Alcohol consumption is generally believed to be a determinant of sexual risk behaviors and other drug use and, therefore, an indirect determinant of HIV acquisition [1,2]. Alcohol consumption is also hypothesized to compromise the immune system, which in turn could allow for increased risk of HIV infection [3–5]. However, more than two decades after the onset of the HIV epidemic, the extant literature [3,6–8] regarding the association between alcohol consumption and HIV acquisition remains inadequate to inform public health policy or preventive interventions. Lack of definitive evidence for a link between alcohol consumption and HIV acquisition is of great concern in part because many populations at highest risk for HIV infection also consume high levels of alcohol, such as injection drug users , men who have sex with men [7,10], and sub-Saharan Africans [8,11].
Most existing studies using prospective data to examine whether alcohol consumption increases the risk of HIV infection have not accounted for expected confounding by sexual activity and drug use [8,12,13]. Studies [14–16] that have accounted for time-updated reports of these potential confounders by standard adjustment ignore the fact that alcohol consumption may affect subsequent sexual activity and drug use [17,18]. If sexual activity and drug use are time-varying confounders that are affected by prior alcohol consumption, then standard adjustment for time-updated sexual activity and drug use is flawed. Specifically, such an approach may remove the indirect effect of alcohol consumption on HIV acquisition mediated through these behaviors [19–21], as well as induce a selection bias [22,23]. Therefore, appropriate statistical methods are needed to obtain unbiased estimates of the effect of alcohol consumption on HIV acquisition. Marginal structural models allow for estimation of total (direct and indirect) effects while appropriately accounting for time-varying confounding without inducing selection bias .
Here we use rich, prospective data on over 1500 African–American injection drug users from the AIDS Link to Intravenous Experience (ALIVE) cohort study  and marginal structural models to characterize the association between alcohol consumption and risk of HIV infection.
The ALIVE cohort enrolled 3627 adults in Baltimore, Maryland between 1988 and 2008. Potential participants were recruited through extensive community outreach and were screened for inclusion based on a history of active or recent injection drug use. Among the 3627 enrollees, 1683 were seronegative at entry and had at least one seronegative follow-up visit subsequent to baseline. One enrollee was excluded due to a lack of demographic information. The remaining 157 non-African–Americans were excluded from the analysis due to small numbers, leaving 1525 African–American participants.
The committee on human research at the Johns Hopkins Bloomberg School of Public Health approved study protocols and informed consent forms, which were completed by all study participants. Participants attended a study visit every 6 months at a University-affiliated, freestanding clinic. At visits, participants provided blood and completed an interviewer-administered questionnaire.
Ascertainment of HIV infection
The outcome of interest was incident infection with HIV. Participants were followed from their first visit through HIV seroconversion, death, or their last follow-up visit before 1 January 2008. HIV seroconversion was determined from participants' blood specimens, which were tested for HIV antibodies by enzyme-linked immunoassay at each visit: reactive specimens were confirmed by western blot. Dates of death were ascertained from death certificates. Participants not observed for any period of more than 1 year were censored at the minimum of their date of death (if applicable), 1 year after their last visit prior to nonattendance, and 1 January 2008. Censored participants were classified as lost to follow-up if they were lost prior to 1 January 2007 and not seen at a subsequent ALIVE visit.
Assessment of alcohol consumption and covariates
The study questionnaire elicited information on demographic characteristics, drug and alcohol use history and practices, and sexual behaviors within the prior 6 months. Alcohol consumption, as the number of drinks per week, was obtained as the product of the reported number of drinking days in a typical week and the usual number of drinks per drinking day. A drink was defined to be a can, bottle or glass of beer, a glass of wine, a shot of liquor, a mixed drink with a shot of liquor, or any other kind of alcoholic beverage. Less than 3% of alcohol consumption reports were of more than 20 drinks per drinking day and were set to 20 drinks per drinking day. For analysis, alcohol consumption is quantified as the number of drinks per week averaged over the prior four visits (approximately 2 years), as well as binge drinking in the prior 6 months. We averaged the prior four visits to reduce random error and capture both short-term and moderate-term alcohol effects. Average drinks per week over the prior 2 years were categorized as 0, 1–5, 6–20, 21–50, and 51–140 to divide the distribution of HIV cases into approximate quintiles. Binge drinking was considered to be drinking at least once a week and consuming more than five drinks on a given drinking day. In secondary analyses, we explored a four-level composite measure of drinks per week and binge drinking with categories defined as follows: 0 drinks per week, 1–20 drinks per week with no binge drinking, 1–20 drinks per week with binge drinking, and 21–140 drinks per week. We also explored alternate exposure windows, namely, drinks per week in the most recent 6 months and cumulatively averaged over the entire follow-up.
