ONE OF THE CONCERNS raised regarding the introduction of any new human immunodeficiency virus (HIV)-prevention measure is the potential for risk disinhibition or sexual risk compensation.1–3 Concerns center around the possibility that the availability of a partially effective prevention technology will foster an overly optimistic sense of protection among users and lead to increases in other forms of risk behavior. This issue has been raised regarding various HIV prevention efforts, including vaccines,4 male circumcision,5 and vaginal microbicides.6
Concerns over risk disinhibition are partially grounded in past HIV and sexually transmitted infection-prevention studies. Wong et al.7 documented a decrease in condom use among Cameroonian women participating in a clinical trial on the effectiveness of nonoxynol-9 and results from hepatitis-B vaccine efficacy trials in the US indicated that placebo recipients were more likely to become infected with hepatitis B after their final injection.8 These studies point to the potential for risk disinhibition. However, risk reduction counseling in HIV-prevention trials has also been shown to be effective and have a positive effect on reducing sexual risk. Bartholow and colleagues showed a general decline in HIV risk behavior among gay male participants in an HIV vaccine efficacy trial,9 a finding corroborated qualitatively by Guest et al.10 These findings are, at least partially, consistent with other clinical studies in which HIV/STI risk behavior declined significantly from baseline.11
The oral tenofovir, PrEP, HIV-prevention trial, in particular, has stimulated much international discussion.12–16 Media from various countries reported concerns that the trials lacked adequate HIV prevention counseling for participants and that sexual risk behavior would increase as a result of trial participation.17,18 Some critics argued that the tenofovir disoproxil fumarate (TDF) trial participants—who were primarily sex workers—would feel protected and use condoms less often, or have sex with more clients.19–21 This article directly addresses these concerns and presents sexual behavior data from the Tenofovir HIV prevention trial in Ghana.
Materials and Methods
The oral tenofovir trial was a randomized, double-blind, placebo-controlled study conducted between June 2004 and March 2006 in Tema, Ghana; Douala, Cameroon; and Ibadan, Nigeria.22 The data presented here are from the Ghana site only, due to premature study closures in the other sites.23 The Ghanaian portion of the study was approved by the Ghana Health Service Ethical Review Committee, Accra, Ghana and the Protection of Human Subjects Committee, Family Health International, USA.
The trials were designed to test the safety and efficacy for prevention of HIV acquisition of a 300 mg daily oral dose of TDF. At the Ghana site 400 women were enrolled into the trial over a 6 month period and data from 361 participants were analyzed. Eligibility criteria for the study included being HIV-antibody negative, 18- to 35-years old, not being pregnant or breastfeeding, not desiring pregnancy during 12 months of study participation, and having an average of 3 or more coital acts per week and 4 or more sexual partners per month. All participants completed written informed consent at enrollment.
Women were recruited from the port city of Tema using outreach workers. They were tested for HIV at their screening visit, their enrollment visit, and each subsequent monthly visit. They also received pre–post-test and HIV prevention counseling at all of these visits.
Eligible women were randomized into 1 of 2 arms—placebo pill or TDF pill—and were counseled to take 1 pill every day. All participants received HIV risk-reduction counseling, male condoms, and treatment for STIs at monthly clinic visits. Women were counseled at each visit to use male condoms to prevent pregnancy, HIV, and STIs. At each monthly follow-up visit, participants also received free male condoms.
All participants were administered a structured questionnaire at enrollment and at each of the monthly visits. The primary 2 risk outcomes from this questionnaire that were used for this analysis were the number of different male sexual partners they had in the past 30 days and the proportion of unprotected coital acts in the past 7 days. The proportion of unprotected coital acts variable was calculated by dividing the participants' numeric responses for 1 item (In the last 7 days, how many times did you have vaginal sex?) by a second item (In the last 7 days, how many times did you use a condom for vaginal sex?). The value derived from this procedure was subtracted from 1.0. A subsample of women were also asked, in separate in-depth interviews, open-ended questions (after their 6-month visit) about changes in sexual behavior.
