Understanding behavioral changes in response to HIV seroconversion and treatment could provide crucial inputs into HIV policymaking. Yet, there is little consensus regarding the ways in which HIV seroconversion impacts subsequent health behaviors, particularly in the context of evolving HIV treatment access and actual use. Prior research has identified various behavioral responses to HIV seroconversion and diagnosis including increased and decreased risk of onward HIV transmission [1–4]. Previous studies on the impact of HIV treatment access on risk behaviors have also had divergent results [4–8], limiting our understanding of how HIV seroconversion and treatment interact to shape subsequent health behaviors. With global efforts to improve access to HIV testing and treatment technologies , the question of how individuals behave following HIV seroconversion in contexts with and without adequate treatment access has become increasingly relevant for public policy.
There are at least three possible ways that HIV seroconversion could impact subsequent health behaviors and health status. First, seroconversion could reduce the perceived cost of engaging in risk behaviors associated with increased onward HIV transmission (e.g. risky sexual behaviors, recreational use of poppers), thereby increasing engagement in these behaviors . Second, seroconversion may increase risk for certain mental health disorders  that could induce coping behaviors (e.g. heavy drinking, marijuana use). Third, increased drinking and recreational drug use following seroconversion could interfere with HIV medication use , which, along with smoking, could worsen physical health among people living with HIV .
Debates on behavioral responses to HIV treatment and new biomedical prevention methods have focused on the possibility of ‘risk compensation’, the notion that improved access to antiretrovirals reduces the perceived cost of engaging in risk behaviors including sexual risk behaviors for onward HIV transmission . Thus, by reducing mortality and morbidity from HIV [13,14], HAART access and use following seroconversion could trigger increased engagement in transmission risk behaviors. However, it is also possible that by introducing a sense of hope and increasing the perceived benefits of reducing engagement in harmful behaviors, treatment access and use following seroconversion could induce positive changes in transmission risk behaviors and substance use (e.g. heavy drinking, smoking, recreational drug use behaviors that are known to be harmful to health). This ‘competing risks’ hypothesis implies that individuals avoid harmful behaviors to reduce other mortality risks that would undermine the added health benefits of HIV treatment .
Our article sheds new light on behavioral responses to HIV seroconversion and treatment using rich longitudinal data collected from men who have sex with men (MSM) during the historical expansion of HAART in the United States. Among participants who seroconverted during observation (i.e. ‘seroconverters’), we assess changes in health behaviors in response to seroconversion and then compare these behavioral responses before and after the availability and actual use of HAART, which contributes to viral suppression  and dramatically improves individual health [13,14]. Specific behaviors of interest include those that promote onward HIV transmission (i.e. risky sexual behaviors, recreational use of poppers), behaviors indicative of mental status and potential self-medication or coping (i.e. heavy drinking, marijuana use), and substance use behaviors harmful to general health and well being (i.e. heavy drinking, smoking).
A key innovation of our study is the use of fixed effects that account for person-specific determinants of risk behaviors. We focus on a sample of seroconverters in which each person can be used as their own control, allowing us to credibly isolate the associations of HIV seroconversion with subsequent behaviors while accounting for possible unobservable time-invariant determinants of individual risk.
Materials and methods
We drew from the Multicenter AIDS Cohort Study (MACS; Public Data Set Release P21, 1 November 2012), an ongoing study of HIV and health behaviors among US MSM . With 49 semiannual study visits conducted from 1984 to 2008, a time period characterized by considerable advances in HIV treatment technology, MACS provides a unique opportunity to study behavioral responses to seroconversion with treatment access and use . Eligible MSM at least 18 years old were enrolled during three recruitment period: 1984–1985, 1987–1991, and 2001–2003. In the first period, when HIV lab tests were not available, men were enrolled if they were free of AIDS-defining illnesses; later, both HIV-infected and age-matched uninfected men were enrolled. Our final analytic sample of 558 individuals (13 018 person-visits) included men who seroconverted during observation (n = 589), were not missing seroconversion dates (n = 14), and were not administratively censored (n = 17). Participants provided written informed consent. Ethics review boards of participating institutions approved all study procedures.
