Circumcision Status and Risk of HIV and Sexually Transmitted Infections Among Men Who Have Sex With Men:  A Meta-analysis FREE

Randomized controlled trials (RCTs) conducted with men in Africa have shown that male circumcision reduces the likelihood of female-to-male transmission of human immunodeficiency virus (HIV) infection by 50% to 60%.^1- 3 Observational studies also suggest that male circumcision may protect heterosexual men against acquisition of other sexually transmitted infections (STI), such as syphilis, chlamydial infection, or genital ulcer disease.^4- 5 The protective effect of circumcision among heterosexual men has generated discussion about the potential role of circumcision in reducing the transmission of HIV and other STIs among men who have sex with men (MSM).^6- 7

Several factors may influence the protective effect of circumcision among MSM. Male circumcision may be most effective as an HIV prevention strategy in countries where HIV/STI prevalence is high and circumcision prevalence is low.⁸ Apart from HIV/STI and male circumcision prevalence, sexual position also plays an important role in the degree to which circumcision protects against disease acquisition among MSM. Circumcision among MSM may protect against HIV infection only among those who primarily or exclusively take the insertive role during unprotected anal intercourse⁷ because unprotected receptive anal intercourse—the riskiest sexual behavior for contracting HIV infection⁹—is independent of any protective effect afforded by circumcision.

An expert panel convened by the US Centers for Disease Control and Prevention (CDC) for a consultation on male circumcision and HIV infection recommended that evidence from existing observational studies be systematically reviewed before determining the usefulness of an RCT of male circumcision among MSM.¹⁰ Only 1 quantitative review has examined circumcision data from studies of MSM.¹¹ However, that review included only 2 studies of male circumcision and HIV infection and did not consider unpublished studies or studies of other STIs.¹¹ Given the increasing interest in circumcision as an HIV prevention strategy, a more thorough synthesis of available data on circumcision and HIV/STI risk among MSM is critical.

We systematically searched the scientific literature and performed a meta-analysis to examine the strength of the association of circumcision status with HIV infection and other STIs among MSM.

We searched widely used databases (ie, MEDLINE, ERIC, Sociofile, PsycInfo, EMBASE, Web of Science, Google Scholar) for relevant reports from the beginning of indexing for each database through February 2008. Searching key words and Medical Subject Headings (US National Library of Medicine) relevant to circumcision (ie, circumcision, circumcised, and uncircumcised), we cross-referenced the male circumcision citations and citations pertinent to homosexual men (ie, homosexual, bisexual, men who have sex with men, MSM, and gay). In addition, we searched the Web sites of HIV/STI–related conferences (ie, International Society for Sexually Transmitted Diseases Research, Conference on Retroviruses and Opportunistic Infections, International AIDS Conference, National HIV Prevention Conference) for relevant abstracts dating back to 1989. Last, we checked the reference lists of pertinent articles for additional citations and contacted investigators of published reports and conference abstracts to identify other possible sources.

Studies that met each of the following criteria were included in the review: (1) recruited MSM as part of the study; (2) included male circumcision as a study variable; and (3) reported quantitative data (either bivariate percentages or results of statistical tests) reflecting the association of circumcision status and HIV or STI prevalence among MSM. When necessary, study authors were contacted to obtain additional information or additional data to calculate effect sizes.

Pairs of reviewers independently abstracted data from eligible articles. The study team used standardized abstraction sheets for recording study authors, publication year, enrollment period, study location, overall sample size of MSM, sample size of circumcised and uncircumcised MSM, racial/ethnic composition of the sample, study design (cross-sectional vs prospective), determination of circumcision status (self-report vs genital examination), data source (published vs unpublished), and type of analysis (univariate and/or multivariate). Although infrequent, disagreements between reviewers during the abstraction process were resolved by discussion.

Abstracted data were entered into a spreadsheet by one investigator and reviewed by a separate investigator. Several decision rules regarding participant characteristics and thoroughness of the data guided the preparation of data for analyses (Box).