Of the 361 participants, 151 (42%) completed all twelve follow up visits. To account for potential biases from loss to follow-up we conducted an attrition analysis, comparing baseline risk indicators between women who completed varying numbers of follow up visits. We compared women who discontinued the trial before their fourth visit (n = 48) with those who completed 4 or more visits (n = 313). We also compared participants who discontinued before their seventh visit (n = 82), and those who completed 7 or more visits (n = 279). Finally, we compared women who completed all 12 visits (n = 151) with those who completed less than 12 visits (n = 210). Using Chi-square and t tests, we compared these groups on 4 risk indicators reported at the screening visit: history of an sexually transmitted disease, condom use at last sex, number of different sexual partners in the past 30 days, and the condom use ratio in the past 30 days.
To measure participants' sexual behavior over time, we first conducted an exploratory analysis using pairwise tests to examine whether significant differences existed between any 2 time points. We then performed growth curve analyses (GCA)24 by a 2-step analysis procedure to estimate the trend of each outcome variable (number of partners and proportion of unprotected sex acts) over time adjusted with participants' characteristics.
GCA models were first used to examine each of the baseline covariates and time data (visit) to test whether demographic characteristics influenced change for either outcome. Demographic variables included age, education, previous STI diagnosis, previous pregnancy, presently living with partner, condom use during last coital act, previous contraceptive use, and previous practice of anal sex and previous practice of oral sex. Any variable that had a P-value of less than 0.1 was included as a covariate to control for that effect in the later multivariable linear and curvilinear GCA.
A series of random effects models were evaluated to assess changes in each risk behavior over time, at both individual and aggregate levels. Each model contained time and 1 of the covariates identified as significant in the bivariate analysis. The multivariable analysis also included all 2-fold interactions with time. A final model containing all main effects that entered into the multivariable model plus any significant interaction terms was fit to the data. Owing to the high degree of variability between the respondents' reported number of sexual partners and unprotected acts, we further categorized study participants into 2 groups: an extreme quartile, exemplifying the riskiest and most variable behavior (across visits), and the remaining 75% for each outcome variable. A sensitivity analysis was then carried out to examine whether either group was driving the observed change.
In addition to reporting P-values from the final multivariable model for each outcome, pairwise tests of all time-points were evaluated as well, to determine whether significant differences existed between any 2 time points. All testing was 2-sided and all P-values <0.05 were considered significant. Statistical analyses were carried out in SAS v9.1, using Proc GLIMMIX.
For the qualitative component, a sample of 24 women was systematically selected from the entire clinical trial population. For each of the 6 enrollment months, 6 women were randomly selected from the pool of clinical trial participants for each month. For each of these groups, outreach workers recruited 4 of the 6 possible participants based on their availability. A total of 26 women were approached and 2 refused. Qualitative interviewers were not associated with the clinical trial.
A semistructured open-ended instrument was used to interview participants. All interviews were recorded on audio tapes, translated into English, and transcribed verbatim by field staff in accordance with a transcription protocol.25 Thematic codebook development followed a standardized process.26 As per this iterative process, themes were identified from the transcript content by 2 researchers who reached agreement on the definition of each theme. Code (theme) frequencies were generated in AnSWR.27 For this analysis, theme frequencies were generated only for themes that emerged in response to 2 open-ended questions and subsequent probes:
- Has participating in the TDF study affected your condom use? If so, please explain. How do you feel about these changes? What do you think are the reasons for these changes?
- Has participating in the TDF study affected the number of sexual partners you have? How do you feel about these changes? What do you think are the reasons for these changes?
The demographic characteristics of the clinical trial participants at enrollment are presented below in Table 1. The mean age and level of education for participants was 23.7 and 7.1 year, respectively. For our analysis, age and education were dichotomized based on their medians.
A small percentage of women (9.7%) reported living with a partner at enrollment. Less than 45% of the women reported previous use of contraceptives (including condoms), or using condoms during their last sexual act. More than 70% of participants had been pregnant previously and about 44% reported ever having been diagnosed with an STI. Close to half of the sample (41%) reported ever performing oral sex on humans. Far fewer (17.7%) reported ever having practiced anal sex.
Comparisons were done between groups of women who discontinued the clinical trial at different points in time. The only significant difference between the groups outlined above was found in the condom use ratio. Women who completed all 12 visits had significantly higher rates of condom use at screening. All other comparisons generated nonsignificant findings, suggesting that no systematic bias resulted from loss-to-follow-up during the clinical trial.