Sex, age, and years of formal education were obtained at the first visit for all participants. Number of male and female sexual partners, self-reported sexually transmitted infections (STIs; i.e., genital herpes simplex virus, genital warts, gonorrhea, and syphilis), cocaine use, shooting gallery attendance, and number of drug injections per day were obtained at each visit with reference to the prior 6 months, based on prior research . Less than 3% of the number of sexual partner reports were more than eight, and were set to eight partners. Relatively rare missing data (2, 7, <1, <1, <1, and <1% on alcohol consumption, sexual partners, cocaine, shooting gallery attendance, injections per day, and STIs, respectively) were set to zero. Inferences were invariant if missing data were set to observed medians or modes.
Characteristics of participants at study entry and averaged over follow-up are presented as percentages or medians and quartiles, as appropriate. The association between drinks per week and binge drinking was assessed using Spearman's rank correlation. Incidence rates were calculated as the number of HIV cases divided by the number of person-years at risk.
Hazard ratios were used to quantify associations between alcohol consumption and HIV incidence; 95% confidence limits were used to quantify precision. Hazard ratios were obtained from Cox proportional hazard models  with time-on-study as the time scale. Confidence limits were obtained using the standard large-sample approximation for the variance in crude and adjusted data, and the robust variance  for weighted data (details below). Drinks per week was included in the Cox model as either indicator variable for categories or a restricted quadratic spline with knots at the 5, 35, 65, and 95th percentiles. Wald χ 2 trend tests were used across groups with the median assigned for categories. The complement of the weighted Kaplan–Meier  curve is presented by categories of drinks per week. Proportional hazards was assessed by a statistical test of the product term between the indicator for the highest drinks per week category and time (P value = 0.932) as well as log time (P value = 0.567); there was no evidence of nonproportionality.
Observed data were weighted by the product of stabilized inverse probability-of-exposure-and-censoring weights to account for confounding and selection biases by measured characteristics. Weights were multiplied over time to account for histories of exposure and censoring. Drinks per week at each visit, as described previously, was modeled using a cumulative logistic regression model , whereas death and dropout were modeled using logistic regression models. All logistic models were pooled over visits . Covariates included age at entry, sex, years of education, time-varying drug use (i.e., cocaine use, shooting gallery attendance, and injections per day), and sexual activity (i.e., sexual partners and STIs). Time-varying covariates were lagged one visit. Continuous covariates were fit using splines as defined previously. Weights were stabilized to improve efficiency by a function of drinks per week, age, and education. The resultant weights had a mean (SD) of 1.02 (0.33) with a range from 0.21 to 7.33. To estimate incidence rates and assess nonlinearity in the association between alcohol consumption and HIV acquisition, weights were stabilized solely by a function of drinks per week. These latter weights had a mean (SD) of 1.03 (0.43) with a range from 0.12 to 9.17. An adjusted model is provided for comparison and accounts for the same confounders as the weighted model as well as concurrent values of time-varying confounders to reflect practices in the existing literature .
Based on the observed number of HIV cases and distribution of drinks per week, we had 80% statistical power to detect a crude hazard ratio of 1.58 comparing drinkers to nondrinkers for a sample size of 1525, wherein 25% of the sample population is nondrinkers, the cumulative incidence of HIV at the end of follow-up among nondrinkers is 0.15, and the two-sided type I error is 0.05. All analyses were conducted using SAS (SAS Institute, Cary, North Carolina, USA).
