There are several points with which we wish to take issue in the recent article by Van Howe, which reviews the scientific literature on male circumcision and risk for HIV infection. He concludes that male circumcision is associated with an increased risk of acquiring and transmitting HIV, which runs contrary to other published reviews[2,3]. There are two main aspects to the article, a geographic analysis and a review of patient studies.
In the geographic analysis, Van Howe begins by identifying three ecological studies from Africa which have reported that in areas where men are generally uncircumcised, there seem to be higher rates of HIV infection in the general population than in areas where circumcision is widely practised. He points out the limitations of ecological studies (which the authors of the studies have also explicitly acknowledged), then goes on to examine AIDS prevalence in first world countries with respect to male circumcision prevalence and finds the opposite result. What he shows essentially is that AIDS prevalence in the United States (where the majority of men are circumcised) is higher than in many European countries (where the majority of men are uncircumcised). However, the majority of AIDS cases in the United States are attributable to injection drug use or male with male sexual activity. Circumcision status is obviously irrelevant for HIV infection acquired through sharing contaminated needles. Some studies among gay men have shown a protective effect of male circumcision against HIV infection, but the reduction in risk has been relatively small compared to that associated with heterosexual transmission[4-6]. The most important risk factor for male to male sexual transmission of HIV is receptive anal intercourse, which is not related to the circumcision status of the receptive partner.
The most relevant ecological analysis is between male circumcision and rates of heterosexually acquired HIV infection, but it is difficult to estimate the extent of the latter in countries where there is significant transmission through other means. In Africa, the vast majority of HIV transmission is through heterosexual intercourse, and the findings of the African studies apply in that context.
Van Howe then conducts a "meta-analysis" of published studies, which are divided into those involving high risk patients (truck drivers, STD clinic attenders and tuberculosis patients), partner studies and population surveys. The exposure/outcome cells are then pooled; first stratified by the type of population studied and then altogether. The first point to make here is that Van Howe has pooled together data from cross-sectional, retrospective and prospective studies, studies involving men directly with those involving women who report on their male partner's circumcision status, and studies with widely varying exposure/risk levels (circumcision rates range from under 6% to over 96%). This approach is highly unsound methodologically. As has been observed in a state-of-the-art review of meta-analysis, "the comparison being made, between treatment groups, or between exposed and unexposed groups, is always within the original study. Simply pooling results over studies, ignoring study as a stratification factor, is not generally accepted as a valid technique for combining results" (Reference , p. 159). The erroneous conclusions derived from Van Howe's approach of indiscriminately pooling raw data become clear when looking more closely at the individual studies. All 15 presented which involve high risk populations clearly show that lack of male circumcision is associated with increased risk for HIV infection, with odds ratios ranging from 1.6 to 10.2, almost all statistically significant at the 0.05 level. Yet simply pooling the data from the studies yields a composite odds ratio of only 1.18. This is a classic case of Simpson's paradox, which is a type of confounding that can occur in epidemiological analyses when data from different strata with widely divergent exposure levels are combined, resulting in a combined measure of association that is not consistent with the results for each of the individual strata. The methodologically correct way of dealing with this situation in meta-analysis is through a stratified analysis, wherein a weighted average of the individual odds ratios is computed. The Mantel-Haenszel method is one widely accepted approach, which in this instance yields a summary odds ratio of 3.0 (95% confidence interval 2.6-3.4).
Another important issue concerns the results presented for the random population studies, which are at best misleading. Eleven studies are presented, five of which were reported together in the same paper by Urassa et al. In three of the studies, the data presented suggest statistically significant associations between male circumcision and increased risk for HIV infection[10-12] (sub-study 1 in Urassa et al.). However, Van Howe presents only the results of univariate analysis. Multivariate analysis was in fact performed in all three studies, and it resulted in no statistically significant associations between circumcision status and HIV infection; odds ratios went from 0.62, 0.53 and 0.61 to 0.8. 0.8 and 1.2 respectively. In another study, the odds ratio on univariate analysis was 0.94 and on multivariate analysis 1.5, P = 0.11, close to statistical significance. This is in the direction of lack of circumcision increasing HIV risk and might be referred to as a trend. The non-experimental nature of data from observational studies requires multivariate analysis to correctly estimate effect sizes, an issue recognized by the original study authors. It is critical in a review paper to report multivariate analysis results, and inappropriate and highly misleading to report only the unadjusted results of univariate analysis. In virtually all of the above papers, the authors comment that lack of circumcision is more common in rural areas, and circumcised men are more likely to be urban and mobile, factors which are associated with an increased risk of exposure to HIV. It is essential therefore to conduct multivariate analysis to adjust for such confounding.