Number of Sexual Partners
The overall trend of change in number of partners over time (Fig. 1) was a curvilinear (quadratic) trend. Controlling for age, previous contraceptive use, and previous pregnancy (variables found to be significant bivariate predictors) we tested nonsignificant general linear (P = 0.14) and curvilinear (P = 0.94) changes over time.
Chi-squared and t tests indicated that women whose mean scores and individual standard deviations for the overall number of partners placed them in the riskiest quartile (n = 78) reported engaging in less anal sex than the remaining 75% (P <0.05; Table 2). Trend lines were plotted and reviewed for the 2 subgroups and reflected dissimilar trends (Fig. 1). The extreme group had substantially higher overall averages than the remaining 75%. Testing this relationship using GCA, we found a statistically significant difference in the change of these 2 groups. The extreme group decreased their number of partners when compared with the remaining 75% (P <0.01). Separate adjusted linear and curvilinear GCA models were then tested for each subgroup. In these 2 separate subanalyses neither the trend for the extreme quartile nor for the remaining 75% revealed a significant change in the number of men per month. These results suggest that the statistical significance found in the overall group was driven by the juxtaposition of these 2 groups.
Despite slight increases in the reported rate of unprotected sex acts at visits 2, 3, and 5, the general trend across all visits revealed that participants reported decreases in unprotected sex as the study continued (Fig. 2). Unadjusted pairwise analysis showed no significant differences between any 2 consecutive visit points. However, in the pairwise comparisons from baseline to follow-up, visits 4 and 6 to 12 all showed significantly lower rates of unprotected sex [visit 4 (P = 0.04), 6 (P = 0.04), 7 (P <0.01), 8 (P <0.01), 9 (P = 0.02), 10 (P <0.01), 11 (P <0.01), and 12 (P <0.01)]. Reported increases in unprotected sex at visits 2, 3, and 5 were not statistically different from baseline. Controlling for condom use at last sex and previous pregnancy, GCA revealed a nonsignificant linear change (P = 0.16) and a nonsignificant quadratic trend over time (P = 0.98).
In our subgroup analysis for this variable, an unadjusted GCA curvilinear comparison between the riskiest quartile (n = 78) and the remaining 75% (n = 283) revealed higher rates of reported condom use by the riskier quartile across the duration of the study. Women in the riskier quartile were also less likely to have been pregnant before enrolling (P <0.05), and more likely to report having used a condom during their last sexual encounter at screening (P <0.01).
For women in the riskier quartile the rate of unprotected acts decreased from 22.2% to 10.5%, a highly significant curvilinear change (P < 0.001). In the remaining 75%, the mean change in unprotected sex decreased less dramatically from the baseline rate of 3.7% to the final rate of 1.5%, a significant linear change over time (P <0.001). When tested, we found that the rates of change were significantly different for the 2 groups of women (P <0.001). The extreme group exhibited a greater decrease in unprotected acts. This could reflect a regression towards the mean, but given that all groups showed similar overall trends, we cannot assume that convergence over a longer period of time would occur.
Table 2 presents the frequencies for the most common content-driven themes related to changes in sexual behavior. From this table and the larger thematic analysis that was conducted, we developed a provisional model of the effect of counseling on HIV risk behavior. This model is presented in the narrative below. Verbatim quotes from in-depth interview participants are provided in Table 3 for each of the postulated causal pathways described in the narrative. Note that the model represents the maximum variability of potential pathways within the qualitative data (i.e., in some cases only 1 participant expressed a given pathway). The number of participants exemplifying an individual pathway varied from 1 to 18 (n = 24).
The counseling provided during the trial was perceived by the majority of participants to affect their risk behavior in a positive (protective) direction (Table 2). For some women, getting tested for HIV in the trial and finding out they were HIV-negative (a requirement of the study) gave them added motivation to use condoms (path #1, Table 3). In a similar fashion, counseling made some participants more health conscious in general, increasing their knowledge and awareness of HIV/AIDS (#2). Related to the above, the counseling also increased participants' risk perception of contracting HIV (#3).
