There is compelling evidence that male circumcision (MC) reduces susceptibility to HIV infection. Early evidence was based on ecological and observational studies.1^–9 Results from a meta-analysis of observational studies showed reductions of 50% and 70% in HIV risk among circumcised men from the general population and higher-risk groups, respectively.6 The most compelling evidence comes from three recent non-blinded randomised control trials conducted among adult men in Kenya, Uganda and South Africa10^–12 which suggested a reduction of 50–60% in HIV risk among circumcised men across the three trials (table 1).

Given the overwhelming evidence and the limited preventive options available, WHO/UNAIDS have published recommendations for countries to consider scaling up access to MC in seronegative men in areas of high HIV prevalence where MC is rare.13 However, a number of important issues such as the safety, cost, feasibility, acceptability, ethics and potential increase in risky behaviour following circumcision should be considered, ideally before the large-scale implementation of MC.13 A better understanding of the potential population-level impact of MC is needed in order to identify who should be circumcised. Further research is needed to guide programme implementation and to better understand additional benefits or risks of MC, including the protective effects of MC on other sexually transmitted infections (STIs).13 Without conducting community-based randomised trials, the impact of circumcision at the population level can be addressed with mathematical modelling if we have a clear understanding of the protective biological mechanisms of MC at the individual level. To achieve this, it is useful to understand clearly the results of the circumcision trials and to make a distinction between efficacy and effectiveness of MC.

In this paper we first review and discuss the efficacy and effectiveness of MC on HIV, in particular the incremental benefit of MC on HIV infection due to its indirect protection against cofactor STIs in the context of the three aforementioned randomised trials.10^–12 In addition, we present new results on the incremental benefit of MC at the population level owing to its efficacy against other STIs.

METHODS

The impact of MC on the prevalence of HIV at the population level was assessed using a previously validated stochastic compartmental model which simulates transmission of HIV and one STI in the heterosexual population in the Kisumu district of Kenya.12 14 The modelled population was stratified into six sexual activity classes with specific rates of sexual partner acquisition. In absence of specific data, the mixing between activity classes was assumed to be proportionate. The STI was modelled with two compartments representing infected or not infected individuals. Infected individuals could recover from the STI and be reinfected. Without loss of generalisability, the modelled STI should be thought of as “generic” rather than representing any one specific STI. HIV infection has been modelled with five stages representing susceptible, acute, asymptomatic, pre-AIDS and AIDS. The progression of individuals between states was based on specific disease stage parameters or on the force of infection (for susceptible) which depended on the sexual activity, HIV or STI infection status, HIV and STI prevalence in the pool of partners, the strength of the HIV-STI interaction and the efficacies of MC against HIV (E_S^HIV) and STI (E_S^STI). Upon commencement of the circumcision intervention, a fixed number of susceptible men were recruited from the uncircumcised to the circumcised susceptible compartment. The model structure, the equations and the Monte-Carlo simulation process are fully described in the online supplement.

High (scenario A) and low (scenario B) STI prevalence scenarios were modelled. The distribution by sexual activity classes and rates of sexual partner acquisition for scenario B were selected to agree with infection rates in the Kisumu UNIM male circumcision trial population12 14^–17 and the 2003 Kenyan Demographic and Health Survey15 corresponding to years 15–17 of the simulated epidemic. Remaining parameters for HIV14 18^–23 and STI24^–26 transmission probabilities, duration of the different HIV states14 18^–23 27^–29 and duration of STI infection24^–26 were based on published studies (table 2).