Among the 1525 participants followed for 8181 person-years, 155 acquired HIV. Among the remaining 1370 HIV-seronegative participants who were censored, 127 died, 559 were censored due to an interval of 1 year or more without a study visit, 362 were lost to follow-up, and 322 completed follow-up alive. Loss to follow-up and censoring due to missed visits were greater among participants who did not report cocaine use (hazard ratio = 1.22; 95% confidence limits: 1.02–1.45) as well as those who reported more than one sexual partner (1.25; 1.02–1.53) and injected more than once a day (1.56; 1.30–1.88). Although nonsignificant, there was a trend toward poorer follow-up among participants with less alcohol consumption [e.g., 1.25; 0.95–1.64 for 0 drinks per week versus 51–140 drinks per week] (see supplementary Appendix Table S1, http://links.lww.com/QAD/A97). Such observed differences in loss to follow-up and censoring due to missed visits were accounted for in the analysis.
Table 1 describes the characteristics of participants at study entry and over follow-up. At entry, 28% of 1525 participants were women with a median (quartiles) age of 37 (32–42) years and 10 (10–12) years of education. Consumption of alcohol and use of illicit drugs was heavy at entry and during follow-up, though less so during follow-up. Male sex with men and reported STIs were rare, but multiple sex partners were common.
Figure 1 depicts alcohol consumption by years of follow-up. Alcohol consumption decreased with increasing time since enrollment. Seventy-six and 23% of the person-years occurred while participants were consuming any alcohol in the prior 2 years or binge drinking in the prior 6 months, respectively. At entry and over follow-up, participants who had a higher number of drinks per week in the prior 2 years also tended to be binge drinkers. Specifically, among those reporting 0 drinks per week, 1–5 drinks per week, 6–20 drinks per week, 21–50 drinks per week, and 51–140 drinks per week in the prior 2 years over follow-up, 0, 2, 21, 55, and 87% reported binge drinking in the prior 6 months, respectively. The rank correlation for number of drinks per week in the prior 2 years and binge drinking in the prior 6 months was 0.60.
Figure 2 shows the weighted hazard ratio of HIV acquisition with 95% confidence limits by number of drinks per week in the prior 2 years. The hazard ratio appears to nonlinearly increase with the number of drinks per week. Crude and weighted hazard ratios with 95% confidence limits for alcohol consumption on HIV acquisition are shown in Table 2. In crude analyses, hazard ratios for participants reporting 1–5, 6–20, 21–50, and 51–140 drinks per week in the prior 2 years compared to participants who reported 0 drinks per week were 1.22 (95% confidence limits: 0.69–2.15), 1.41 (0.82–2.42), 2.06 (1.20–3.54), and 2.96 (1.67–5.23), respectively. The weighted marginal structural Cox model resulted in hazard ratios of 1.09 (0.60–1.98), 1.18 (0.66–2.09), 1.66 (0.94–2.93), and 2.12 (1.15–3.90), respectively. The P values for the Wald tests-of-trend are also shown in Table 2: the P values for the crude and weighted tests were 8.2 × 10–6 and 9.7 × 10–4, respectively. Figure 3 illustrates cumulative proportions of HIV-positive participants over years of follow-up stratified by alcohol consumption categories for the number of drinks per week in the prior 2 years. The cumulative proportion of HIV-positive participants increased with greater alcohol consumption.
As also shown in Table 2, the weighted hazard ratio for binge drinking was 1.70 (1.22–2.37). For the composite metric, light-to-moderate drinking (i.e. 1–20 drinks per week) in the absence of binge drinking did not demonstrate an elevated hazard of HIV acquisition in the weighted analysis. Despite not achieving statistical significance, light-to-moderate drinking in the presence of binge drinking did appear to elevate the hazard of HIV acquisition. As shown in both the crude and weighted analyses in supplementary Appendix Table S2 (http://links.lww.com/QAD/A97), the association between alcohol consumption in the prior 6 months and HIV acquisition was similar to that of consumption in the prior 2 years and cumulative over follow-up.
For comparison, standard adjusted analysis did not demonstrate a statistically significant association for the highest drinks per week category compared to the lowest, whereas the weighted analysis did. Adjustment diminished the crude hazard ratios for drinks per week in the prior 2 years to 1.00 (0.56–1.79), 1.07 (0.61–1.87) 1.36 (0.76–2.41), and 1.70 (0.92–3.14), respectively. The P value for the Wald test-of-trend for the standard adjusted analysis was 0.02.
