among studies conducted in Africa (40 %; 95 % CI = 33, 47; k = 3) compared to South-East Asia (22 %; 95 % CI = 18, 27; k = 4), Q B (1) = 19.32, p < .001. (Two studies, one conducted in the European region [27] and another in the Western Pacific region [20], were excluded from the analyses.) The prevalence of alcohol use before sex was higher when studies sampled more women (β = 0.13, p = .043, k = 9), fewer MSM (β = -0.44, p < .001, k = 8), more drug users (β = 0.49, p < .001, k = 6), and participants who reported having multiple sexual partners (β = 0.79, p < .001, k = 7). Alcohol use before sex was less prevalent among PLWHA who were currently on ART (β = -0.60, p < .001, k = 8) and adherent to their HIV medication (β = -0.99, p = .010, k = 3).

Condom Use

Condom use was more prevalent among studies conducted in South-East Asia (76 %; 95 % CI = 71, 80; k = 4) and the Americas (73 %; 95 % CI = 70, 76; k = 4) compared to Africa (64 %; 95 % CI = 59, 68; k = 2), Q B (2) = 15.02, p < .001. (Two studies, one conducted in the European region [27] and another in the Western Pacific region [20], were excluded from the analyses.) The prevalence of condom use was higher when studies sampled more women (β = 0.17, p = .007, k = 12), fewer MSM (β = -0.48, p < .001, k = 11), fewer alcohol users (β = -0.67, p < .001, k = 12), fewer drug users (β = -0.19, p = .003, k = 10), and fewer participants who reported having multiple sexual partners (β = -0.64, p < .001, k = 9). Condom use was more prevalent among PLWHA who were currently on ART (β = 0.57, p < .001, k = 10). Self-reported HIV medication adherence was not a significant predictor of the prevalence of condom use (β = -0.05, p = .779, k = 4).

Multiple Sexual Partners

The prevalence of multiple sexual partners was greater among studies conducted in the Americas (50 %; 95 % CI = 46, 53; k = 4) compared with Africa (17 %; 95 % CI = 12, 23; k = 2) and South-East Asia (14 %; 95 % CI = 11, 18; k = 3), Q B (2) = 145.15, p < .001. (Two studies, one conducted in the European region [27] and another in the Western Pacific region [20], were excluded from the analyses.) The prevalence of participants with multiple sexual partners was higher when studies sampled fewer women (β = -0.22, p < .001, k = 11), more MSM (β = 0.24, p < .001, k = 11), more alcohol users (β = 0.21, p < .001, k = 11), and fewer drug users (β = -0.15, p < .001, k = 8). Multiple sexual partners were less prevalent among PLWHA who were currently on ART (β = -0.23, p < .001, k = 10) and were adherent to their HIV medication (β = -0.71, p < .001, k = 5).

Association between Alcohol Use and Unprotected Sex among PLWHA

Five studies [19, 20, 22, 24, 31] of the 17 studies (N = 1,589) assessed the association between any alcohol consumption (yes, no) and unprotected sex (yes, no). Alcohol use was associated with unprotected sex, d + = 0.23 (95 % CI = 0.05, 0.40). The effect sizes were homogeneous indicating consistency across studies, Q(4) = 4.50, p = .34, I 2 = 11 % (95 % CI = 0, 53).

Discussion

This systematic review and meta-analysis of 17 recent studies assessing alcohol use and sexual risk behaviors among nearly 6300 PLWHA shows that approximately one-third of PLWHA use alcohol and engage in sexual risk behaviors. Our findings also show that any alcohol consumption is associated with unprotected sex among PLWHA (d + = 0.23). The magnitude of the effect size for the alcohol-risky sex association was small to medium and is consistent with an earlier review of the literature (see [15], d + = 0.30, k = 9). Although these global association studies (i.e., studies that assess the extent to which an individual’s overall frequency of alcohol use and sexual risk behavior are related) cannot establish a causal link between alcohol and risky sexual behavior, our findings provide support for the growing consensus that alcohol use is an important risk factor for the transmission of HIV [12•, 46]. The studies included in the meta-analysis represent diverse samples of PLWHA from multiple geographical regions and significant heterogeneity was observed for the prevalence estimates. Weighted regression analyses indicated that the variability across effect sizes is moderated by study and sample characteristics (discussed below). Overall, our findings provide on-going support for the involvement of alcohol use in sexual risk-taking among PLWHA.

