Highly Active Antiretroviral Therapy and Sexual Risk Behavior: Title and subTitle BreakA Meta-analytic Review FREE

Highly active antiretroviral therapy (HAART) became available in the United States and other developed countries in 1996. Although HAART regimens may not be effective for all infected persons due to drug-resistant strains of human immunodeficiency virus (HIV) and unmanageable adverse effects,^{1 - 3} many HIV-positive persons receiving HAART have substantially lowered their HIV RNA levels (ie, viral load) through strict adherence to treatment regimens.^{4 - 7} Consequently, the incidence of AIDS and deaths due to AIDS have decreased considerably in countries in which HAART has been widely available.^{8 - 10}

Apart from beneficial clinical effects, treatment advances may have unintended effects on sexual behavior. Some evidence suggests that since HAART became available, the prevalence of unprotected sex^{11 - 14} and the incidence of sexually transmitted infections (STIs),^{15 - 16} including HIV,^{17 - 20} have increased, mostly among men who have sex with men (MSM). Some HIV-positive persons receiving HAART, especially those with a low viral load, may feel protected from transmitting HIV sexually. Recent evidence does suggest that a low viral load may reduce the level of infectiousness of HIV-positive persons receiving HAART.^{21 - 23} As this information moves into the public domain, it may influence people's beliefs about HIV transmission and lessen concern about engaging in unsafe sex. People who hold these beliefs may be more likely than their counterparts to engage in unprotected sex. Given that taking HAART and having an undetectable viral load do not eliminate the possibility of transmitting HIV,^{24 - 26} it is important to examine whether use of HAART and beliefs about HAART and viral load are associated with sexual risk taking.

Several studies conducted since 1996 have examined these associations. The empirical findings have not been examined as a whole. Consequently, different and potentially conflicting conclusions may be reached when different subsets of studies are considered. Meta-analytic techniques quantitatively combine findings from multiple studies so that integrative conclusions can be reached.^{27 - 29} Accordingly, we conducted 3 meta-analyses to answer the following questions: (1) Among HIV-positive persons in care, is the group of persons being treated with HAART more likely than the group not being treated with HAART to engage in sexual intercourse without using a condom? (2) Among HIV-positive persons aware of their seropositive status, is the group with an undetectable viral load more likely than the group with a detectable viral load to engage in unprotected sexual intercourse? (3) Among HIV-positive and HIV-negative persons and persons of unknown serostatus, are HAART-related beliefs about HIV transmission and reduced concerns about unsafe sex associated with unprotected sexual behavior? Furthermore, we examined whether associations were moderated by participant characteristics (sex, sexual orientation, and HIV serostatus [for belief studies]), types of beliefs (transmission beliefs, reduced concern about unsafe sex), and other study characteristics (sample size, country of study).

We searched electronic databases including AIDSLINE, MEDLINE, PubMed, CINHAL, PsycInfo, ERIC, EMBASE, and Sociofile from January 1996 through August 2003 to identify relevant articles, book chapters, and conference abstracts. We crossed multiple search terms (ie, keywords and Medical Subject Headings terms) reflecting 2 categories: (1) HAART and viral load variables (HAART, highly active antiretroviral therapy, protease inhibitor, PI [protease inhibitor], antiretroviral therapy, combination therapy, HIV treatment, viral load, undetectable, CD4 counts, beliefs or attitudes about HIV treatment) and (2) sexual behavior (sexual risk behavior, risk behavior, unsafe sex, unprotected sex, condom use, sexually transmitted disease, STD, sexually transmitted infection, STI, HIV, AIDS). Additional articles were identified from the references of pertinent articles. We also conducted a hand search of 12 journals published between August 2002 and August 2003 and contacted authors for additional articles. English-language articles, conference proceedings, doctoral dissertations, and book chapters identified in the searches were screened independently for inclusion by pairs of reviewers (N.C., T.H., and G.M.).