During 20 years of follow-up, 155 of 1525 African–American injection drug users in the ALIVE cohort acquired HIV. Alcohol consumption was highly prevalent and heavy in this cohort, but lessened with time. The marginal structural model analysis indicated a strong dose–response relationship between alcohol consumption and subsequent HIV acquisition, independent of prior drug use and sexual activity. The hazard ratios for alcohol consumption were on par with some prior identified risk factors for HIV acquisition in this cohort such as younger age, not completing high school, shooting gallery attendance, self-reported STIs, and cocaine use, but weaker than frequent injections and male homosexual behavior .
Weaker evidence for a dose–response relationship of alcohol on HIV acquisition was obtained from standard adjusted analysis. Specifically, the excess hazard from the weighted analysis was 38% (=1–0.70/1.12) stronger than results from the standard adjusted model. In standard analysis, indirect effects of alcohol use on HIV infection mediated through sexual activity and drug use were likely blocked with adjustment for these time-varying confounders affected by prior alcohol use.
The role of binge drinking in HIV infection remains unclear, particularly in the setting of an injection drug using population. Given the strong correlation between number of drinks per week and binge drinking, the observed association between number of drinks per week and HIV acquisition may be partially explained by binge drinking. The fact that the independent association between number of drinks per week in the prior 2 years and HIV acquisition was muted in the presence of binge drinking in the prior 6 months provides evidence for a role of binge drinking. However, given that binge drinking is likely to have acute effects and in this analysis the strength of the observed association between number of drinks per week and HIV acquisition was independent of the exposure window, the effect of number of drinks per week may not be explained entirely by binge drinking.
As suggested by Dingle and Oei , alcohol consumption may lead to lowered inhibition and in turn increase risky sexual and drug use behaviors. Steele and Josephs  similarly suggest alcohol may impair decision-making and in turn hinder risk reduction techniques. The low observed prevalence of STIs may be due to underreporting. However, if the STI prevalence is indeed low, alcohol consumption may not be working primarily through risky sexual behavior to increase the hazard of HIV infection in this population of injection drug users. Instead, high-risk drug use practices may be the primary mediating factor. Alcohol may also work to compromise the immune system and in turn allow for increased risk of HIV infection [3–5]. In a series of studies, Bagasra et al. [30–32] showed alcohol to increase HIV-1 replication in human peripheral blood mononuclear cells. This increase in HIV-1 replication coincided with lower production of soluble immune response suppressor activity and interleukin-2 attributed to functional impairment of both suppressor (CD8+) and helper (CD4+) T lymphocytes. CD4+ and CD8+ T-lymphocytes regulate the immune response. CD8+ T lymphocytes have been shown to inhibit HIV-1 replication.
There are limitations to the present research. For valid inference, we must assume the absence of unmeasured confounding as well as no informative censoring by unmeasured factors. If unmeasured factors for confounding or informative censoring exist, then the reported incidence rates and hazard ratios will be biased. Admittedly, the indicators of high-risk sexual activity and drug use behaviors such as number of sexual partners, any sexually transmitted infections, as well as attendance at a shooting gallery may not fully capture the extent of risk behaviors in this population of injection drug users. Therefore, residual confounding of effect estimates may still be present. If sexually transmitted infections were more underreported in heavy drinkers or heavy drinkers were more likely to share drug injection equipment, then this residual confounding would bias effect estimates away from the null. Bias can also occur in the presence of nonpositivity, model misspecification, or lack of consistency . We do not account for the fact that variables as assessed are typically imperfect measures of the true underlying characteristic. In addition, many heavy drinkers were HIV-infected at entry. If heavy drinkers who were HIV-negative at entry were less prone to risky sex and drug use activities as compared to heavy drinkers who seroconverted prior to entry, then the association between heavy drinking and HIV acquisition would be muted.
There are several strengths to the present work. Use of prospective data helps ensure temporal order between alcohol consumption and HIV infection. Lagging covariates further facilitates temporal order for confounders, and in turn proper causal inference. Use of time-updated reports minimizes systematic bias due to measurement error . Use of marginal structural models avoids bias associated with standard adjustment techniques in the presence of time-varying confounders affected by prior exposure . Finally, randomized evidence is not feasible due to ethical concerns, or lack of compliance in the case of risk reduction trials [35,36]. Without randomized evidence, thorough analysis of prospective observational studies with repeated assessments of exposures and outcomes provides the best evidence for estimation of etiologic effects.
