Male circumcision for the prevention of heterosexually acquired HIV infection: a meta-analysis of randomized trials involving 11 050 men


  • *See editorial by Lazarus et al. [1] on pp. 327–328 and article by Rice et al. [2] on pp. 329–331 in this same issue.

Dr Edward Mills, St Paul's Hospital, British Columbia Centre for Excellence in HIV/AIDS, 608-1081 Burrard Street, Vancouver, BC, Canada V6Z 1Y6. Tel: +1 604 806 3727; fax: +1 604 806 9044; e-mail:



Observational studies and a small collection of randomized controlled trials (RCTs) suggest that male circumcision may significantly reduce HIV transmission between sero-discordant contacts. The Joint United Nations Programme on HIV/AIDS (UNAIDS) and World Health Organization have recently announced recommendations to scale up male circumcision in countries with generalized epidemics and low levels of male circumcision. However, no meta-analysis has been conducted to determine the effectiveness of this intervention.


We conducted a systematic review of medical literature, and included any RCTs assessing male circumcision to prevent heterosexually acquired HIV infection among males. We used the DerSimonian–Laird random effects method to pool study outcomes. We calculated the relative risk (RR), risk difference, number needed to treat (NNT) and I2, all with 95% confidence intervals (CIs).


We identified three RCTs that met our inclusion criteria, involving a total of 11 050 men. The pooled RR was 0.44 (95% CI 0.33–0.60, P<0.0001, I2=0%, 95% CI 0–35%). The risk difference was 0.014 (95% CI 0.07–0.21), yielding a NNT of 72 (95% CI 50–143).


Male circumcision is an effective strategy for reducing new male HIV infections. Its impact on a population level will require consistently safe sexual practices to maintain the protective benefit.


In 2007, there were an estimated 33.2 million (plausibility range 30.6–36.1) people living with HIV/AIDS globally, with some 22.5 million (range 20.9–24.3) of these adults and children living in sub-Saharan Africa. We have every reason to believe that this figure will continue to increase unless effective interventions can slow the progress of the epidemic [3]. In March 2007, the Joint United Nations Programme on HIV/AIDS (UNAIDS) and the World Health Organization announced recommendations to scale up male circumcision in countries affected by generalized epidemics that currently have low levels of male circumcision [4]. Male circumcision represents an important intervention in combating HIV/AIDS: systematic reviews of observational studies have indicated the important protective effects of this intervention [5,6]. Recently, the first randomized clinical trials were completed. We have conducted the first meta-analysis of the recently completed randomized trials.


We included any randomized trial assessing male circumcision to prevent heterosexually acquired HIV infection among males. We systematically reviewed the medical literature and searched electronic databases (MedLine, EMBASE, CINAHL), electronic conference websites (IAS, CROI) and clinical trial registries (, Meta-Register). The completed trials have been well publicized and we contacted the study authors. We abstracted data independently, in duplicate, addressing trial setting, participants, trial duration and number of infections in each group (active and controls). Additionally, we abstracted data on methods of allocation concealment, randomization and adherence to the intention-to-treat principle.

For meta-analysis, we used the DerSimonian–Laird random effects method to pool the study outcomes. This method recognizes and anchors studies as a sample of all potential studies, and incorporates an additional between-study component to the estimate of variability [7]. Our primary endpoint was the number of patients seroconverting. We excluded patients who were HIV-positive at baseline but were still randomized [8]. We calculated the relative risk, risk difference and number needed to treat (NNT), all with 95% confidence intervals (CIs). Baseline risks across the groups were similar (1.78–4.2% over 24 months). We also calculated the I2 statistic as a measure of the proportion of the overall variation that was attributable to between-study heterogeneity, with its 95% CI [9]. We used Comprehensive Meta-Analysis Version 2.1 (Biostat Inc., Englewood, NJ, USA) and StatsDirect Version 2 (StatsDirect Ltd, Manchester, UK) for all calculations.


We identified three randomized trials that met our inclusion criteria [8,10,11]. Table 1 presents the study characteristics and individual study outcomes. All studies adequately reported the a priori determined methodological issues. All trials stopped prior to complete enrolment because of Data Safety and Monitoring Board recommendations of clear effectiveness.

Table 1.   Study characteristics and outcomes
StudyDesignSettingPopulationnOutcomesRelative risk
(95% confidence
  • *

    3274 randomized, 3128 included in analysis.