Assessment of outcomes and exposures
All behavioral outcomes were self-reported every 6 months via Audio Computer-Assisted Self-Interview (ACASI) interviews. We dichotomized outcomes for regression analyses to facilitate interpretation of coefficients . Risky sex was operationalized as reporting at least two male sexual partners (including oral sex partners) or insertive or receptive anal sex with at least two male partners (among those with two or more male sexual partners) in the past 6 months. Heavy drinking was defined as consuming at least three drinks when drinking. Heavy smoking was defined as smoking at least 1/2 packs of cigarettes daily. Additional recreational drug use outcomes that were consistently measured included monthly or more frequent use of marijuana or poppers (nitrite inhalants; past 6 months). Questions for all of these outcome variables were administered consistently across the entire study period except questions on risky sex, which started in wave 8.
Exposures of interest included HIV seroconversion (new HIV diagnosis determined via lab tests from blood specimens tested by ELISA and confirmed by western blot) and exposure to HIV treatment, which was operationalized as two binary measures: first, HAART availability, defined using a post-1996 dummy variable ; and second, actual treatment initiation, defined as self-reported initiation of any medication/drug for HIV/AIDS, which could include pre-1996 (i.e. non-HAART) treatments (via ACASI).
We used a fixed effects framework to estimate changes in time-varying risk behaviors following HIV seroconversion with and without exposure to HIV treatment technology. Fixed effects control for individual-specific, time-invariant determinants of risk behaviors that are unobservable but may be confounded with HIV status and treatment exposure such as individual propensity for risk . This approach limited the sample to seroconverters but enabled us to credibly identify within-person changes in behaviors before and after seroconversion rather than relying solely on comparisons across diverse individuals with different HIV statuses.
We first examined the overall effects of HIV seroconversion using a model of the form (1):
where behaviori,t+1 was one of the risk behaviors of interest in t + 1. Postconversioni,t was a dummy variable equal to one for all periods after and including t, the survey wave in which individual i seroconverts. Xit was a vector of covariates including a time trend and its interaction with high school degree, age interacted with high school degree, and a time trend interacted with the baseline level of each corresponding risk behavior. We included these variables to account for idiosyncratic serial correlation in individual behaviors. μi was individual fixed effects as described above. The time dummies δt, removed survey wave fixed effects. Finally, εi,t+1, the idiosyncratic error term, was clustered at the person level to account for correlated unobservable determinants of risk behaviors over time. The parameter of interest, α1, then identified the average within-person difference in risk behaviors between the preconversion and postconversion period.
To examine the role of HIV treatment exposure in shaping behaviors before and after seroconversion, we also estimated an augmented model of the form (2):
where exposurei,t was one of the two variables capturing HIV treatment exposure. All the variables included in Xit in model (1) were also considered here. α1 and α2 were the parameters of interest indicating how risk behaviors respond to seroconversion before and after exposure to treatment, respectively. As a baseline, we estimated Eqs. (1) and (2) using conditional logistic regressions but found similar results using linear probability models based on ordinary least squares.
Further analyses assessed the sensitivity of our results to controlling for time-varying physical and mental health factors that could affect HIV infection, treatment, or observed changes in risk behaviors, including CD4+ cell count, which signals disease progression and influences treatment recommendations, and depression symptomatology [Center for Epidemiological Studies-Depression Scale (CES-D 20)] .
We investigated differential attrition using the following equation:
where attritioni,t+1 was the indicator of dropping out at t + 1 conditional on being surveyed at time t. Behaviori,s,t was risk behavior of type s at time t. We estimated Eq. (3) with risk behaviors of interest gradually added. In the baseline, Xit included a time trend and its interaction with high school degree, age interacted with high school degree, and a time trend interacted with the baseline level of each one of the included risk behaviors. We also assessed an alternative model in which Xit additionally included post-1996 indicator and its interaction with postconversion indicator. Attrition bias would arise if risk behaviors were interacting with seroconversion in systematic ways leading individuals to drop out of the study due to mortality, moving, or otherwise losing contact with the study.
We also ran pooled regressions of Eq. (1) – omitting individual fixed effects – on the full sample including always HIV-infected, always HIV negative, and seroconverters. These estimates identify average differences in behaviors between always HIV-uninfected individuals and seroconverters preconversion, and always HIV-infected and seroconverters postconversion. While controlling for time-invariant demographic factors (e.g. ethnicity, education, recruitment wave indicator), these results help clarify the importance of including individual fixed effects, when possible, to assess behavior change.
Statistical analyses were performed using STATA v15 (College Station, Texas, USA).