Our findings provide compelling evidence for a dose–response relationship between alcohol consumption and HIV acquisition in populo. This evidence lends support for the enhancement of HIV risk reduction strategies with alcohol-specific interventions, including incorporation of alcohol-related prevention into programs tailored for substance users and for prevention programs among HIV positives . It is estimated that lowering drinking from the highest drinks per week category (i.e., 51–140) to the lowest (i.e., 0) for two-thirds of the person-years contributed in the highest category would reduce HIV incidence in this population of injection drug users by 29%. Future work should thoroughly explore the pathways by which alcohol consumption increases risk of HIV infection. Identifying mediating factors will be central to identifying new targets for HIV prevention interventions. In addition, marginal structural models should be used to examine the association between alcohol consumption and HIV acquisition in other high-risk populations such as men who have sex with men and sub-Saharan Africans.
All authors contributed to the design of the study. C.J.H. and S.R.C. undertook the analysis and drafted the article. All authors provided feedback on drafts and approved the final version.
This work was supported by the National Institute on Alcohol Abuse and Alcoholism through R01-AA-01759 and the National Institute on Drug Abuse through R01-DA-04334 and R01-DA-12568.
The authors would like to thank Ms Jacquie Astemborski for assistance with the ALIVE data, Dr James Robins and Ms Petra Sander for expert advice, and the ALIVE study staff and participants.
There are no conflicts of interest.
Data presented previously at 14th International Workshop on HIV Observational Databases in Sitges, Spain, abstract 14_38, 3/25/10 and at the 43rd Annual Meeting of the Society for Epidemiologic Research, abstract 557, 6/25/10. Data previously published as an abstract in Am J Epidemiol, 1 June 2010; 171:S1–S157.
1. Steele CM, Josephs RA. Alcohol myopia. Its prized and dangerous effects. Am Psychol 1990; 45:921–933.
2. Dingle GA, Oei TP. Is alcohol a cofactor of HIV and AIDS? Evidence from immunological and behavioral studies. Psychol Bull 1997; 122:56–71.
3. Shuper PA, Neuman M, Kanteres F, Baliunas D, Joharchi N, Rehm J. Causal considerations on alcohol and HIV/AIDS: a systematic review
. Alcohol Alcohol
4. Szabo G. Alcohol's contribution to compromised immunity. Alcohol Health Res World 1997; 21:30–41.
5. Wang Y, Watson RR. Is alcohol consumption a cofactor in the development of acquired immunodeficiency syndrome? Alcohol 1995; 12:105–109.
6. Fisher JC, Bang H, Kapiga SH. The association between HIV infection and alcohol use: a systematic review and meta-analysis of African studies. Sex Transm Dis 2007; 34:856–863.
7. Woolf SE, Maisto SA. Alcohol use and risk of HIV infection among men who have sex with men. AIDS Behav 2009; 13:757–782.
8. Baliunas D, Rehm J, Irving H, Shuper P. Alcohol consumption and risk of incident human immunodeficiency virus infection: a meta-analysis
. Int J Public Health
9. Galai N, Safaeian M, Vlahov D, Bolotin A, Celentano DD. Longitudinal patterns of drug injection behavior in the ALIVE Study cohort, 1988-2000: description and determinants. Am J Epidemiol 2003; 158:695–704.
10. Stall R, Paul JP, Greenwood G, Pollack LM, Bein E, Crosby GM, et al
. Alcohol use, drug use and alcohol-related problems among men who have sex with men: the Urban Men's Health Study. Addiction 2001; 96:1589–1601.
11. World Health Organization. Alcohol use and sexual risk behaviour: a cross-cultural study in eight countries
. Geneva: World Health Organization; 2005.
12. Howe CJ, Sander PM, Plankey MW, Cole SR. Effects of time-varying exposures adjusting for time-varying confounders: the case of alcohol consumption and risk of incident human immunodeficiency virus infection
. Int J Public Health
13. Sander PM, Cole SR, Ostrow DG, Mehta SH, Kirk GD. Determinants of alcohol consumption in HIV-uninfected injection drug users
. Drug Alcohol Depend
14. Celentano DD, Nelson KE, Suprasert S, Eiumtrakul S, Tulvatana S, Kuntolbutra S, et al
. Risk factors for HIV-1 seroconversion among young men in northern Thailand. JAMA 1996; 275:122–127.