Odds ratios (ORs) were used to estimate effect sizes. Abstracted data for circumcised MSM and uncircumcised MSM were converted, when necessary, into percentages that represented yes vs no for a given outcome. Standard meta-analytic methods were used to aggregate effects across studies.¹² To estimate the overall effect size, each natural log OR (lnOR) was weighted by the inverse of its variance, the weighted lnOR summed across samples, and then divided by the sum of the weights.¹³ For the purpose of presentation and ease of interpretation, we converted effect sizes and 95% confidence intervals (CIs) back to ORs. An OR of less than 1 indicates a decreased odds of HIV infection or STI among circumcised compared with uncircumcised MSM. We used the I² index to examine heterogeneity of individual effect sizes in the overall aggregations for HIV and STI and in stratified analyses.¹⁴ According to Higgins et al, I² values of near or less than 25% indicate low heterogeneity, values near 50% indicate moderate heterogeneity, and values near 75% or higher indicate high heterogeneity.¹⁴ Under conditions of low heterogeneity (I² ≤ 25%), we used fixed-effects models and under conditions of higher heterogeneity (I² = 25%) we used random-effects models.^15- 16 Using methods developed by Hedges and Pigott¹⁷ to calculate the statistical power of our meta-analysis, we specified a small effect size (15% relative reduction in odds of HIV infection among circumcised vs uncircumcised MSM) as an estimate of the true population value for the overall and stratified aggregations. Power was calculated using variance components derived from the primary studies, a 2-tailed test, and α of .05.

Per recommendations from the Cochrane Collaborative Review Group on HIV Infection and AIDS,¹⁸ the quality of observational studies in this meta-analysis was assessed using the Newcastle-Ottawa Scale.¹⁹ This instrument assesses the quality of nonrandomized studies in 3 broad categories (patient selection [4 criteria], comparability of study groups [1 criterion], and assessment of the outcome [3 criteria]). Following quality assessment standards of previous meta-analyses,²⁰ studies in our meta-analysis that met 5 or more of the Newcastle-Ottawa Scale criteria were considered to be of higher quality. The aggregated effect size of the higher-quality studies were compared with the aggregated effect size of all other studies. We also examined the association of the Newcastle-Ottawa Scale (sum score across the 3 categories) for each study with the respective effect sizes using a weighted generalized least squares meta-regression model.²¹ Study quality was examined further by focusing on a subset of studies with moderate (22%-66%) circumcision prevalence because studies with very high or very low male circumcision rates may not have had enough variability to detect an association with HIV and STI. Last, we assessed study quality by limiting analyses to studies with large analytical samples (> 1000 participants), and studies that determined circumcision by genital examination and also used diagnostic tests to assess HIV status.

The effect of potential outliers was examined by comparing the aggregated effect size with estimates obtained after iterations using k−1 findings (k = the number of independent effect sizes). Sensitivity analyses of the HIV results indicated that there was no evidence of an outlier among the studies. Sensitivity analyses of the STI results indicated that 2 studies^22- 23 may be outliers, but we did not treat either study as a statistical outlier because each of the k−1 estimates produced a 95% CI that overlapped with the 95% CI of the full STI sample. To evaluate the presence of publication bias, we used linear regression methods proposed by Sterne and Egger²⁴ to investigate funnel plot asymmetry. There was no evidence of publication bias among the set of HIV studies in these analyses (β, 0.056; P = .84), but we found evidence of potential publication bias among the set of studies that examined STI (β, 0.824; P = .01).

We identified 4337 citations of which 33 were considered relevant and obtained for further screening (Figure 1). Of these 33 studies, 18 quantitatively examined the association between male circumcision status and HIV or STI among MSM (number of studies and count of reference citations may not sum because Templeton et al has separate citations for HIV³⁴ and STI³⁵).^{22- 23,25- 41} Seventeen of the 18 studies examined the association between circumcision and HIV,^{22- 23,25- 31,33- 34,36- 41} and 7 of the 18 studies examined the association between circumcision and STI.^{22- 23,27,29,32- 33,35}

Of the 17 studies that examined the association between circumcision and HIV infection among MSM, 9 of the original reports^{25- 28,30- 31,33- 34,36} found no statistically significant association with HIV, 5 reported^23,37- 40 that circumcision had a significant protective association with HIV, 1 reported²² a nearly significant protective association, 1 reported⁴¹ that circumcised MSM had a significantly greater odds of HIV infection, and 1 reported no statistically significant association in the overall sample but found a significantly protective effect among men who only engaged in insertive anal sex.²⁹