Studies were included in the meta-analysis if they met all 3 of the following criteria:

Information from eligible articles was independently abstracted by pairs of the reviewers (N.C., T.H., and G.M.). Using a standardized spreadsheet, each study was coded for the following variables: authors, year of publication, data collection setting, study time period, HIV serostatus of participants, length of time since having tested as HIV-positive (if applicable), length of time taking HIV therapy, sex, ethnicity, focus of study (HAART, viral load, belief), participation rate, sample size, data collection methods (self-report vs medical record), sexual behavior measured, recall period, type of comparison (between-group, within-subjects), type of analysis (univariate, multivariate), and statistical tests used and findings reported. There was 98% agreement between reviewers across all variables coded. Discrepancies were reconciled by a third independent reviewer.

The following rules guided the data abstraction:

Effect sizes were estimated with odds ratios (ORs) because the majority of the studies compared 2 groups (eg, receiving HAART vs not) on a dichotomous sexual behavior variable. Three studies^{38 ,52 - 53} provided data on a continuous scale (eg, instances of a behavior, strength of belief). In such cases, we calculated a standardized mean difference (d) and then converted it into OR.⁵⁴ An OR greater than 1 indicates increased likelihood of unprotected sex among those receiving HAART, among those with an undetectable viral load, or among those believing that HAART or an undetectable viral load protects against transmitting infection or lessens concerns about engaging in unprotected sex.

Standard meta-analytical methods were used.^{28 - 29} A fixed-effects model was initially used to aggregate effect sizes across samples. We used the natural logarithm to obtain the log OR (lnOR) and calculated its corresponding weight (ie, inverse variance) for each independent sample. In calculating the overall effect size, we multiplied each lnOR by its weight, summed the weighted lnOR across samples, and then divided by the sum of the weights.

Next, in each meta-analysis, we calculated the magnitude of heterogeneity of the individual effect sizes by using the Q statistic. Q was significant in 2 of the 3 meta-analyses indicating significant random variation across effect sizes. Accordingly, the final aggregation used a random-effects model, which provides a more conservative estimate of variance and generates more accurate inferences about a population of studies beyond those included in the review.⁵⁵ In presenting the results, the aggregated lnOR was converted back to OR by exponential function and a 95% confidence interval (CI) was derived.

Another analytical issue pertains to the belief variables. Some studies^{41 ,47 ,53 ,56 - 59} had more than 1 belief item. We first calculated an effect size for each item and then aggregated those effect sizes within each sample included in the study⁵⁴ (some studies had more than 1 independent sample). Thus each sample contributed 1 belief finding.

Stratified analyses were conducted to examine whether effect sizes differed by 2 types of beliefs: (1) beliefs about whether HAART or viral load reduces HIV transmission and (2) beliefs reflecting reduced concern about unsafe sex given the availability of HAART. If a study had data on each type of belief, it contributed 2 findings for the stratified analyses.⁴⁵ Studies^{53 ,60} combining these 2 types of beliefs into an overall index in the primary analysis could not be included in the stratification. Additional stratified analyses examined whether effect sizes differed by participant characteristics; eg, MSM, women, heterosexual, injection drug user, country of study (United States vs other countries), definition of undetectable viral load, data collection method for viral load (self-report vs medical record), HIV serostatus of participants (for belief variables only), and sample size (based on a median split).

Two types of sensitivity analyses were conducted in each of the 3 meta-analyses. First, to examine the effect of outliers, we compared the overall effect size with estimates obtained after iterations using k−1 findings (k = number of independent samples). We removed a finding and calculated the overall ES. Then, we replaced that finding, removed another, and repeated the process. Second, we recalculated the overall effect size using data on different sexual behavior outcomes, partners, and statistical methods, if available, to see if our conclusions changed. Finally, publication bias was ascertained by inspection of a funnel plot⁶¹ and a linear regression test.⁶²

Twenty-five published studies met the inclusion criteria (Figure 1). Table 1, Table 2, and Table 3 together provide a descriptive summary of each study. All studies were conducted between 1996 and 2001 with the majority published after 1999. Sixteen (64%) of the studies were conducted in the United States. More than half of all studies were conducted with MSM^{38 ,41 - 42 ,46 - 47 ,51 - 53 ,56 - 57 ,59 - 60} or with a sample composed mostly of MSM.⁶⁶ Age of participants ranged from the late 20s to mid 40s. Only 5 studies^{39 ,53 ,56 ,59 ,69} reported participation rates (range, 66% to 100% with a median of 83%). The majority of the studies (n = 16) used a 6-month recall for sexual behavior outcomes.