Characteristics of the study population and analytical sample
For the full MACS study population (n = 4616), Table 1 summarizes differences in baseline demographic characteristics, health, and behavioral outcomes of interest for participants who were always HIV-uninfected (n = 1176), always HIV-infected (n = 2882), or seroconverted during observation (n = 558). Compared with always uninfected or infected participants, seroconverters were younger, more educated, more likely to be non-Hispanic white, and generally physically and mentally healthier. Greater proportions of seroconverters reported sex with multiple partners, receptive anal sex, heavy drinking, heavy smoking, and monthly or more frequent use of poppers, whereas always HIV-infected participants had the highest proportion of monthly or more frequent marijuana use. Together, these initial differences in risk profiles across groups suggest the need for a fixed effects research design.
Results from regression analyses
Table 2 reports associations of HIV seroconversion with subsequent behaviors among seroconverters. The number of observations is less than the full sample and changes across columns because of missing values in the dependent variables, which differ across columns due to changes in reporting by individuals and survey items. Table 2 shows that HIV seroconversion was associated with reduced odds of subsequent engagement in sex with at least two partners [aOR: 0.371; confidence interval (CI): 0.263–0.523], reduced odds of insertive anal sex with at least two partners (aOR: 0.360; CI: 0.219–0.591), and reduced odds of heavy drinking (aOR: 0.704; CI: 0.508–0.977). These findings suggest reduced engagement in risky sex and heavy drinking, behaviors associated with increased HIV transmission , following seroconversion. Also of note, findings point to little or no change in other risk behaviors.
Table 3 reports the results of individual fixed effects regressions examining how the associations of HIV seroconversion with subsequent behaviors varied with exposure to HIV treatment using the two binary treatment measures (HAART availability and actual treatment initiation). If risk compensation were present, we would expect to observe increased risky sex behaviors (or offsets in any reductions in risky sex) following seroconversion in the post-1996 era and with actual treatment access. Instead, individual fixed effects regressions (panel A) showed that seroconversion after 1996 was associated with further reduced odds of insertive anal sex with at least two partners (aOR: 0.219; CI: 0.121–0.398), compared with before 1996 (aOR: 0.411; CI: 0.241–0.703). In addition to enhanced reductions in risky sex, we also observed similarly reduced odds of heavy smoking after 1996 (aOR: 0.388; CI: 0.160–0.939) compared with before (aOR: 1.447; CI: 0.793–2.641), and reduced odds of monthly or more frequent popper use following seroconversion in the post-1996 era (aOR: 0.301; CI: 0.128–0.704) compared with the pre-1996 period (aOR: 0.780; CI: 0.512–1.188). Because regressions using this post-1996 dummy variable may capture perceptions of treatment availability and other changes unrelated to HIV treatment, we also examined actual treatment initiation (panel B). These regressions showed that actual treatment initiation after seroconversion was associated with similarly reduced odds of insertive anal sex with at least two partners, heavy smoking, and monthly or more frequent popper use (results were qualitatively similar to those in panel A). Actual treatment initiation after seroconversion was also associated with further reduced odds of any sex with at least two partners (aOR: 0.239; CI: 0.157–0.364) in the post-1996 period compared with the pre-1996 era (aOR: 0.392; CI: 0.276–0.557), and heavy drinking (aOR: 0.421; CI: 0.264–0.673) in the post-1996 period compared with before 1996 (aOR: 0.759; CI: 0.551–1.047). Overall, findings from these regressions (panels A and B) reveal further reductions in risk behaviors following seroconversion in the presence of HIV treatment.
Additional analyses assessing the sensitivity of our results to the inclusion of time-varying physical and mental health factors suggested that CD4+ cell count and depression did not significantly affect the behavioral changes associated with seroconversion (Appendix Table A1, https://links.lww.com/QAD/B387), before or after treatment availability (Appendix Tables A2 and A3, https://links.lww.com/QAD/B387). Table A1 Panel A and Table A2, https://links.lww.com/QAD/B387 controlled for individuals’ CD4+ cell counts, which left the main results mostly unchanged. The same coefficient stability was observed in Table A1 Panel B and Table A3, https://links.lww.com/QAD/B387, which controlled for individuals’ CESD scores. Although statistical significance changed for several outcomes, the main conclusion of lower risk-taking after seroconversion, especially in the context of treatment availability, remained intact. Additional analyses showed that most risk behaviors did not predict postdiagnosis attrition (Table A4, https://links.lww.com/QAD/B387). Although there was some periodic attrition among those having receptive anal sex with multiple partners and those using poppers monthly or more frequently, these particular risk behaviors were associated with a lower likelihood of attrition after seroconversion. For other risk behaviors, the likelihood of attrition was similar before and after seroconversion. In addition, adjusting baseline estimates for these outcomes by the inverse probability of attrition left the main results in Table 3 unchanged. Overall, we conclude that our findings cannot be fully explained by CD4+ cell count or depression or differential attrition among higher risk individuals.