Of the 7 studies that examined circumcision and STI among MSM, all except 1 reported data for both HIV and STI.³² Among the 7 STI studies, 4 of the original reports^22,27,29,33 found no statistically significant association between male circumcision and STIs, 2 reported a significantly protective effect for syphilis and no statistically significant association for other STIs,^23,35 and 1 reported no association for most STIs but a significantly greater odds of nonchlamydial nongonococcal urethritis.³²

After excluding 3 of the 18 total studies due to insufficient data,^28,30,38 a final set of 15 studies^{22- 23,25- 27,29,31- 37,39- 41} were included in the meta-analysis (Table 1). The 15 studies were conducted between 1989 and 2007. Nine studies^{22- 23,25- 27,32- 33,36- 37} took place in North America, 9 were conducted^{22- 23,25,27,32,34- 37,41} with primarily white participants, 4 had a prospective cohort design,^26,34- 37 and 9 were unpublished abstracts or reports or previously unreported circumcision data.^{25- 27,29,31,34- 36,39,41} The average quality of the studies, based upon the Newcastle-Ottawa Scale, was moderate (Table 1). Five studies that met 5 or more study quality criteria^{29,34- 37,40} were considered the highest quality; the remaining studies fulfilled fewer than 5 of the scale's 8 study criteria. Across all studies, the prevalence of male circumcision ranged from 4% to 88% (median = 60%). A total of 53 567 MSM were included in our analytical sample, 52% of whom were circumcised.

Fourteen studies^{22- 23,25- 27,29,31,33- 34,36- 37,39- 41} contributed 15 findings for the association of circumcision and HIV infection (Figure 2). Our analysis included a total of 27 816 circumcised MSM and 25 751 uncircumcised MSM. The weighted overall effect size reflecting the association between circumcision and HIV infection among MSM was protective, but statistically nonsignificant (OR, 0.86; 95% CI, 0.65-1.13; k = 15). The power for the overall meta-analysis of HIV outcome studies was 0.78. There was moderate to high heterogeneity among the 15 findings (I² = 64%), which warranted further examination via stratified analyses. Most of the stratified analyses revealed associations that were moderately protective, but not statistically significant (Table 2). Circumcision was not significantly associated with HIV infection when stratified by prevalence of HIV among study samples, prevalence of male circumcision among study samples, method used to determine either circumcision or HIV status, cross-sectional or prospective studies, bivariate or multivariate analysis, published or unpublished data, country in which the study was conducted (United States vs non United States), World Bank country classification (developed vs developing),⁴² or racial/ethnic composition of the study sample. However, there was moderately high heterogeneity among studies in each of these stratified analyses (I² range, 33%-80%).

In contrast, a statistically significant protective association (OR, 0.47; 95% CI, 0.32-0.69; k = 3; I² = 0%) of circumcision with HIV infection was found for MSM studies conducted prior to the introduction of highly active antiretroviral therapy (HAART) in 1996. Of studies conducted after HAART, the association of circumcision and HIV infection was not statistically significant and heterogeneity among those studies was much higher (I² = 47%). We also found statistically nonsignificant results when we examined studies with moderate circumcision prevalence (22%-66%) and studies with large analytical samples (>1000 participants), but slightly more protective results for studies that used both a genital examination to determine circumcision status and a diagnostic test to determine HIV infection. The association between circumcision and HIV infection among the subset of higher-quality studies (determined by the Newcastle-Ottawa Scale) was protective, but statistically nonsignificant (OR, 0.79;95% CI, 0.44-1.40; k = 5; I² = 53%). However, in the meta-regression, being circumcised was associated with a reduced odds of HIV infection as study quality scores increased (β, −0.415; P = .01).

A separate analysis (not shown in Table 2) of 4 findings from 3 studies^29,33- 34 reporting HIV infection and circumcision status for MSM who engaged exclusively or primarily in insertive anal intercourse (n = 2238) was protective, but not statistically significant (OR, 0.71; 95% CI, 0.23-2.22; k = 4). The power for this analysis was 0.94. Although there was high heterogeneity among these 4 findings (I² = 90%), too few findings were available for a stratified analysis.