  • **

    ** 2784 randomized, 2780 included in analysis.

Auvert et al. (2005)*Randomized trialOrange farm, South AfricaMales aged 18–24 years312820/154649/15820.42 (0.25–0.70)
Bailey et al. (2007)**Randomized trialKisumu, KenyaMales aged 18–24 years278019/138846/13920.41 (0.24–0.70)
Gray et al. (2007)Randomized trialRakai district, UgandaMales aged 15–49 years499622/247445/25220.50 (0.30–0.83)

Our pooled analysis indicates a RR of 0.44 (95% CI 0.33–0.60, P≤0.0001, I2=0%, 95% CI 0–35%; see Fig. 1) in favour of circumcision, corresponding to a RR reduction of 56% (95% CI 40–67%). The risk difference is 0.014 (95% CI 0.07–0.21), yielding a NNT of 72 (95% CI 50–143).

Figure 1.

 Random effects meta-analysis.


Our meta-analysis should be of interest to policy-makers, clinicians and the public. There is a large and consistent effect of circumcision in the prevention of heterosexually acquired HIV infection. As new prevention technologies prove challenging, circumcision appears to be an inexpensive and effective prevention strategy. Challenges will now exist in expanding access to circumcision and addressing cultural concerns about the acceptability of the intervention [12].

There are several strengths and limitations to consider in our analysis and in the included trials. Strengths of our review include our extensive searching, contact with authors and policy-makers, and use of the random-effects model to provide a more conservative estimate. Limitations of our review include our inability to conduct sensitivity analysis because of the low number of trials included. However, given the consistency of findings, it is unlikely that any trial-level characteristics influenced the outcomes. Our meta-analysis used binary estimates. It is possible that our results would change marginally had we used time-to-event data.

Limitations of the trials also exist. All trials were stopped early; it is likely that, because of stopping early, results represent an over-estimate of the true reduction in infection rates [13,14]. However, the magnitude of the estimates is so great, and the results so consistent, that this limitation does not threaten the inference that, when administered to similar populations in a similar fashion, circumcision results in an appreciable RR reduction.

Participants in these trials received education to reduce their likelihood of infection, including safe-sex counselling and the recommendation of abstinence during the healing period. It is possible that this education will impact the generalizability of the trials because it may reduce the number of exposures that participants have in comparison to the general population. The trials were unable to blind participants to the intervention and control and, as a result, some participants in the control groups received the intervention outside of the trial. To counter this effect, the investigators also performed a per-protocol analysis and found a similar magnitude of effect.

Media attention has focused on the ‘60%’ reduction in infections observed initially in the trial by Auvert et al. [11]. However, our pooled analysis indicates that the protective effect of male circumcision may be somewhat different and that the actual population effect of the intervention may be less compelling. The NNT of 72 (95% CI 50–143) suggests that approximately 72 circumcisions will have to be conducted over a 2-year period to prevent a new infection, although this will differ in populations with varying baseline risks. Researchers have modelled the cost-effectiveness and population impact of widespread circumcision. Recent models assessing circumcision in South Africa and Uganda found savings of US$2411 and US$2631 per infection averted, respectively [15,16]. While the costs of the surgery will vary by country, the lives saved from prevented infections and prevented antiretroviral provision are likely to overwhelm the costs of the brief surgery.

A number of considerations suggest that the effect seen in the trials may not be reproduced with widespread dissemination of the intervention. With circumcision, this is observed as often longer than 6 weeks [17]. Men who engage in sex during the healing period may place themselves at increased exposure to infection. Counselling patients on the abstinence period may prove to be a challenge for successful implementation.

In addition, circumcised men may have an exaggerated sense of protection from sexually transmitted diseases including HIV that could influence their behaviour. Currently, we do not know how circumcision will impact upon behaviours; however, a modelling study from Uganda indicated that an increased number of sexual partners will counteract the beneficial impact of circumcision [16].

This situation raises important considerations for the conduct of preparedness studies addressing new prevention strategies. Despite the plethora of preparedness studies assessing potential HIV vaccines, comparatively few studies have addressed circumcision. While vaccines remain elusive, circumcision now represents an important tool in preventing new infections. As more circumcision data are developed there will be greater pressure on ministries to promote circumcision, potentially challenging local traditions. For interventions to have an important population-level effect, there is a need for local buy-in. Assessing cultural acceptability and strategies to overcome barriers will represent one of the greatest challenges for prevention strategies this year.