As seen in Table A5, https://links.lww.com/QAD/B387, in contrast to the fixed effects model, HIV infection was associated with increased odds of receptive anal sex with at least two partners (aOR: 1.725; CI: 1.448–2.056), and increased odds of monthly or more frequent marijuana use (aOR: 1.402; CI: 1.174–1.675). The discrepancies between these two models reflect pooled regressions’ inability to account for individuals’ inherent propensities for risk-taking. Moreover, comparing always HIV-uninfected with always HIV-infected individuals makes it difficult to distinguish the actual effects of seroconversion relative to other persistent, confounding differences between these groups over the study period.
We investigated behavioral responses to HIV seroconversion without and with access to treatment among MSM during the historical expansion of HAART in the United States. After accounting for unobservable time-invariant individual propensity for risk, we found reduced engagement in specific risk behaviors following seroconversion. These reductions were more pronounced following HAART availability (i.e., the post-1996 era) and actual treatment initiation, highlighting the behavioral benefits of improved access to treatment. In particular, HIV seroconversion was associated with reduced engagement in specific health risk behaviors including insertive anal sex with multiple partners and heavy drinking. Following HAART availability, seroconversion was associated with even further reduced odds of insertive anal sex and additional substance use behaviors (heavy smoking, frequent popper use) that have been associated with HIV risk and adverse health impacts in this population . As discussed below, our findings carry implications for health policy and practice as well as future research on health behaviors and biomedical advances in HIV testing and treatment.
Although our study cannot identify the precise mechanisms through which HIV seroconversion leads to reductions in risk behaviors in contexts of treatment availability and actual use, our results suggest that seroconversion is plausibly associated with an increase in the perceived benefits of reducing risky sexual behaviors and a re-evaluation of overall health and well being leading to reduced engagement in risky sex behaviors and heavy drinking. It is also possible that linkage to high-quality medical care and lifesaving treatment contributes to a ‘new lease on life’ or sense of hope . In contrast to risk compensation, this would trigger reductions in risky sex behaviors as well as other risk behaviors not directly transmitting HIV (e.g. substance use behaviors) to reduce mortality risks aside from HIV, in line with the notion of ‘competing risks’ . In addition, HIV testing and clinical care encounters often provide opportunities for risk reduction counseling and patient education, screening, and referrals for other health-promoting services (e.g. smoking cessation programs), suggesting a need for continued support for the HIV care continuum . Research on contemporary HAART-related attitudes  and experiences with HIV care in this population could help pin down mechanisms involved in these patterns of behavior change.
In addition to the use of rich longitudinal data, a key methodological distinction between our study and prior work is the use of a fixed effects framework that enabled comparing within rather than across diverse individuals. By focusing on a sample of seroconverters, we were able to identify within-person changes in behaviors before and after HIV seroconversion with credibility. One prior study similar to ours, focused on MSM in Amsterdam, found that reductions in sexual risk behaviors following HIV seroconversion were mitigated by the introduction of combination antiretroviral therapy . However, these analyses did not fully exploit the longitudinal structure of the data to control for time-invariant, individual-specific determinants of risk behaviors. These unobservable components of behavior could be confounded with HIV seroconversion and treatment exposure, making it difficult to assess the drivers of behavior change. Within the MACS study population, the baseline demographic, health, and behavioral differences we observed between the always HIV-infected, uninfected, and seroconverter groups confirmed the need for our fixed effects approach. As our initial analyses revealed, failing to use such an approach led to biased conclusions about behavioral responses (e.g. increased odds of receptive anal sex with multiple partners and frequent marijuana use) that would have implied very different policy implications than the more appropriate fixed effects approach used here. This methodological distinction, and more generally, appropriate methods will be essential in the broader line of inquiry regarding behavioral changes following access to treatment innovations.