The presence of STI in the HIV-infected sexual partner was assumed to increase HIV per partner infectivity by fourfold, while STI in the HIV-susceptible increased per HIV susceptibility by threefold. This reflected the results from a meta-analysis of observational studies where the relative risk of HIV due to STI in men was less than 4.4 for genital ulcer disease (GUD), 3.1 for all STIs, 2.7 for herpes, 2.5 for syphilis, 3.9 for gonorrhoea and 0.8 for chlamydia.30^–33 HIV infection was also assumed to increase susceptibility to and infectivity with STI by 1.5-fold.30^–32 Scenario A was obtained from scenario B by increasing the average duration of STI infection from 6 months to 1.3 years. Because this change also increased HIV infection rates as a result of the STI and HIV interaction, we simultaneously decreased HIV transmission probabilities during the early stage and slightly modified the rates of sexual partner change in order to hold HIV infection rates to reasonable levels. The parameter values used for both scenarios and the key characteristics of the simulated STI and HIV epidemics are summarised in table 2. In both scenarios the male HIV prevalence was 28% at year 15 and declined as the epidemic progressed owing to a strong dependence on the prevalence of STI which declined due to AIDS differential mortality. In scenario A the overall prevalence of STI and HIV and the incidence of HIV averaged 22.3%, 29.8% and 5.9 per 100 person-years over the 20 years following the intervention (introduced at year 15), respectively, compared with 3.7%, 16.4% and 1.9 per 100 person-years, respectively, for scenario B.1^–3 12 14^–17

To assess the individual-level impact of MC, Desai et al14 simulated the Kisumu MC trial by recruiting and following 2750 initially uncircumcised seronegative men from the overall simulated population and randomly assigning them to the circumcision or control arm. The simulated follow-up was 24 months, HIV incidence rate was 2.5% per year in control subjects and the prevalence of STI averaged 8.2% in controls over the 2 years, which compares with the overall prevalence of bacterial infections observed in the UNIM trial participants at baseline.

To assess the population-level impact of MC, new simulations were performed where a large-scale mass circumcision programme was initiated in a mature HIV epidemic in low and high STI prevalence settings (scenarios A and B). Low to high values of MC efficacy against STI (six scenarios in all) were assumed in MC interventions offered to 75% of HIV susceptible men (either STI positive or STI negative) instead of offering it to a minority of men (as in the simulated trials) where coverage was too small to generate herd immunity.

Theoretical context

Individual-level effect of MC: efficacy vs effectiveness

In the clinical trial literature, “efficacy” is typically defined as the individual-level clinical/biological benefit of the intervention used under ideal conditions (eg, with 100% compliance and adherence), reflecting the maximal effect it can have. Individual-level “effectiveness”, often termed “real world” effectiveness (not to be confounded with the population-level effectiveness) refers to the effect of the intervention achieved under more realistic conditions of use (eg, imperfect adherence) and relates more closely to the potential benefit of the intervention to individuals when widely used in practice.33 34

Determining the efficacy of MC is more complex because MC can have different individual-level efficacies, reflecting different biological/clinical protective mechanisms conferred to individuals directly against HIV or indirectly against other STIs,14 34 such as:

• Reduction in susceptibility to HIV infection (E_S^HIV) or to other cofactors STI (E_S^STI);

• Reduction of the infectiousness of HIV-infected (E_I^HIV) or STI-infected (E_I^STI) circumcised men for their sexual partners;

• Modification of the natural history of STI infection by reducing the frequency of ulcers or by accelerating the natural clearance (E_nh^STI).

In randomised controlled trials where HIV-negative individuals are randomised and followed up, it is only possible to estimate a reduction in acquisition to HIV (E_S^HIV) and STIs (E_S^STI) independently.34

Because cofactors STI were on the causal pathway to HIV infection, the primary outcome of the three circumcision trials was the overall effectiveness of MC on HIV acquisition rather than the independent efficacies against HIV and STIs.11 Thus, the overall effectiveness (ie, the overall reduction in HIV incidence) measured in the trials could have been the result of a combination of direct protection (efficacies) against HIV acquisition (E_S^HIV) and/or against acquisition of cofactor STI (E_S^STI). For example, in the South African MC trial the overall effectiveness of 60% could be due to an efficacy against HIV only (E_S^HIV = 60%, E_S^STI  = 0%) or to some degree of protection against both HIV and STI (0%<E_S^HIV<60%, 0%<E_S^STI <60%).

