Approximately 1.7 million people were newly infected with HIV worldwide in 2019, most of whom resided in southern and eastern Africa.1 Substantial scale-up of prevention and treatment programs over the past decade has made possible the goal of containing the epidemic and ultimately eliminating HIV. Male circumcision (MC) reduces the risk of heterosexually acquired HIV infection by approximately 60% among men^2–5 and is an essential component of the HIV prevention program in southern and eastern Africa, where HIV is transmitted primarily heterosexually and MC prevalence is relatively low. Since 2009, the US President’s Emergency Plan for AIDS Relief (PEPFAR) has supported Ministries of Health and Defense to offer voluntary medical male circumcision (VMMC) services for HIV prevention in 15 southern and eastern African countries with generalized HIV epidemics and low MC prevalence. Through 2020, approximately 25 million (of 27 million total) VMMC services had been supported by PEPFAR.⁶

Following years of observational studies showing an association between MC and lower HIV prevalence, 3 randomized controlled trials published from 2005 to 2007 definitively established the HIV preventive benefit of MC^3–5; the durability of effect has been demonstrated in longer-term follow-up of participants from these randomized controlled trials.^8–10 More recently, observational data from community cohorts where VMMC programs have been implemented have demonstrated greater HIV incidence reduction among circumcised men and, in some instances, their female sexual partners than among uncircumcised men.^11–15 These studies did not account for concomitant antiretroviral treatment (ART) scale-up, although modeling has shown that ART scale-up largely explains near-term declines in HIV incidence whereas MC explains longer-term declines.¹⁶ Nationally representative data to quantify the decline in HIV incidence from the VMMC program, to determine its contribution toward regional epidemic control in the context of ART scale-up, have not previously been available.

The population-based HIV impact assessments (PHIAs) are nationally representative surveys conducted in several African countries with generalized HIV epidemics. The PHIA surveys were designed to directly estimate national HIV incidence among adults using HIV recency testing. This article describes the impact of the VMMC program on male HIV incidence after approximately 10 years of VMMC program scale-up in 8 countries with completed PHIA surveys.

From 2015 to 2017, among the 8 countries, 90,088 men aged 15–59 years were rostered, of whom 80,154 (89.0%) were deemed eligible for participation (see Figure, Supplemental Digital Content, http://links.lww.com/QAI/B621); 68,758 (85.8%) participated in the interview [range: 80.8% (Zambia)–93.7% (Uganda)], and 63,088 (91.8%) of those interviewed participated in the biomarker portion of the survey [range: 86.8% (Malawi)–98.5% (Uganda)]. Overall, 62,344 (98.8%) men with blood results reported their MC status, and among circumcised men, 25,786 (96.6%) reported their MC provider type. Overall, MC prevalence was 32.3% [95% confidence interval (CI): 31.8% to 32.7%] medically circumcised; 19.4% (95% CI: 19.0% to 19.8%) nonmedically circumcised; and 48.3% (95% CI: 47.9% to 48.7%) uncircumcised. Recency testing was performed in 99.9% of specimens of men with positive HIV test results (n = 6089). There were a total of 75 recent HIV infections among men with known MC status: 12 in men who reported medical MC, 17 in men who reported nonmedical MC, and 46 in uncircumcised men.

Medically circumcised men were more likely to be younger, single, and have higher levels of education than uncircumcised men [15–24 years: 47.4% (95% CI: 46.3% to 48.5%) versus 38.6% (95% CI: 37.9 to 39.2) (Table 1); single 50.3% (95% CI: 49.2% to 51.3%) versus 38.4% (95% CI: 37.7% to 39.0%); and higher education 8.4% (95% CI: 7.9% to 9.0%) versus 7.1% (95% CI: 6.8% to 7.5%), respectively]. They were also more likely to live in urban areas than uncircumcised men [44.9% (95% CI: 44.1% to 45.7%) versus rural 26.6% (95% CI: 26.2% to 27.0%), respectively]. Tanzania contributed the largest proportion of men (weighted) to the analysis [39.4% (95% CI: 39.3% to 39.5%)].

