Helen A Weiss

Maria A Quigley

Richard J Hayes

Medical Research Council Tropical Epidemiology Group

London School of Hygiene & Tropical Medicine

Corresponding author (and address for reprint requests):

Helen A Weiss

Infectious Disease Epidemiology Unit

Department of Infectious and Tropical Diseases

London School of Hygiene & Tropical Medicine

Keppel Street

London

WC1E 7HT

UK

Running title: Meta-analysis of circumcision and HIV

3874 words excluding references and summary

Objective: To systematically review studies of male circumcision and risk of HIV-1 infection in men in sub-Saharan Africa, and to summarize the findings in a meta-analysis.

Design: Meta-analysis of observational studies

Methods: A systematic literature review of studies published up to April 1999 that included circumcision as a risk factor for HIV-1 infection among males in sub-Saharan Africa was carried out. A random-effects meta-analysis was used to calculate a pooled relative risk (RR) and 95% confidence interval (CI) for all studies combined, and stratified by type of study population. Further analyses were conducted among those studies which adjusted for potential confounding factors.

Results: 27 studies were included. Of these, 21 showed a reduced risk of HIV among circumcised men. HIV risk among circumcised men was around half that in uncircumcised men (crude RR=0.52, CI 0.40-0.68). In the 15 studies that adjusted for potential confounding factors, the association was even stronger (adjusted RR=0.42, CI 0.34-0.54). The association was stronger among men at high risk of HIV (crude RR 0.27; adjusted RR=0.29, CI 0.20-0.41) than among men in general populations (crude RR 0.93; adjusted RR=0.56, CI 0.44-0.70).

Conclusions: Male circumcision is associated with a significantly reduced risk of HIV infection among men in sub-Saharan Africa, particularly among men at high risk of HIV. These results suggest that consideration should be given to the acceptability and feasibility of providing safe services for male circumcision as an additional HIV prevention strategy in areas of Africa where men do not traditionally circumcise.

Keywords: HIV-1 infection, male circumcision, heterosexual transmission, meta-analysis, sub-Saharan Africa.

HIV prevalence in the general population in Africa varies widely both within and between countries. The magnitude of the variation seems to be only partly explained by different sexual behaviour patterns or factors known to influence HIV transmission, such as presence of sexually transmitted diseases (STDs) or condom use [1]. The hypothesis that male circumcision may reduce the risk of acquiring HIV infection was first suggested early in the HIV epidemic [2], and many epidemiological studies have since included circumcision as a potential risk factor in studies of HIV infection.

There is substantial evidence that circumcision is associated with a reduced risk of ulcerative STDs such as chancroid and syphilis [3]. As STDs, both ulcerative and non-ulcerative, are known to enhance risk of acquiring and transmitting HIV [4], it is likely that circumcision has an indirect effect on HIV infection. Circumcision may also protect against HIV directly, as viral entry may occur through micro-traumatic lesions or mini-ulcerations of the foreskin [5] or through trauma to the non-keratinized inner mucosal surface of the foreskin [6]. Further, the foreskin contains a higher density of Langerhans cells than the urethra or rectum [7], and these cells may be primary target cells for HIV transmission [8]. Genital ulcers may also be less easily recognised in uncircumcised men, thus delaying treatment and increasing susceptibility to HIV [6].

Male circumcision is practiced in many parts of Africa, but there is wide regional variation [5, 9]. In particular, there are many ethnic groups in eastern and southern Africa in which male circumcision is not traditionally practiced. This area of non-circumcision covers most of Uganda, parts of western Kenya, western Tanzania, north-eastern Democratic Republic of the Congo, virtually all of Rwanda, Burundi, Zambia, Malawi and Zimbabwe, and parts of Botswana, Namibia, Mozambique and South Africa. Circumcision is almost universally practised in most parts of West Africa, including Nigeria. There is, however, a large area of traditional non-circumcision covering central/eastern Ivory Coast and western/central Ghana.

Within Africa, there is a broad correlation between areas where there is little circumcision, and those with high HIV rates [5, 9]. This ecological association means little however, without looking at individuals within populations, and taking into account other factors associated with circumcision status. There may be other factors associated with HIV risk (such as sexual mixing patterns, or presence of other STDs) which are less prevalent among circumcising than non-circumcising populations.