Our study was limited by several factors. First, pertaining to external validity, even after enrichment efforts , MACS underrepresents racial and ethnic minority MSM who currently experience the highest incidence and burden of HIV in the United States . Future research on behavioral responses to HIV seroconversion should include racial/ethnic minorities, women, and populations in resource poor settings also experiencing high burdens of HIV. Second, our data precede the evidence on treatment as prevention and recent campaigns around ‘U = U’ (i.e. undetectable viral load equates to untransmissible); thus, research on the mechanisms we investigated here is needed in contemporary contexts. Third, data were subject to reporting errors and socially desirable responses. Biological markers of risk behaviors could be helpful in future studies. Fourth, we lacked data on potential mechanisms such as individuals’ knowledge of treatment and various treatments’ efficacy or other factors that could drive results (e.g. reduced social ties following diagnosis due to the historical and persistent stigma surrounding HIV/AIDS). Fifth, although our distinction between insertive and receptive anal sex speaks to behavior changes associated with specific types of risky sex, our analyses did not include other important types of risky sex due to limited availability of data spanning the full period of interest. Nevertheless, using the limited data on condom use, we found results consistent with our main findings (i.e. less condomless insertive anal sex following seroconversion, particularly after treatment availability; data not shown). Finally, although we controlled for a rich set of individual time-varying covariates, fixed effects approaches cannot fully control for individual-specific risk propensities that change over the life course (e.g. during individuals’ early 20s vs. late 30s). Because our approach identified differential changes in behavior between those who seroconverted before HAART vs. after, we cannot compare long-term with newly infected individuals. Future work should investigate whether the long-term infected behave similarly following HAART access.
Despite these limitations, our study provides evidence that positive behavior change is more likely to follow HIV seroconversion in the presence of treatment availability and actual use. Importantly, this implies that continued investment in the HIV care continuum, including support for engagement in clinical care and the provision of treatment, is clearly needed to boost the behavioral benefits of HIV testing. Although our study focused on US MSM, the design and findings may inform future research and policy debates in high-HIV-burden developing countries that are currently undergoing dramatic expansions of HIV testing and treatment, general health infrastructure, health insurance, and social safety nets. Understanding how expanded health and treatment infrastructures affect health behaviors in these settings is an important question for future research using longitudinal empirical strategies.
The work was supported by the National Institute on Drug Abuse grant K01DA043412 (PI: A.R.B.) and the Boston University Peter Paul Career Development Professorship. Data in this article were collected by the Multicenter AIDS Cohort Study (MACS). Studies were approved by the ethics review boards of participating institutions. MACS institutions, Principal Investigators, and funding include the following: Johns Hopkins University Bloomberg School of Public Health (Joseph Margolick), U01-AI35042; Northwestern University (Steven Wolinsky), U01-AI35039; University of California, Los Angeles (Roger Detels, Oto Martinez-Maza), U01-AI35040; University of Pittsburgh (Charles Rinaldo), U01-AI35041; the Center for Analysis and Management of MACS, Johns Hopkins University Bloomberg School of Public Health (Lisa Jacobson, Gypsyamber D'Souza), UM1-AI35043. The MACS is funded primarily by the National Institute of Allergy and Infectious Diseases (NIAID), with additional cofunding from the National Cancer Institute (NCI), the National Institute on Drug Abuse (NIDA), and the National Institute of Mental Health (NIMH). MACS data collection is also supported by UL1-TR001079 (JHU ICTR) from the National Center for Advancing Translational Sciences (NCATS) and NIH Roadmap for Medical Research. The contents of this publication are solely the responsibility of the authors and do not represent the official views of the National Institutes of Health (NIH) or MACS investigators. The MACS website is located at http://aidscohortstudy.org/.
W.Z. conceptualized the study design and performed data analysis. S.B. and A.R.B. contributed to study design. All authors contributed to the analysis and interpretation of data. All authors wrote and revised initial drafts of the article, and reviewed and approved the final article.
Conflicts of interest
There are no conflicts of interest.
1. Fonner VA, Denison J, Kennedy CE, O’Reilly K, Sweat M. Voluntary counseling and testing (VCT) for changing HIV-related risk behavior in developing countries
. Cochrane Database Syst Rev
2. Dombrowski JC, Harrington RD, Golden MR. Evidence for the long-term stability of HIV transmission-associated sexual behavior after HIV diagnosis
. Sex Transm Dis
3. Gong E. HIV testing and risky sexual behaviour
. Econ J
4. Heijman T, Geskus RB, Davidovich U, Coutinho RA, Prins M, Stolte IG. Less decrease in risk behaviour from pre-HIV to post-HIV seroconversion among MSM in the combination antiretroviral therapy era compared with the precombination antiretroviral therapy era
5. Eaton LA, Kalichman S. Risk compensation in HIV prevention: implications for vaccines, microbicides, and other biomedical HIV prevention technologies
. Curr HIV/AIDS Rep
6. Dukers NH, Goudsmit J, de Wit JB, Prins M, Weverling GJ, Coutinho RA. Sexual risk behaviour relates to the virological and immunological improvements during highly active antiretroviral therapy in HIV-1 infection
7. Lakdawalla D, Sood N, Goldman D. HIV breakthroughs and risky sexual behavior
. Q J Econ
8. Eaton L, Westercamp N, Abraham A. Eaton LA, Kalichman S. Risk compensation in response to HIV prevention
. Biomedical advances in HIV prevention
. New York, NY: Springer; 2014. 109–137.