Sixteen studies contributed 21 independent effect sizes reflecting the association between receiving HAART (vs not) and unprotected sex in HIV-positive persons. The percentage of HIV-positive persons receiving HAART or protease inhibitors during the study period ranged from 18% to 78% with a median of 59%. Only 1 article⁴⁹ reported the length of time that patients were receiving HAART, and only 4 articles^{38 - 39 ,58 ,70} provided information on the length of time since patients had tested positive for HIV (ranging from 1 to 199 months with a median of 48 months). The prevalence of unprotected sex ranged from 9% to 56% (median, 33%) in the HAART group and from 11% to 77% (median, 44%) in the non-HAART group.

The results of the random-effects models in Table 4 demonstrate that the weighted overall effect size reflecting the association between receiving HAART and engaging in unprotected sex was not significant (OR, 0.92; 95% CI, 0.65-1.31; k = 21). The likelihood of engaging in unprotected sexual behavior was not higher in the group of HIV-positive persons receiving HAART compared with the group not receiving HAART. Figure 2 shows the wide heterogeneity of results (Q₂₀ = 105.21, P<.001). However, sensitivity tests did not reveal any individual effect size that exerted influence on the overall effect size. The majority of the 21 effect sizes included 1.0 in their 95% CIs. Of the 7 significant effect sizes, 4 showed that receiving HAART was associated with a reduced likelihood of engaging in unprotected sex.

In stratified analyses shown in Table 4, the null association was seen in US and non-US studies and in studies with small and large sample sizes. Interestingly, in primary studies conducted with mixed samples (eg, MSM and heterosexuals pooled), heterosexual samples, or men and women combined, the group receiving HAART was significantly less likely than the group not receiving HAART to have engaged in unprotected sex. There was no evidence of publication bias.

Twelve studies contributed 13 independent effect sizes reflecting the association of viral load and unprotected sex in HIV-positive persons. In all of these studies, participants were classified by the original authors into 1 of 2 groups: undetectable or detectable viral load. Six studies provided information on detection thresholds including fewer than 50 copies/mL,⁶⁵ fewer than 200 copies/mL,⁴⁰ fewer than 400 copies/mL,^{39 ,46 ,57 ,68} or fewer than 500 copies/mL⁴⁰ depending on tests used. The other 6 studies grouped participants into undetectable and detectable viral load categories without providing detection thresholds. Two thirds of all studies abstracted viral load information from medical records.^{39 - 40 ,46 ,49 ,57 ,63 ,65 ,68} The remaining relied on self-reported information on viral load.^{34 ,42 ,52 ,70} The percentage of HIV-positive persons with undetectable viral load ranged from 22% to 59% with a median of 37% according to studies in which information was available.^{34 ,39 - 40 ,42 ,63 ,65 ,68 ,70} The prevalence of unprotected sex ranged from 10% to 68% (median, 39%) in the undetectable group and from 14% to 70% (median, 42%) in the detectable group.

The overall association (Table 4) between having an undetectable viral load and sexual risk behavior was nonsignificant (OR, 0.99; 95% CI, 0.82-1.21, k = 13). The group of HIV-positive persons with an undetectable viral load was not more likely than the group with a detectable viral load to engage in unprotected sex. The 13 effect sizes showed highly uniform results (Figure 3). A test of heterogeneity was nonsignificant (Q₁₂ = 14.27, P = .28). There were only 2 significant effect sizes: one showed that the group with an undetectable viral load was more likely to have engaged in unprotected sex and the other showed the opposite.

The magnitude of the overall effect size did not vary by participant risk group, country of study, whether viral load data were based on self-report or medical record, whether articles provided information on a detection threshold, or by size of study sample. Sensitivity tests did not reveal any changes in the findings, and there was no evidence of publication bias.

Ten studies contributed 18 independent effect sizes reflecting the association of unprotected sex with beliefs about HIV therapy or viral load in HIV-positive persons, HIV-negative persons, and persons of unknown serostatus. A consistent pattern of significant results was observed. The likelihood of unprotected sexual behavior was significantly higher in people who believed that HAART reduces HIV transmission or who were less concerned about engaging in unsafe sex given the availability of HAART (OR, 1.82; 95% CI, 1.52-2.17, k = 18). Twelve of the18 individual effect sizes were significant in this direction (Figure 4). The prevalence of unprotected sex ranged from 17% to 81% (median, 49%) in the reduced transmission or concern group and from 9% to 68% (median, 38%) in their counterparts.