Population-level and long-term impact

MC has direct effects on HIV at the individual level and additional indirect effects at the population level due to herd immunity (ie, a reduced exposure to infection among uncircumcised men or women due to reduced prevalence of STIs or HIV).31 The total population-level impact is a function of the different individual-level efficacies (E_S^HIV, E_S^STI, E_I^HIV, E_I^STI, E_nh^STI), thus justifying the need to quantify each of them when possible.

It is also important to evaluate the incremental role played by MC in reducing the incidence of new STI infections and indirectly preventing new HIV infections. Knowledge of E_S^HIV and E_S^STI can help to interpret the MC effectiveness results on HIV and to extrapolate the results to other communities. If most of the protection against HIV was obtained indirectly via the protection against STIs, the trial results would depend more strongly on the epidemiology of STIs and the trial population and would therefore have a poorer external validity. The potential impact of MC at the population level would also depend on the prevalence of the different STIs and the associated synergetic impact with HIV.

Biologically, MC may reduce susceptibility to HIV infection because the underside of the foreskin is rich in HIV target cells (CD4+ T cells). It may also reduce the risk of abrasion, micro tearing and inflammatory conditions suffered by the inner mucosa of the foreskin in uncircumcised men.10^–13 35 36 Only one observational study in Rakai suggested a reduction in infectiousness of HIV-positive circumcised men for their female partners.8 However, in a very recent clinical trial in Rakai, a higher (but not statistically significant) HIV incidence was observed among the wives of circumcised HIV-positive men, which was attributed to premature resumption of sexual activity following the surgical procedure rather than behavioural disinhibition.37 The protective effect of MC against STIs is more uncertain.37^–45 A meta-analysis suggested a reduction in the risk of syphilis and herpes simplex virus-2 (HSV-2) among circumcised men of 33% and 12%, respectively.38 The estimates across different observational studies varied from −10% to 88% for chancroid. Evidence on the protective effect against gonorrhoea is unclear and mostly based on early studies.9 39 41^–44 One study suggested a reduced rate of chlamydial transmission to their female partners by circumcised compared with uncircumcised men.45 Thus, in the results presented below, the MC efficacies were modelled as a reduction in the susceptibility of men to HIV infection which was fixed to 60% (E_S^HIV). The efficacy against STI (E_S^STI) varied between 0% and 70%. Low E_S^STI (∼0–20%) reflected the efficacy of MC on chlamydia and HSV-2 while high E_S^STI (∼60–80%) reflected the efficacy against chancroid and syphilis.11 38 40 We assumed that male-to-female HIV transmission was unchanged by circumcision status (table 2).

RESULTS

Insights from previous clinical trial simulations

The protective role of MC against cofactor STI on the risk of HIV infection at the individual level was assessed in a study by Desai et al14 where a small fraction of men were followed up for 2 years in the simulated UNIM MC trial in Kisumu (table 1). Under the UNIM simulated conditions, if MC did not protect against HIV (E_S^HIV = 0%) but strongly protected circumcised men (HIV+ or HIV−) against the STI only (E_S^STI = 80%), the overall effectiveness against HIV would be only 13%. If the trial duration was prolonged to 5 years and MC efficacy against STI was increased to 100%, the overall effectiveness against HIV, due to the protection against STI alone, increased to 21%. The effectiveness reached 25% or 30% only when the STI prevalence was increased to 19% or when the relative risk was increased to sixfold, respectively. As these assumptions—especially the 100% efficacy against the STI—appeared unrealistically high, the authors concluded that the effectiveness above 50%, as observed in the field randomised controlled trial, could not have been due solely to the protection against cofactor STI. MC needed to strongly protect directly against HIV.14 In addition, under the STI conditions, MC needed to have an HIV efficacy of at least 40% and 50% to generate the observed overall effectiveness of 50% or 60% against HIV, respectively, even if the efficacy against STI was as high as 60%.14 This also meant that, if the MC efficacy against HIV was 40%, not more than 20% of the HIV infections prevented in the circumcised arm of the trial could be attributable to the efficacy against the STI (rather than efficacy against HIV). This proportion decreased as the efficacy against HIV increased (fig 1A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 (A) Individual-level effectiveness. Estimated fraction of new HIV infections prevented over 2 years of follow-up in the circumcised arm of the simulated Kisumu trial that is due to the indirect protection of male circumcision (MC) against the sexually transmitted infection (STI) rather than to a direct protection against HIV (y axis). In these simulations the efficacy due to a reduction in susceptibility to both STIs and to HIV (E_S^STI and E_S^HIV) was varied. The fraction of HIV infection prevented by the MC intervention in the simulated trials is fixed to 100%. Thus, if the E_S^STI component prevented 20% of new infections, the remaining 80% prevented was due to E_S^HIV. As E_S^HIV increased, the contribution of E_S^STI to the total infections prevented declined. (B) Population-level effectiveness. Overall male and female HIV prevalence over time without and with the circumcision intervention in scenario A when E_S^HIV  =  60% and E_S^STI varied from 0% to 70%. 75% of uninfected men are reached and circumcised at the beginning of the intervention introduced in a mature epidemic. Scenario A: STI prevalence averaged 24.3% and 20.2% in women and men, respectively.