The trial counseling was perceived by the majority of participants to increase their condom use (e.g., #2, #4). More specifically, the counseling gave women greater access (free and unlimited supply) to condoms (#5) that participants, and in some cases their clients, perceived to be of higher quality than other condoms (#6). Women also learned how to better negotiate condom use with clients and how to use condoms properly, thereby improving their condom self-efficacy (#7).
The most common theme expressed in the text analyzed was increased condom use (Table 2). In sum, women reported that knowing their HIV-negative status, combined with increased knowledge and risk perception regarding HIV, made them more likely to use condoms with clients and (in a few cases) regular partners. Better condom efficacy (i.e., better knowledge and skills for effectively using condoms) and greater access to condoms also increased condom use.
Counseling also had an effect on the number of sexual partners women had. Some participants suggested that the education from the counseling and enhanced HIV risk perception led to a decrease in the number of partners (#8, #9, respectively). Even the simple knowledge of an HIV-negative status could indirectly result in a reduction in the number of sexual partners (#10). Trial participation, however, could also facilitate an increase in number of sexual partners. In a few cases, women reported that having a confirmed HIV-negative status (and being associated with the trial in general) made them more attractive to their male clients who perceived them to be disease-free (#11). Some women capitalized on this situation and reported an increase in their number of sexual partners.
Number of sexual partners was also influenced in both directions by the reported increase in condom use. In some instances, women reported that clients perceived a woman who insists on condom use as safer, and therefore, more desirable (#12). In other situations, women took on more clients because they themselves felt a sense of protection afforded by the condom (#13). Some women took advantage of this perception of enhanced desirability and accepted more partners. Conversely, some clients simply did not want to use condoms, resulting in an overall reduction in number of partners for some women (#14).
One of the main conclusions that can be drawn from our study is that, overall, being in an HIV prevention study does not necessarily lead to risk disinhibition. In fact, in the TDF study risk behavior, on average, decreased over the course of the trial. These findings are congruent with other studies showing similar reduction of risk in clinical trials.8–11 Results further confirm that consistent counseling and associated condom provision can be effective risk reduction strategies and, in this trial, were likely the main reason behind the largely positive behavioral changes observed. Participants clearly ascribed their changes in behavior to various aspects of the counseling provided throughout the trial.
These findings have implications for required sample sizes for future HIV prevention trials where seroconversion is the main outcome. Indeed, the HIV incidence for the TDF study in Ghana was much lower than predicted (based on existing epidemiologic data), and, combined with the closure of the other 2 study sites, rendered the efficacy data statistically inconclusive.22 This raises questions about how and where future effectiveness trials should be conducted for PrEP, vaccines, and vaginal microbicides. The notion that HIV prevention trials can be conducted more efficiently in high incidence populations needs to be reconsidered in a more nuanced way. For example, if high incidence is reflective of individual-level factors such as condom skills and condom access then participation in a prevention trial is likely to have a significant impact on incidence. Alternatively, if high incidence is reflective of structural barriers to risk reduction, such as gender power dynamics and economic vulnerability, then trial participation will probably have less of a direct effect.
Results from the growth curve analysis showed significant differences between women in the riskiest quartile and the remaining 75% of the trial participants in terms of changes in risk behavior across the trial. Our subanalysis also showed that although the “risky” group of women had more sexual partners they were less likely to engage in other types of risk behavior such as anal sex. They also tended to use condoms more often, suggesting they were engaging in risk management based on the context of the sexual activity. These findings, and the fact that not all risk behaviors cluster together, warrant further investigation, as they have implications for how counseling or other interventions are developed. The data suggest that a one-size-fits-all model of prevention messaging and delivery may not be the most effective for this population.
The other implication of the findings presented is that counseling may have a differential effect on groups with unique risk profiles. More research is needed to better understand these differences and to describe the intricate relationship between counseling, risk perception, individual characteristics, and risk behavior, both within and outside of the clinical trial context.
1. Cassell M, Halperin D, Shelton J, et al. Risk compensation: the Achilles' heel of innovations in HIV prevention?. BMJ 2006; 332:605–607.
2. Pinkerton S. Sexual risk compensation and HIV/STD transmission: empirical evidence and theoretical considerations. Risk Anal 2001; 21:727–736.