Similar proportions of medically circumcised and uncircumcised men reported having sex in the past 12 months [83.9% (95% CI: 83.0% to 84.8%) and 84.6% (95% CI: 84.1% to 85.1%), respectively] (Table 1). Medically circumcised men more frequently had ≥2 sex partners in the past 12 months than uncircumcised men [34.8% (95% CI: 33.6% to 36.1%) versus 26.8% (95% CI: 26.1% to 27.5%), respectively,] and were less likely to know the HIV status of their last sex partner(s) than uncircumcised men [50.2% (95% CI: 48.9% to 51.5%) versus 41.0% (95% CI: 40.2% to 41.7%), respectively]. Men aged 25–34 years had the highest mean risk score among all groups, and for each age group, medically circumcised men had higher mean risk scores than uncircumcised men (see Table 1, Supplemental Digital Content, http://links.lww.com/QAI/B622). HIV prevalence was 6.2% (95% CI: 3.9% to 9.5%) among medically circumcised men, 7.9% (95% CI: 5.1% to 12.1%) among nonmedically circumcised men, and 11.4% (95% CI: 7.5% to 17.1%) among uncircumcised men (Table 2).

Among men aged 15–34 years, medically circumcised men had lower annualized HIV incidence than uncircumcised men [0.04% (95% CI: 0.00% to 0.10%) versus 0.34% (95% CI: 0.10% to 0.57%), respectively; adjusted incidence difference: 0.23% (95% CI: 0.06% to 0.40%); P value = 0.01] (Table 3). Among men aged 35–59 years, the incidence difference was not statistically significant between medically circumcised men and uncircumcised men [1.36% (95% CI: 0.32% to 2.39%) versus 0.55% (95% CI: 0.14% to 0.97%), respectively; adjusted incidence difference: −0.34% (95% CI: −0.79% to 0.11%); P-value = 0.14]. For men aged 15–34 and 35–59 years reporting nonmedical MC, the incidence differences were not statistically significant versus uncircumcised men.

A majority (56.5% [range: 12.0% in Zimbabwe to 71.9% in Tanzania]) of medically circumcised men reported undergoing the MC procedure before VMMC program implementation in their country (see Table 2, Supplemental Digital Content, http://links.lww.com/QAI/B622). As expected, this was more common among men aged 35–59 years than 15–34 years (84.1% versus 47.2%, respectively). Among the 12 recent HIV infections in medically circumcised men, 8 occurred in those reporting to be circumcised before VMMC program scale-up and 4 in those circumcised afterward.

In this large, multicountry analysis, medical MC was associated with reduced HIV incidence among men aged 15–34 years compared with uncircumcised men, even though medically circumcised men in this age group had higher sexual risk scores. This finding demonstrates the effectiveness of the VMMC program in a nationally representative setting, lending strong support for continued VMMC scale-up in countries with generalized HIV epidemics and low MC prevalence. No such country has yet reached the fast-track target MC prevalence of 90% in men 10–29 years.³⁶ Although the HIV prevalence pattern in the PHIA surveys was consistent with numerous cohort and controlled studies and demographic and health surveys showing the protective effect of MC on HIV acquisition among men, prevalence associations alone cannot exclude the possibility of reverse causality (ie, HIV-negative men self-selecting to undergo MC). The lower HIV incidence among medically circumcised men aged 15–34 years found in this study helps address this issue.

The incidence difference was nonsignificant for 35–59-year-olds. This could be a consequence of few recent cases occurring in the PHIAs resulting in large variance estimates, or the short amount of time elapsed between when MCs scale-up began and the PHIAs were conducted, as the lifelong benefits (and population-level effect) of this one-time intervention are expected to accrue with time.16 A risk concentrating phenomenon—wherein as a result of the HIV preventive benefit of MC, among all men at high risk for HIV, uncircumcised men differentially move out of the HIV-negative category more quickly than circumcised men, leaving a substantial risk discrepancy at older ages—could also be a factor. Spurious waning—wherein as a result of the partial preventive benefit of MC, among men at high risk for HIV, the uncircumcised ones become infected more quickly than the circumcised ones, leaving a substantial risk differential at older ages—could also bias this analysis.³⁷ True spontaneous biological mitigation of the protective effect of MC seems unlikely given the strong evidence for MC.^3–5 Offsetting of the protective effect by other factors (eg, herpes simplex virus infection and development of high-risk behaviors later in life) are other possible explanations. As discussed below, a lack of significance could also result in part from MC-type misclassification, which our findings suggest was more frequent in this age group.