Of the 11 population surveys referenced by Van Howe, on multivariate analysis, 3 showed a protective effect of circumcision on risk for HIV infection, 1 showed a trend towards an association, 7 showed no association and none showed a negative effect. The five studies in the paper by Urassa et al were conducted in the same region of northwestern Tanzania, and when the authors combined the results, the composite multivariate odds ratio was a statistically significant 1.5 in the direction of lack of circumcision increasing HIV risk. Simpson's paradox is not likely to have been as important in this analysis, as the exposure level (proportion of men circumcised) was not very different in the five surveys.
Of all 33 studies referenced by Van Howe in his meta-analysis, after multivariate analysis, 16 showed an association between lack of male circumcision and risk for HIV infection, 4 showed a trend towards an association, 12 showed no association (of which 4 were sub-studies of a larger study, as indicated above), and only one (a cross-sectional study from Rwanda) showed circumcision associated with an increased risk for HIV infection. This last study examined HIV infection in women, who were asked the circumcision status of their male partners. The authors comment that few male partners in the study (6%) were circumcised, and that these men may have constituted a population with high-risk sexual behaviour. The authors also state that it is likely that some men were circumcised as a result of treatment for venereal disease, which would have introduced additional bias.
Prospective study designs are the strongest ones for investigating the association between circumcision and HIV infection, and there are now eight reported studies which have examined the association in such a fashion[2,6]. Six of the eight have found statistically significant associations between lack of male circumcision and risk for HIV infection and the two others have found trends towards an association (relative risks of 3 and 3.5, P values of 0.11). The evidence that lack of male circumcision increases risk for HIV infection appears compelling, contrary to the contention in Van Howe's paper.
- Van Howe RS. Circumcision and HIV infection: review of the literature and meta-analysis. Int J STD AIDS 1999; 10: 8-16
- Moses S, Bailey RC, Ronald AR. Male circumcision: assessment of health benefits and risks. Sex Transm Inf 1998; 74: 368-73
- de Vincenzi I, Mertens T. Male circumcision: a role in HIV prevention? AIDS 1994; 8: 153-60
- Kreiss JK, Hopkins SG. The association between circumcision status and human immunodeficiency virus infection among homosexual men. J Infect Dis 1993; 168: 1404-8
- Whittington WL, Jacobs B, Lewis J, Lee F, Edwards T, Nahmias A. HIV-1 in patients with genital lesions attending a North American STD clinic: assessment of risk factors [Abstract No. T.A.P.118]. In Abstracts, V International Conference on AIDS, Montreal, Canada, June 4-9, 1989
- Buchbinder S, Heagerty P, Mayer K, et al. Risk factors for HIV seroconversion in a contemporary cohort of high risk men who have sex with men (MSM) [Abstract No. 23350]. In Conference Record, XII World AIDS Conference, Geneva, Switzerland, June 28-July 3, 1998
- Dickerson K, Berlin JA. Meta-analysis: state-of-the-science. Epidemiol Rev 1992; 14:154-76
- Rothman KJ. Modern Epidemiology. Boston: Little Brown and Company, 1996
- Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst 1959; 22: 719-48.
- Urassa M, Todd J, Boerma JT, Hayes R, Isingo R. Male circumcision and susceptibility to HIV infection among men in Tanzania. AIDS 1997; 11: 73-80
- Barongo LR, Borgdorff MW, Mosha FF, et al. The epidemiology of HIV-1 infection in urban areas, roadside settlements and rural villages in Mwanza Region, Tanzania. AIDS 1992; 6: 1521-8
- Grosskurth H, Mosha F, Todd J, et al. A community trial of the impact of improved sexually transmitted disease treatment on the HIV epidemic in rural Tanzania: 2. Baseline survey results. AIDS 1995; 9: 927-34
- Quigley M, Munguti K, Grosskurth H, et al. Sexual behaviour patterns and other risk factors for HIV infection in rural Tanzania: a case-control study. AIDS 1997; 11: 237-48
- Chao A, Bulterys M, Musanganire, F, et al. Risk factors associated with prevalent HIV-1 infection among pregnant women in Rwanda. Int J Epidemiol 1994; 23: 371-80