3. Blower S, McLean A. Prophylactic vaccines, risk behavior change, and the probability of eradicating HIV in San Francisco. Science 1994; 265:1451–1454.
4. Chesney M, Chambers D, Kahn J. Risk behavior for HIV infection in participants in preventive HIV vaccine trials: a cautionary note. J Acquir Immune Defic Syndr 1997; 16:266–271.
5. Lagarde E, Dirk T, Puren A, et al. Acceptability of male circumcision as a tool for preventing HIV infection in a highly infected community in South Africa. AIDS 2003; 17:89–95.
6. Foss AM, Vickerman PT, Heise L, et al. Shifts in condom use following microbicide introduction: should we be concerned?. AIDS 2003; 17:1227–1237.
7. Wong E, Roddy R, Tucker H, et al. Use of male condoms during and after randomized, controlled trial participation in Cameroon. Sex Transm Dis 2005; 32:300–307.
8. Farley TA, Cohen DA, Kahn RH, et al. The acceptability and behavioral effects of antibiotic prophylaxis for syphilis prevention. Sex Transm Dis 2003; 30:844–849.
9. Bartholow BN, Buchbinder S, Celum C, et al. HIV sexual risk behavior over 36 months of follow-up in the world's first HIV vaccine efficacy trial. J Acquir Immune Defic Syndr 2005; 39:90–101.
10. Guest G, McLellan-Lemal E, Matia DM, et al. HIV vaccine efficacy trial participation: men who have sex with men's experiences of risk reduction counselling and perceptions of risk behaviour change. AIDS Care 2005; 17:46–57.
11. Kaul R, Kimani J, Nagelkerke N, et al. Reduced HIV risk-taking and low HIV incidence after enrollment and risk-reduction counseling in a sexually transmitted disease prevention trial in Nairobi, Kenya. J Acquir Immune Defic Syndr 2002; 30:69–72.
12. Ahmad K. Trial of antiretroviral for HIV prevention on hold. Lancet Infect Dis 2004; 4:597.
13. Bonn D. Tenofovir: a pill to prevent HIV?. Lancet Infect Dis 2004; 5:78.
14. Loff B, Jenkins C, Ditmore M, et al. Unethical clinical trials in Thailand: a community response. Lancet 2005; 365:1617–1619.
15. Richards T. Conduct of drug trials in poor countries must improve. BMJ 2005; 330:1466.
16. Dechambenoit G. Ethics in the north and in the south: the African elites should not be silent. Afr J Neurol Sci 2005; 24.
17. Elias P. Activists demand drug maker Gilead halt AIDS trial. San Jose Mercury News
July 4, 1914.
18. Jack A. AIDS drug pioneers in talks with backers. Financial Times Com
August 3, 2005.
19. Abidjan NC. Cameroon activists want protection of drug test victims. VOA Com
February 15, 2005.
20. Pyne S. Hun Sen halts controversial HIV drug trial. Cambodia Daily
August 13, 2004.
21. Johnson S. Protests may slow test of HIV drug: doctors see promise in Foster City firm' product. Mercury News
June 13, 2005.
22. Peterson L, Taylor D, Roddy R, et al. Tenofovir disoproxil fumarate for prevention of HIV infection in women: a phase 2, double-blind, randomized, placebo-controlled trial. PLoS Clin Trials 2007; 2:e27.
23. Grant R, Buchbinder S, Cates W, et al. AIDS: promote HIV chemoprophylaxis research, don't prevent it. Science 2005; 309:2170–2171.
24. Singer J. Using SAS PROC MIXED to fit multilevel models, hierarchical models, and individual growth models. J Educ Behav Stat 1998; 23:323–355.
25. McLellan E, MacQueen KM, Niedig J. Beyond the qualitative interview: data preparation and transcription. Field Methods 2003; 15:63–84.
26. MacQueen KM, McLellan E, Kay K, et al. Codebook development for team-based qualitative analysis. Cultur Anthropol Methods J 1998; 10:31–36.
27. Centers for Disease Control and Prevention. AnSWR: Analysis Software for Word-based Records, Version 6.4. Altanta, GA: Centers for Disease Control and Prevention, 2004.