There have been four previous reviews of the literature on HIV and male circumcision [3, 10-12]. The present review differs in being restricted to female-male transmission in sub-Saharan Africa, where heterosexual transmission is the predominant mode of transmission. It thus addresses the specific role of male circumcision in the heterosexual acquisition of HIV in Africa. Only one previous meta-analysis of the role of circumcision in HIV transmission has been performed [12], and this had several statistical and epidemiological limitations [13]. A re-analysis of the studies included in that report found that uncircumcised men were at a significantly increased risk of HIV infection, but concluded that a systematic review of the literature was urgently needed to minimize bias, and to clarify the issue [14].

Materials and Methods

The aim of our literature search was to identify all published studies of risk factors for HIV-1 infection among men in sub-Saharan Africa that included circumcision as a potential risk factor. The Medline, Pre-Medline, HealthStar and Popline databases were searched for papers published up to April 1999 which included ‘circumcision’ and ‘HIV’ as keywords or text in the abstract. 22 original research papers were identified in this way, 20 of which appeared in Medline. A further Medline search was carried out using keywords to search for all published studies of HIV risk factors in men in sub-Saharan Africa using the search condition (‘HIV-infections (epidemiology, etiology, transmission)’ OR ‘HIV-seroprevalence’ OR ‘HIV-seropositivity (epidemiology, transmission)’) AND ‘sub-Saharan Africa’ AND (‘risk factors’ or ‘odds ratios’ or ‘risk’)). 397 articles were identified in this way, including all 20 papers identified in the initial search. Of these, the abstracts of 59 papers referred to risk factors for female-male transmission of HIV, and these papers were manually scanned for references to circumcision, yielding a further 7 eligible papers. Finally, the reference lists of all 29 papers were checked, and 9 further potential papers were identified. However, none of these included circumcision as a risk factor. We did not include papers that used a proxy for circumcision, such as Muslim religion. Conference abstracts were also excluded.

Of the 29 eligible papers, two were studies of risk factors for HIV-2 only, and were excluded from the analysis [15, 16]. A further three studies contained insufficient details to calculate a crude relative risk and were excluded from the meta-analysis [17-19]. These studies were relatively small (study sizes of 125, 81 and 63 men respectively) and thus unlikely to have influenced the meta-analysis substantially. One study [20] did not report a crude relative risk, but did report an adjusted relative risk, and was included in the adjusted meta-analysis.

Several studies from Mwanza, Tanzania appeared in more than one publication. To avoid duplication of data in the meta-analysis, study 1 and study 5 from the Urassa paper [21] were excluded because they were included elsewhere [22, 23]. A study by Barongo et al. [24] was a subset of Urassa study 4, and was excluded. Another eligible paper [25] was not included as the study population formed the basis for a case-control study which allowed adjustment for confounding variables [22].

Three papers included stratified analyses of different population groups, and these strata were included as separate studies in the meta-analysis to reduce confounding [23, 26, 27].

It is plausible that the effect of circumcision varies according to background prevalence of HIV and of co-factors such as ulcerative STDs. Three broad groups of studies were identified - population-based studies [21-23, 26, 28-31]; studies of men at high risk of HIV [20, 27, 32-41]; and other studies (of factory workers and volunteers) [21, 42, 43]. Sub-group analyses were carried out on the population-based and high-risk studies.

All types of study design (cohort, case-control and cross-sectional studies) were included in the main analysis, and analyses were also carried out for cross-sectional studies alone. In case-control and cross-sectional studies, the odds ratio was used as an estimate of the risk ratio. The odds ratio was used in cross-sectional studies because it is not possible to obtain an adjusted risk ratio from published data as these were calculated using logistic regression.

However, when prevalence of HIV in the population is greater than about 20-25%, the odds ratio will be more extreme than the risk ratio. The sensitivity of the meta-analysis to the measure of effect was assessed by re-analysing the data for cross-sectional studies using the risk ratio. Statistical and graphical analyses were performed using Stata 6 [44]. A random effects meta-analysis was used to calculate a pooled relative risk (RR) [45, 46]. This model assumes a different underlying effect for each study, and heterogeneity between studies was also assessed [46]. Publication bias was assessed with a funnel plot and Begg’s Test for correlation between the effect estimates and their variances [47, 48].