Seven studies^{22- 23,27,29,32- 33,35} contributed 8 findings for the analysis of STIs other than HIV (Figure 3). All but 1 of the studies³⁵ were cross-sectional and conducted in North or South America and the analysis included 15 233 circumcised MSM and 11 003 uncircumcised MSM. A study³² of 899 MSM that provided an effect size for the association between circumcision status and STI, but did not provide the prevalence of circumcised and uncircumcised MSM, was also included in the analytical sample. The weighted overall effect size for the association between circumcision status and STI among MSM was not statistically significant (OR, 1.02; 95% CI, 0.83-1.26; k = 8). The power for the meta-analysis of STI studies was 0.67 and heterogeneity was moderate (I² = 36%). In stratified analyses (Table 3), circumcision was not significantly associated with STIs (bacterial or viral) generally, with specific STIs (ie, syphilis, herpes simplex virus) or with any of the other study or sample characteristics in Table 3. There was also no statistically significant association between study quality, based upon the Newcastle-Ottawa Scale, and the effect sizes of the STI studies (β, 0.265; P = .47).

In this meta-analysis of 15 observational studies of the association of circumcision status and HIV infection among 53 567 MSM, the odds of being HIV positive were 14% lower among MSM who were circumcised than among MSM who were uncircumcised, but the difference was not statistically significant. When we restricted the analysis to studies of MSM who reported primarily engaging in insertive anal sex, the aggregated findings also were statistically nonsignificant. The STI analyses similarly revealed no statistically significant association by circumcision status among MSM. Additionally, we had sufficient power for the HIV analyses and adequate power for most STI analyses. Taken together, these findings indicate insufficient evidence among available observational studies conducted with MSM of an association between circumcision and HIV infection or other STIs.

Several important findings emerge from the results of our meta-analysis. First, we found a statistically significant protective association for circumcision among MSM in studies conducted before the advent of HAART, but a statistically nonsignificant association for studies conducted after HAART. A possible explanation for this difference may be related to an increase in the sexual risk behaviors of MSM after HAART. It has been well documented that beliefs that HAART limits HIV transmissibility are associated with increases in sexual risk behavior among MSM,⁴³ and that the era since the advent of HAART has been defined by higher rates of sexual risk behaviors among MSM,^44- 48 outbreaks of STIs,^49- 51 and increasing rates of HIV infection.^52- 56 Higher rates of sexual risk behavior among MSM since the availability of HAART may diminish the relative effectiveness of male circumcision, and is supported by studies of MSM that report that behavioral risk factors (eg, unprotected anal sex) contribute comparatively more to HIV seroconversion than circumcision status.³⁷

Second, circumcision was not associated with STI in the overall or the stratified analyses of study and sample characteristics but the power for several STI analyses was low. Stratified analyses among higher-quality studies and among studies with moderate circumcision prevalence indicated that circumcision was associated with a nearly significant greater odds of acquiring STI, but caution should be undertaken in interpreting these results because only 2 studies were included in each stratified analysis. Further, these findings may be the result of uncontrolled confounding since STIs can be transmitted through routes other than insertive penile intercourse (ie, oral, digital-anal, skin-to-skin, receptive anal sex),^57- 61 thus bypassing any effect associated with circumcision status.

Third, our review indicates that the available scientific literature examining circumcision status and HIV/STI among MSM are of varied methodological quality. Studies in our meta-analysis may not have been designed specifically to examine HIV/STI in relation to male circumcision and our statistically nonsignificant findings may be due to variability among studies and not the absence of an effect. Nearly two-thirds of the reports in our meta-analysis did not have sufficiently high methodological quality, the point estimates as well as the widths of the CIs included a wide range of values, and between-study heterogeneity was moderate to high in the overall and stratified analyses. A separate analysis of HIV outcome studies with the highest methodological quality revealed a protective but statistically nonsignificant association between circumcision and HIV infection. However, we found that being circumcised was associated with a reduced odds of HIV infection as study quality increased.