The results of our moderator tests show that geographical region is an important predictor of the prevalence of alcohol use and sexual risk behaviors among PLWHA. The prevalence of alcohol consumption was higher among the WHO regions of the Americas and South-East Asia compared to Africa but the prevalence of condom use was lower in Africa relative to the Americas and South-East Asia. Furthermore, alcohol use before sex was higher in Africa relative to South-East Asia (40 % vs. 22 %). These findings corroborate prior reviews that showed a high prevalence of alcohol use in sexual contexts in Africa [47, 48]. High-risk alcohol use may also be associated with the lower prevalence of condom use among Africans [47]. Greater attention of the influence of alcohol use on sexual risk-taking, especially among PLWHA in Africa, is warranted.

Moderator tests suggest that sample characteristics are important predictors of the prevalence of alcohol use and sexual risk behaviors among PLWHA. Consistent with prior research [49], alcohol use was less prevalent among studies that sampled women (vs. men). In contrast, women were more likely to consume alcohol prior to sex, perhaps because women consume alcohol as a means to reduce social inhibitions [50•]. MSM and drug users report high rates of alcohol use and sexual risk behaviors. The prevalence of condom use was lower among MSM and drug users; nonetheless, unlike drug users, MSM were more likely to have multiple sexual partners. Overall, alcohol users were more likely to report less condom use and more sexual partners. The pattern of results suggests that alcohol use and sexual risk behaviors may be more prevalent for certain populations and, overall, suggests that interventions tailored to subpopulations of PLWHA can help to reduce alcohol use and sexual risk behaviors that enhance HIV transmission [4].

Antiretroviral therapy (ART) can play an important role in preventing the transmission of HIV. “Treatment as prevention” has garnered much attention due to the success of the National Institute of Health HPTN 052 trial, which found that ART for PLWHA reduced HIV transmission by 96 % between serodiscordant heterosexual couples [51]. The role of ART in reducing sexual risk behaviors in the broader community has been largely absent from subsequent discussions (see [52•] for a commentary on this issue). Moderator results from the current meta-analysis suggest that PLWHA who are currently on ART consume less alcohol and engage in fewer sexual risk behaviors. Of the few studies (k = 5) reporting relevant data, adherence to HIV medication was also associated with lower alcohol use, alcohol use before sex, and fewer sexual partners. Thus, it appears that engagement in medical care leads not only to a reduction in infectiousness but may also be associated with less risky behavior.

Limitations

Several limitations should be considered when interpreting these findings. First, using electronic bibliographic databases to identify relevant studies are restricted by publication source and authors’ choice of keywords [53]. Second, our meta-analysis was restricted to studies published between January 1, 2012 and April 1, 2013, which may not be representative of other studies of PLWHA. Nonetheless, our findings were consistent with prior reviews of the earlier literature with PLWHA [5, 6, 15]. Third, all outcomes involve self-reports, which are vulnerable to measurement, cognitive (e.g., memory), and social (e.g., self-presentation) biases [54]. Researchers typically use methods to minimize these biases and maximize data quality [55]. Fourth, our moderator tests were limited to the data available in the individual studies. Inconsistencies in reporting ‘problematic alcohol use’ (see description of measures in Table 1 note) as well as clinical and immunological markers (e.g., CD4 counts, viral load) prevented us from fully exploring potentially relevant moderators that may elucidate our findings. Finally, causality cannot be determined from the available data. The relation between alcohol and sexual risk may be explained by ‘third variables’ (e.g., partner drinking, drinking environment).[56]

Conclusion

Alcohol use is a prominent risk factor in the transmission of HIV. Our meta-analysis confirms that PLWHA frequently use alcohol and engage in sexual risk behaviors and that alcohol use is associated with sexual risk-taking. The accumulated evidence suggests that efforts to address the co-occurrence of alcohol use and sexual behavior are needed to reduce the transmission of HIV.