Insight from the three MC circumcision trials

The effect of circumcision on the incidence of STI was not reported in the South African trial10 and only the baseline prevalence of STI was reported in the Kenyan trial.12 Gray et al11 reported a baseline prevalence of 7% and a reduction of 47% (95% CI 36% to 57%) in self-reported GUD in the circumcised arm during part of the trial (table 1). In subgroup analysis, an effectiveness of 40% (95% CI 8% to 66%) against HIV was reported in the GUD negative group compared with an effectiveness of 51% (95% CI 16% to 72%) for the whole cohort. In line with the meta-analysis results for high-risk individuals,6 the Rakai subgroup analyses reported an effectiveness of 71% (95% CI −29% to 97%) in men reporting more than two partners and 70% (95% CI 15% to 91%) in those with self-reported GUD. Based on our simulation results, the difference between an effectiveness of 70% and an effectiveness of 50–60% among high-risk individuals compared with the general population could partly be explained by the additional protection of MC against STIs if the STI prevalence (>20%) or the recurrence of ulcers among high-risk individuals is high. In addition, under the simulated UNIM trial conditions, MC efficacy against HIV was predicted to be at least 40–50% given that the observed individual-level MC effectiveness in the three trials was 50–60%.14

Insights from new simulation results: population-level and long-term impact

Because the trials were of short duration and only captured the reduction in susceptibility to HIV infection (at the individual level), the field and simulated trials do not necessarily reflect adequately the incremental impact of MC on HIV at the population level owing to its protective impact against cofactors STI. The full incremental impact of MC efficacy against STIs may be larger at the population level and over a longer time scale when coverage increases. To understand the extent of the impact of MC efficacy against STI, we simulated a high (A) and low (B) STI scenario with a strong STI-HIV interaction.

Figure 1B shows the impact of MC on the prevalence of HIV over time for scenario A. The figure shows that MC can reduce the long-term prevalence of HIV in men and that women would benefit somewhat from the herd immunity effect, but not noticeably until 5–10 years after the intervention. Importantly, the figure also highlights the small incremental impact of MC due to MC efficacy against STI in the long term, even if E_S^STI was as high as 70% and the prevalence of STI was high. The incremental benefit of MC E_S^STI was slightly greater for women than for men because the prevalence of STI declines faster than the prevalence of HIV in men, and women benefit from the reduced HIV infectivity of men who avoid STIs. Figure 2 summarises the percentage reduction in the prevalence of HIV 5 and 15 years after the introduction of the MC intervention. MC efficacy against the STI has a limited impact, even in the long term, unless the prevalence of STI is high (scenario A) and E_S^STI is higher (>70%) than currently suggested by data. In the long term, women could benefit substantially from the herd immunity effect of MC (20% reduction in HIV prevalence). Even at the population level, MC efficacy against HIV produced most of the benefit on HIV in both men and women. Only under specific conditions of very high STI prevalence, high synergetic interaction or large E_S^STI could MC efficacy against STI produce a noticeable incremental benefit for men and women.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Percentage reduction in overall prevalence of HIV in men and women in uncircumcised and circumcised populations (A) 5 years and (B) 15 years after a circumcision intervention. 75% of uninfected men are reached and circumcised at the beginning of the intervention which is delivered 15 years after the beginning of the HIV epidemic in scenarios A and B. Scenario A: STI prevalence averaged 24.3% and 20.2% in women and men, respectively. Scenario B: STI prevalence averaged 2.9% and 4.5% in women and men, respectively. MC, male circumcision; STI, sexually transmitted infection.