9. UNAIDS. HIV testing and treatment. 2016. Available at: https://open.unaids.org/sites/default/files/documents/SRA1_HIV_test%26treat_2016.pdf
. [Accessed 12 September 2018].
10. National Institute of Mental Health. 2016. HIV/AIDS and mental health. Available at: https://www.nimh.nih.gov/health/topics/hiv-aids/index.shtml
. [Accessed 12 September 2018].
11. U.S. Department of Veterans Affairs. Drugs and alcohol: interactions with your HIV meds. Available at: https://www.hiv.va.gov/patient/daily/alcohol-drugs/interactions-with-meds.asp
. [Accessed 12 September 2018].
12. Kelly SG, Plankey M, Post WS, Li X, Stall R, Jacobson LP, et al. Associations between tobacco, alcohol, and drug use with coronary artery plaque among HIV-infected and uninfected men in the Multicenter AIDS Cohort Study
. PLoS One
13. Wada N, Jacobson LP, Cohen M, French A, Phair J, Munoz A. Cause-specific life expectancies after 35 years of age for human immunodeficiency syndrome-infected and human immunodeficiency syndrome-negative individuals followed simultaneously in long-term cohort studies, 1984–2008
. Am J Epidemiol
14. Wada N, Jacobson LP, Cohen M, French A, Phair J, Munoz A. Cause-specific mortality among HIV-infected individuals, by CD4(+) cell count at HAART initiation, compared with HIV-uninfected individuals
15. Dow WH, Philipson TJ, Xavier S-i-M. Longevity complementarities under competing risks
. Am Econ Rev
16. McMahon JH, Elliott JH, Bertagnolio S, Kubiak R, Jordan MR. Viral suppression after 12 months of antiretroviral therapy in low- and middle-income countries: a systematic review
. Bull World Health Organ
17. Kaslow RA, Ostrow DG, Detels R, Phair JP, Polk BF, Rinaldo CR Jr. The Multicenter AIDS Cohort Study: rationale, organization, and selected characteristics of the participants
. Am J Epidemiol
18. Chan TY, Hamilton BH, Papageorge NW. Health, risky behaviour and the value of medical innovation for infectious disease
. Rev Econ Stud
19. Riedl M, Geishecker I. Keep it simple: estimation strategies for ordered response models with fixed effects
. J Appl Stat
20. Bazzi AR, Rangel G, Martinez G, Ulibarri MD, Syvertsen JL, Bazzi SA, et al. Incidence and predictors of HIV and sexually transmitted infections among female sex workers and their intimate male partners in Northern Mexico: a longitudinal, multilevel study
. Am J Epidemiol
21. Radloff LS. The CES-D scale: a self-report depression scale for research in the general population
. Appl Psychol Meas
22. Sander PM, Cole SR, Stall RD, Jacobson LP, Eron JJ, Napravnik S, et al. Joint effects of alcohol consumption and high-risk sexual behavior on HIV seroconversion among men who have sex with men
23. Bradley H, Hall HI, Wolitski RJ, Van Handel MM, Stone AE, LaFlam M, et al. Vital signs: HIV diagnosis, care, and treatment among persons living with HIV – United States, 2011
. MMWR Morb Mortal Wkly Rep
24. Ostrow DE, Fox KJ, Chmiel JS, Silvestre A, Visscher BR, Vanable PA, et al. Attitudes towards highly active antiretroviral therapy are associated with sexual risk taking among HIV-infected and uninfected homosexual men
25. Silvestre AJ, Hylton JB, Johnson LM, Houston C, Witt M, Jacobson L, et al. Recruiting minority men who have sex with men for HIV research: results from a 4-city campaign
. Am J Public Health
26. CDC. HIV in the United States: at a glance. Available at: https://www.cdc.gov/hiv/statistics/overview/ataglance.html
. [Accessed 12 September 2018]