HIV incidence in men with nonmedical MC was not different from incidence in uncircumcised men. This could be due to heterogeneity in the treatment effect, given this is a nonstandardized practice for amount of foreskin removed, unlike medical MC. It could also result from the observed higher prevalence of high-risk behaviors and/or differences in adoption of other HIV prevention and treatment interventions among these groups.

The findings of this study also suggest that relying on self-reported provider type to distinguish medical and nonmedical MC may be unreliable. Many men reporting medical MC said it occurred well before the VMMC program in their country was implemented. Although some amount of medical MC is likely to have been performed before the VMMC program, absence of widespread availability of medical MC was a prerequisite for implementing VMMC programs²⁶; thus, such a large proportion reporting medical MC before VMMC program scale-up is unlikely. This finding could result from a combination of recall bias for men circumcised many years before and social desirability response bias because of potentially stigmatizing characterization for some men of nonmedical MC (eg, “traditional”). Possible misclassification of MC status and type has implications for future PHIAs (as well as other surveys) that draw distinctions between self-reported medical and nonmedical MC, especially because these measures are used as benchmarks for progress of MC programming in respective countries. Including a question that asks about MC setting (ie, clinic versus “traditional”) and providing graphical depictions of partial versus complete MC could potentially improve differentiation of medical and nonmedical MC, although this, too, would need to be validated.

A unique strength of this analysis was inclusion of the impact of ART scale-up in the adjusted analysis given the efficacy of VLS in preventing HIV transmission. Although several observational studies have shown declining HIV incidence in the presence of ART and VMMC scale-up, they did not isolate the contribution of VMMC.^11–13 Our study indicates that VMMC scale-up in sub-Saharan African countries with low baseline MC prevalence is already yielding population health benefits.

There were several limitations to this study. First, there were few recent cases among participants in the PHIAs, which resulted in wide confidence intervals with lower precision on estimates. This potentially impacted our ability to detect incidence differences between groups of men by MC status. In addition, data had to be pooled across countries because of these few recent HIV cases. Thus, although the adjusted model controlled for country, important differences in the drivers of HIV infection between countries may have been unaccounted for. Because of the imbalance in national population sizes, almost 40% of the study population is Tanzanian, from Tanzania where HIV incidence is relatively low and circumcision status relatively high. Residual confounding by age also cannot be discounted as a contributor to the protective association of MC with HIV incidence. However, the few recent cases did not permit finer age disaggregation.

Misclassification of MC status (circumcised versus uncircumcised) and MC type (medical versus nonmedical) is possible with self-reported outcomes.²¹ In both instances, we expect that misclassification would be pre-dominantly one way, related to social desirability bias resulting from a perception of alignment between PHIA survey staff and the VMMC program (ie, both are PEPFAR-supported activities). Furthermore, the observed incidence differences between medically circumcised and uncircumcised men would not be expected if bidirectional misclassification dissociated true and self-reported MC status or MC type. Thus, misclassification from nonmedical to medical MC would likely bias our findings toward the null. No studies have assessed the validity of self-report for MC type (ie, medical or nonmedical MC), whereas this has been studied for MC status—with mixed results—elsewhere; some have found a high level of agreement between self-report and physical examination,^38–42 typically in areas where nonmedical MC is not a local cultural practice, whereas others have indicated substantial self-misclassification.^14,23,43,44 Importantly, Lissouba et al¹⁴ found that the protective association of MC (measured using an HIV recency test) was obscured when relying only on self-reported as opposed to clinically assessed MC status. In surveys as large as PHIAs, the feasibility of clinical examination for MC status is low. Including the timing of medical MC and geographic information about where VMMC scale-up occurred might further refine self-reported MC designations. However, such an approach in this analysis was not possible as a larger sample size would be needed to ensure sufficient values in all cells given the rarity of recent HIV infections in PHIAs.

This multicountry, nationally representative analysis found significantly lower HIV incidence among medically circumcised men aged 15–34 years compared with uncircumcised men. Additional rounds of PHIAs will provide the opportunity to analyze the trend in incidence to further explore the impact of the VMMC program, as well as permit more refined age groups than were possible in this study. Given that many VMMC program clients to date have been 10–14 years old, compounding benefits are expected over time as these boys who are mostly presexual age into higher incidence groups. The findings from this analysis are consistent with the expected ongoing VMMC program impact and highlight the independent importance of VMMC as part of the HIV response in sub-Saharan Africa, including in the context of ART scale-up.