The studies included in the review are shown in Table 1, grouped by type of population. The review incorporated 28 studies (19 cross-sectional, five case-control, three cohort and one partner study) reported in 22 published papers. The studies covered a range of population types, including general populations (11 studies), STD clinic attenders (8 studies), truck drivers (3 studies), factory workers (2 studies), and TB patients, hospital patients, volunteers and married couples (one study of each).

Crude relative risks were reported in 27 studies (see Table 1; Figure 1a). One study [20] reported an adjusted RR but not a crude RR. In 21 of the studies, circumcised men were at lower risk of HIV than uncircumcised men, and the association was statistically significant (p<0.05) in 14 of these. Of the six studies with a positive association between circumcision and risk of HIV, four were from Mwanza [21-23], and none found a statistically significant association. Overall, circumcision was associated with a highly significant reduction in HIV risk (pooled RR=0.52; CI 0.40-0.68; Figure 1a; Table 2). There was significant between-study heterogeneity (p<0.001).

Some adjustment for confounders was reported for 15 studies. Most studies adjusted for age and one or more factors from the following: socio-demographic (marital status, area of residence, ethnic group), sexual behaviour (number of sexual partners in lifetime, last year or last four months, contact with sex workers), and factors associated with risk of transmission (condom use, presence of ulcerative STDs). Details are given in the footnote to Table 1. The study by Barongo [23] was stratified by area of residence for the crude analysis, but residence was included as a confounder in the multivariate analysis, and hence this study contributes only once to the meta-analysis of adjusted effects. In addition, one study reported no significant association after adjustment for confounders but no details were given [39] and this study was not included in the adjusted meta-analysis.

All the 15 studies reporting adjusted relative risks found a protective effect of circumcision on HIV risk (RRs ranging from 0.12 to 0.8), and in 10 studies the effect was statistically significant (Table 1; Figure 1b). In general, the effect of adjustment was to strengthen the association between circumcision and reduced risk of HIV, with the adjusted RR being smaller than or equal to the crude RR in 10 of the 14 studies which reported both measures. For two studies from Mwanza [21, 22], the adjusted RR showed a decreased risk of HIV among circumcised men although the crude RR had shown a non-significant increased risk of HIV associated with circumcision. The four studies where the adjusted RR was closer to one than the crude RR were the studies from the Ivory Coast [32, 36] and of STD clinic attenders in Nairobi [38, 40]. In each of these studies, the difference between the crude and adjusted RRs was small (Table 1).

The adjusted analysis showed a slightly stronger effect than the crude analysis (pooled adjusted RR=0.42; CI 0.34-0.54; Table 2). For the 14 studies reporting both crude and adjusted RRs, the crude RR was 0.54 (CI 0.39-0.74) and the adjusted RR was 0.45 (CI 0.34-0.58).

Analyses were then stratified by type of population. The crude analysis showed little evidence of an association between circumcision and HIV in population-based studies (crude RR=0.93, CI 0.71-1.21). Of these 12 studies, six adjusted for confounders. The crude RR was reported in five of the six studies and was similar to that for all population-based studies (RR=0.90, CI 0.59-1.36). The adjusted RR showed a significantly decreased risk of HIV among circumcised men (adjusted RR=0.56, CI 0.44-0.70, based on six studies; Table 2, Figure 2a). There was no significant heterogeneity between these studies (p=0.21). When analyses were restricted to cross-sectional studies alone, results were very similar (crude RR=0.91, CI 0.63-1.32; adjusted RR=0.55, CI 0.42-0.72; Table 2).

Among the 12 studies of men at higher risk of HIV (STD clinic attenders, truck drivers, TB patients and hospital patients), the crude RR was 0.27 (CI 0.22-0.33). Seven of these studies included adjustment for confounders, which had little effect on the effect estimate (adjusted RR=0.29, CI 0.20-0.41; Table 2, Figure 2b). All adjusted RRs were 0.5 or less, but there was significant between-study heterogeneity (p=0.03). Results for cross-sectional studies alone were similar (crude RR=0.24, CI 0.20-0.29; adjusted RR=0.24, CI 0.18-0.31; p-value for heterogeneity=0.49; Table 2).