The 18 effect sizes were significantly heterogeneous (Q₁₇ = 43.95, P<.001). In stratified analyses, the association was significant in US and non-US studies; in studies with small and larger sample sizes; in HIV-positive, HIV-negative, and combined serostatus samples; in MSM; and in both types of beliefs. The association fell well short of significance in heterosexuals due largely to the small number of samples (k = 3). The findings remained unchanged in sensitivity tests, and there was no evidence of publication bias.

This meta-analytic review of studies conducted since 1996 indicated that the prevalence of unprotected sexual intercourse was not significantly higher in the group of HIV-positive persons receiving HAART (vs not receiving HAART) or in the group of HIV-positive persons with undetectable (vs detectable) viral load. Unprotected sex was, however, associated with people's beliefs about HAART and viral load. Regardless of their HIV serostatus, the likelihood of unprotected sex was higher in people who agreed that receiving HAART or having an undetectable viral load protects against transmitting HIV or that the availability of HAART reduces their concerns about having unsafe sex.

The primary studies that examined the association between beliefs and behavior were correlational in nature. Thus, it is impossible to specify a causal relationship. It is reasonable to assume that beliefs are antecedents of behavior, but it is also possible that beliefs may originate or intensify as a response to one's behavior.^{71 - 72} That is, a person may come to strongly believe that HAART or undetectable viral load reduces HIV transmission after engaging in risky sex.^{51 ,73} In this case, the belief would serve as a psychological coping mechanism to reduce stress and anxiety. However, the belief may still serve to sustain risky behavior in the future.

Beliefs may affect behavior in a complex manner. People may engage in unprotected sex because they believe that HIV/AIDS is a less severe and threatening disease due to the availability of HAART. These beliefs as well as those about HAART-related HIV transmission may reduce people's concern about engaging in unsafe sex, which in turn may affect their behavior. Studies in this review did not examine disease severity beliefs in relation to other beliefs or behavior and thus the potentially complex relations of these beliefs cannot be specified herein. This issue merits attention in future research.

Medical factors such as length of time receiving HAART and stage of disease might elucidate our findings, but most articles did not report these data. For example, there may be a critical threshold of time during treatment before increases in unprotected sex are seen. Some patients who have just begun HAART may be experiencing symptomatic illness that may decrease interest in sex. Additionally, HAART may produce adverse effects that reduce sexual desire.⁷⁴ It is possible that patients receiving HAART who are feeling well may be more prone to engage in unprotected sex. However, it is encouraging that patients with undetectable viral load, who by definition have responded successfully to HAART, were no more likely than patients with detectable viral load to engage in unprotected sexual behavior.

Although the majority of the articles reviewed herein provided information on recruitment settings, most did not report on sampling frames, participation rates, and reasons for refusing participation. If these factors introduced a bias into a specific study, it is unlikely that all or even a majority of the investigations had the same type of bias. Thus, it is unlikely that these methodological variables biased the aggregated results of the meta-analyses. Importantly, we were able to show that effect sizes were not moderated by the sample sizes of the primary studies (ie, power to detect an effect).

Despite the limitations described above, our meta-analytic findings regarding the association of HAART-related beliefs with unprotected sex suggest that HIV and STI patients should receive prevention messages emphasizing that having an undetectable viral load does not eliminate the possibility of transmitting HIV^{24 - 26} nor does it mean that an infected person is "cured" of the virus. Even HIV-positive patients who are engaging in safer sex should hear these messages to reinforce their safer-sex practices. Similar messages are also needed for a broader audience because beliefs about HAART and viral load held by those who are HIV-negative and whose serostatus is unknown were significantly associated with their sexual risk behaviors. A full range of approaches may be needed to disseminate such messages, including mass media campaigns, posting information on Internet Web sites, providing messages to people calling HIV hotlines, and integrating messages into health care settings (eg, posters in clinic waiting rooms, brochures, and clinician counseling). A wide-scale approach to counteracting these beliefs may help prevent HIV infections in the age of HAART.