Some patterns in the stratified HIV analyses are of interest as they point to the potentially protective role of circumcision among MSM. Although not statistically significant, circumcision had a stronger protective association with HIV infection in study samples in which circumcision prevalence was 50% or lower compared with samples with a greater than 50% circumcision prevalence. A similar pattern was apparent in stratified analyses by country. The aggregated US studies, where circumcision prevalence is relatively high,⁶² had a 9% reduction in odds of HIV infection among circumcised MSM. By contrast, the reduction in odds was higher in countries with lower adult circumcision rates than the United States⁶³ (15% reduction in studies conducted in other developed countries and 51% reduction in developing countries). We also found a comparatively stronger, albeit nonsignificant, protective association of circumcision with HIV infection in MSM studies in which circumcision status was determined by genital examination, HIV status was determined by diagnostic test, multivariate analyses adjusted for confounders, and samples were limited to MSM who primarily engaged in insertive anal sex. Moreover, we found a 53% reduction in the odds of HIV infection among circumcised MSM during the era before HAART, which is comparable with the 50% to 60% reduction in the odds of HIV infection among circumcised heterosexual men in the 3 African RCTs^1- 3 and a previously published meta-analysis of heterosexual African men.⁶⁴

Our meta-analysis has several strengths. First, it is the most comprehensive meta-analysis of male circumcision and HIV or STI risk among MSM. Second, we stratified the data by several study design and sample characteristics to gauge the presence or absence of an association. Third, half of our analytical sample (over 26 000MSM participants) was derived from unpublished data, minimizing concern over publication bias. Fourth, our meta-analysis includes domestic and international studies of MSM, including studies from developed and developing countries. Last, our conclusions are informed by power analyses, by analyses of heterogeneity of the aggregated findings, and by analyses that stratified studies according to methodological quality.

There are also several limitations to our review. The studies in our meta-analysis are observational and most did not control for potentially confounding variables. However, 4 studies^23,29,33,37 did provide multivariate findings that controlled for age and HIV risk behaviors, and these aggregated analyses were statistically nonsignificant. Another limitation is that a substantial proportion of the studies in our meta-analyses were cross-sectional, which limits inferences of HIV or STI incidence. Similarly, the analyses of MSM who engaged in insertive anal intercourse were derived from cross-sectional studies and likely include MSM who engaged in receptive anal sex during their lifetime, which may bias the association between circumcision and HIV infection toward the null. An additional limitation is that moderate to high heterogeneity between studies remained in many of our stratified analyses. There were also not enough data to examine STIs separately, aside from syphilis and herpes simplex virus, and aggregating all STI data in our analyses possibly diluted our ability to detect any effects. Last, we found evidence of publication bias in the STI data. However, 3 of the 7 (43%) STI studies were unpublished studies and aggregated analyses revealed no statistically significant associations between circumcision and STI for published or unpublished data.

Considerable gaps remain in the available literature of male circumcision among MSM. Additional prospective studies of MSM need to be conducted that meet high quality assessment criteria (eg, genital examinations, diagnostic tests for HIV/STI, multivariate analyses).⁶⁵ There also needs to be more data on the proportion of MSM who only engage in insertive anal sex, whether such proportions differ by race/ethnicity, age, geography, cultural context or other factors, and the population-level impact of new HIV infections possibly averted by circumcising these men. Further research studies should also examine sexual behavior differences between circumcised and uncircumcised MSM and how these differences may be related to HIV transmission. In addition, future research should compare the relative effectiveness of current HIV prevention strategies (eg, behavioral interventions targeting MSM)⁶⁶ with mathematical models of male circumcision and HIV infection among MSM. Last, it is unknown whether differences exist in viral shedding in the anus vs the vagina, and how potential differences in viral load in either cavity may affect the likelihood of HIV transmission to an uncircumcised, seronegative, insertive male partner per act of unprotected intercourse.¹⁰

Serious consideration should be given to conducting an RCT under conditions that would be optimal (ie, high acceptability of circumcision, high HIV prevalence, low circumcision prevalence) to evaluate the potential prevention implications of circumcision among MSM. The association between circumcision and HIV infection may not be uniform across all groups of MSM and any proposed trial might consider enrolling HIV-negative MSM who primarily engage in insertive anal sex and/or reside in resource-deprived settings without HAART access. However, recruitment challenges and ethical considerations need to be thoroughly addressed before initiating any RCT with MSM.

We found a protective, albeit statistically nonsignificant, association of circumcision with HIV infection in our meta-analysis of MSM observational studies, and a statistically nonsignificant association between circumcision status and STI. Our data revealed that male circumcision conferred a significant protective effect from HIV infection among MSM in studies conducted before HAART but not after, possibly due to documented increases in sexual risk behavior during the era since the availability of HAART.^44- 48 Additional studies are necessary to elucidate further the relationship between circumcision status and HIV infection or STIs among MSM.