There was wide variation in study size, with the largest study [28] consisting of 6821 men (25% of the total subjects included in the meta-analysis), whereas the second largest study consisted of 2603 men [21]. To examine the influence of the largest study on the meta-analysis, we re-analysed the data excluding this study. The resulting relative risks were similar (crude RR=0.52, CI 0.40-0.68; adjusted RR=0.42, CI 0.32-0.55), indicating that this study did not influence the meta-analysis unduly.

Funnel graphs of the data are presented in Figure 3a and 3b. There is no evidence that smaller studies (i.e. those with larger standard error) were more likely to report a positive association, indicating no evidence of publication bias (Begg’s rank correlation test: p-value=0.56 for the crude analysis; p=0.24 for the adjusted analysis).

The effect of using the odds ratio instead of the risk ratio was assessed by re-analysing the cross-sectional studies using the risk ratio. For these 18 studies, the crude odds ratio was 0.51 (CI 0.37-0.69) and the crude risk ratio was 0.59 (CI 0.47-0.73). In the high risk studies, where HIV prevalence was higher, the crude odds ratio was 0.24 (CI 0.20-0.29), and the crude risk ratio was 0.37 (CI 0.33-0.41).

This systematic review and meta-analysis provide compelling evidence that male circumcision is associated with a reduced risk of HIV infection in sub-Saharan Africa. The strongest association is seen among men at high risk of HIV, but circumcised men in the general population are also at significantly reduced risk after adjusting for potential confounding factors. This supports results from the previous meta-analysis [14], which included 33 studies from around the world, and found an overall crude odds ratio of 0.60 (CI 0.45-0.80), with a stronger effect in high risk groups (crude OR=0.33, CI 0.26-0.43) than in general population groups (crude OR=1.04, CI 0.77-1.41). This was a re-analysis of previously collected data and had several major limitations (no systematic literature review, lack of adjustment for confounding factors, and lack of exploration of heterogeneity), which have been addressed in the present analysis.

Overall, there was significant heterogeneity between studies, indicating that the magnitude of the protective effect varies between populations. We would, however, expect any protective effect of circumcision to be heterogeneous across different populations due to differences in a number of factors, including variations in sexual exposure to HIV, timing of circumcision, but particularly differences in the prevalence of STDs. There is evidence that male circumcision protects against some ulcerative STDs (particularly chancroid and syphilis) [3], which in turn enhance HIV transmission, and this could explain the greater protection afforded by circumcision in populations where genital ulcer disease (GUD) is more prevalent. Indeed, several studies adjusted specifically for prior or current GUD, and as GUD is likely to act on the causal pathway, the effect of circumcision on HIV may be stronger than that estimated by an adjusted RR.

When the analysis was stratified by type of population, heterogeneity was reduced substantially and was not significant in the population-based studies. There was still significant heterogeneity among the seven high risk populations, and the lack of reliable and consistent data from these studies on other potential effect-modifiers (such as penile hygiene) prevented further exploration of this heterogeneity. However, the studies all found a strong protective effect (RRs from 0.12 to 0.50), and while this is a wide range, it suggests that a substantial protective effect was a consistent finding in all high risk populations.

The main limitation of observational studies is that the effect of circumcision on HIV infection may be confounded by factors which are associated with HIV risk, and which may differ between circumcised and uncircumcised men. It is therefore important to note that adjustment for confounding strengthened the association in all the population-based studies, and generally made little difference in the high risk studies. This is not surprising, as most studies which have reported behavioural patterns [21, 27, 28, 30, 33, 34, 42, 49] have found that circumcised men tend to report higher risk behaviours than uncircumcised men in the same population. This would tend to result in higher HIV prevalence among circumcised men if there were no protective effect of circumcision. It remains a possibility, however, that there is residual confounding due to behavioural or biological factors that are unknown or poorly measured in some studies.