15. Plankey MW, Ostrow DG, Stall R, Cox C, Li X, Peck JA, et al
. The relationship between methamphetamine and popper use and risk of HIV seroconversion in the multicenter AIDS cohort study. J Acquir Immune Defic Syndr 2007; 45:85–92.
16. Wang C, Vlahov D, Galai N, Cole SR, Bareta J, Pollini R, et al
. The effect of HIV infection on overdose mortality. AIDS 2005; 19:935–942.
17. Stein MD, Charuvastra A, Anderson B, Sobota M, Friedmanna PD. Alcohol and HIV risk taking among intravenous drug users. Addict Behav 2002; 27:727–736.
18. Zule WA, Flannery BA, Wechsberg WM, Lam WK. Alcohol use among out-of-treatment crack using African-American women. Am J Drug Alcohol Abuse 2002; 28:525–544.
19. Robins J. A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. J Chronic Dis 1987; 40(Suppl 2):139S–161S.
20. Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000; 11:550–560.
21. Hernán MA, Brumback B, Robins JM. Marginal structural models to estimate the joint causal effect of nonrandomized treatments. JASA 2001; 96:440–448.
22. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology 2004; 15:615–625.
23. Cole SR, Hernan MA. Fallibility in estimating direct effects. Int J Epidemiol 2002; 31:163–165.
24. Vlahov D, Anthony J, Munoz A, Margolick J, Nelson K, Celentano D, et al
. The ALIVE study, a longitudinal study of HIV-1 infection in intravenous drug users: description of methods. NIDA Res Monogr 1991; 109:75–100.
25. Nelson KE, Galai N, Safaeian M, Strathdee SA, Celentano DD, Vlahov D. Temporal trends in the incidence of human immunodeficiency virus infection and risk behavior among injection drug users in Baltimore, Maryland, 1988–1998. Am J Epidemiol 2002; 156:641–653.
26. Cox DR. Regression models and life tables. J R Statist Soc (B) 1972; 34:187–220.
27. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53:457–481.
28. Armstrong BG, Sloan M. Ordinal regression models for epidemiologic data. Am J Epidemiol 1989; 129:191–204.
29. D'Agostino RB, Lee ML, Belanger AJ, Cupples LA, Anderson K, Kannel WB. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med 1990; 9:1501–1515.
30. Bagasra O, Kajdacsy-Balla A, Lischner HW. Effects of alcohol ingestion on in vitro susceptibility of peripheral blood mononuclear cells to infection with HIV and of selected T-cell functions. Alcohol Clin Exp Res 1989; 13:636–643.
31. Bagasra O, Kajdacsy-Balla A, Lischner HW, Pomerantz RJ. Alcohol intake increases human immunodeficiency virus type 1 replication in human peripheral blood mononuclear cells. J Infect Dis 1993; 167:789–797.
32. Bagasra O, Bachman SE, Jew L, Tawadros R, Cater J, Boden G, et al
. Increased human immunodeficiency virus type 1 replication in human peripheral blood mononuclear cells induced by ethanol: potential immunopathogenic mechanisms. J Infect Dis 1996; 173:550–558.
33. Cole SR, Hernan MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol 2008; 168:656–664.
34. Hu FB, Stampfer MJ, Rimm E, Ascherio A, Rosner BA, Spiegelman D, et al
. Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. Am J Epidemiol 1999; 149:531–540.
35. Kalichman SC, Simbayi LC, Vermaak R, Cain D, Jooste S, Peltzer K. HIV/AIDS risk reduction counseling for alcohol using sexually transmitted infections clinic patients in Cape Town, South Africa. J Acquir Immune Defic Syndr 2007; 44:594–600.
36. Kalichman SC, Simbayi LC, Vermaak R, Cain D, Smith G, Mthebu J, et al
. Randomized trial of a community-based alcohol-related HIV risk-reduction intervention for men and women in Cape Town South Africa. Ann Behav Med 2008; 36:270–279.
37. Centers for Disease Control and Prevention. Prevention is care. Atlanta: Centers for Disease Control and Prevention; 2009. http://www.cdc.gov/PreventionISCare
. [Accessed 27 January 2010].