DISCUSSION

In this paper we distinguished between the overall effectiveness of MC and the efficacies of MC against HIV and STIs in order to assess the population-level impact of MC on HIV. Together, the field and simulated trial results suggested that only a small fraction of the observed MC effectiveness against HIV in the randomised controlled trial could be due to the indirect efficacy against STIs rather than the efficacy against HIV. The fact that E_S^STI contributed little to the overall individual level effectiveness may explain the consistency of the estimates, which varied by less than 10% across trials despite differences in STI prevalence. Thus, the three circumcision trials have good external validity and the results can be extrapolated to settings with different STI epidemiology. To generalise the trial results to other risk populations and in settings outside sub-Saharan Africa, the mechanism of protection against HIV—particularly the relative role of hygiene compared with biological mechanisms—remains to be clarified.

Our results support previous modelling studies suggesting that MC has the potential to help curb the HIV epidemic in the long term in populations where the prevalence of MC is low and in the absence of associated sexual disinhibition.46^–51 If enough men are circumcised, women will benefit from long-term herd immunity effects. In our model, men were circumcised in a short window period. In practice, such high coverage would be reached over a longer period. It is therefore important to determine who should be circumcised first (eg, younger and more sexually active men) in order to scale-up the circumcision programme to achieve maximum impact very rapidly.51 The population-level impact of MC would be even larger, especially for women, if it also reduced the infectiousness of HIV-positive circumcised men. However, it could also have detrimental effects if men were more infectious to their female partners immediately after the procedure.37 46 This has obvious implications for the roll-out of mass circumcision.

Our analysis suggests that, in the long term, the incremental population-level impact of MC on HIV due to a reduction in E_S^STI among men is also likely to be small. Although we only modelled one generic STI, our conclusions remained valid under extreme assumptions of high MC efficacy against STI (higher than has been observed), a strong association between STI and HIV and a very high prevalence of STI (fig 2). The incremental benefit of E_S^STI was marginally better for women than for men, simply because their protection was mediated by herd immunity effects through the rapid decline in the prevalence of STI among men. Based on current knowledge, we assumed that MC reduced acquisition of new STIs. However, if MC also reduced the frequency or duration of ulcers during the course of infection,11 40 this may provide additional incremental benefits. Considerable uncertainty remains regarding MC efficacy against STIs. The Rakai circumcision trial on HIV acquisition reported a 50% reduction in self-reported GUD.40 However, results from a complementary trial observed a 25% reduction in acquisition of HSV-2 among HIV-negative circumcised men and a 25% reduction in GUD.40 These estimates of MC efficacy against HSV-2 were greater than those suggested in Weiss’s meta-analysis.35 38 The incremental effect of MC on HIV in women would be larger if MC also reduced bacterial vaginosis and trichomoniasis in the wives of circumcised men.40

Our results do not indicate that MC does not protect against STIs. They only predict that the incremental impact of MC against HIV due to a reduction in the susceptibility of men to STIs is likely to be relatively small in generalised HIV epidemics. Additional work is needed to identify whether, and under what conditions, MC efficacy against STIs can produce more significant benefits under different epidemic characteristics (eg, concentrated, rising) and to better understand how MC protects against STIs. Even if MC efficacy against STIs did not play an important role in preventing HIV, it would still be beneficial to reduce the burden of STIs.