Religion and ethnicity are potentially important confounding factors in these observational studies, as they are likely to be associated with both circumcision and HIV risk. Three studies explored this by stratifying analyses by religion. Among Muslims in Kigali, Rwanda [42], circumcision was associated with a protective effect (crude RR=0.18, CI 0.02-1.20). Little association was seen among Christians (crude RR=0.79, CI 0.50-1.23) although this could be because, unlike Muslims, most Christians were circumcised post-puberty. A study in Rakai, Uganda [28], where 98.5% of Muslim men were circumcised, found that among non-Muslim men, those circumcised before age 12 had a lower risk of HIV compared with uncircumcised men (crude RR=0.62, CI 0.29-1.28). Further, in Mwanza, Tanzania [22], the association between circumcision and HIV was similar when results were re-analysed excluding Muslims. Hence the evidence suggests that circumcision has an effect on HIV infection independently of the confounding effect of religion.

It has been suggested that prepubertal circumcision (performed for religious or traditional reasons) is more likely to protect against STDs and HIV, whereas postpubertal circumcision is likely to be performed as a consequence of infections, such as balanitis [28]. Only two studies included in the meta-analysis asked about age at circumcision [22, 28]. In Rakai [28], the strongest effect of circumcision on HIV risk was seen among those circumcised below age 12 (adjusted RR=0.39, CI 0.29-0.53). There was also a significantly reduced risk of HIV associated with circumcision between ages 13-20 (adjusted RR=0.46, CI 0.28-0.77), and a non-significant effect above age 20 (adjusted RR=0.78, CI 0.43-1.43). In contrast, among men in rural Mwanza [22], circumcision before age 15 years was associated with an increased risk of HIV (adjusted RR=1.50, CI 0.57-3.90), while circumcision after age 15 was associated with a lower HIV risk (adjusted RR=0.37, CI 0.18-0.74). The reasons for these discrepant findings are unclear, and further work on the effect of age at circumcision on HIV susceptibility is needed.

The observational studies included in this meta-analysis cannot definitively establish a causal role for circumcision in protecting against HIV infection. It is biologically plausible that the foreskin may enhance HIV transmission both directly and indirectly [50] although there is little direct evidence for this. Causality also implies that circumcision occurs prior to infection with HIV, and we cannot be certain of this with retrospective data, although the studies which report age at circumcision [21, 22, 28, 42] indicate that it usually occurs by age 20 and hence is likely to precede HIV infection.

The present analysis was restricted to studies of HIV infection in men. Hence we were able to assess only the effect of circumcision on male susceptibility, and not on male infectiousness. However, the advantages of this restriction are considerable. Firstly, an effect of male circumcision on susceptibility is arguably more plausible on biological grounds than an effect on infectiousness. Secondly, only the circumcision status of the study subject is needed, and not the status of partners, thus reducing potential misclassification of exposure. Thirdly, a study of the effect of male circumcision on male-female transmission is complex, as a woman may have had more than one male sexual partner, with differing circumcision status. However, recent data from Rakai have shown that among discordant couples in which the male partner was HIV positive, there was a non-significant reduction in transmission rate if the man was circumcised (rate ratio=0.44, CI 0.15-1.32), and this reduction in risk was statistically significant among couples with male HIV viral loads < 50,000 copies per mL [51].

Circumcision was determined by self-report in 16 studies, and by clinical examination in eight studies. For four further studies [35-38] the method of ascertainment was not clear. Validity of self-reported circumcision was assessed among factory workers in Mwanza [21]. Of the 111 men who had reported they were circumcised, only 69% were found to be so on examination. Among the 91 men who reported themselves as uncircumcised, 94% were found to be uncircumcised on examination. This suggests that self-reported circumcision may over-estimate the proportion of men circumcised in this population, and suggests that genital examination should be carried out whenever possible in future studies. Assuming this misclassification is non-differential with respect to HIV status, the effect would be to under-estimate the association between lack of circumcision and HIV. All but one study performed HIV tests by one or two ELISA tests, with confirmatory Western blot for discrepant or indeterminate ELISA results. The remaining study [37] performed only one ELISA with no confirmatory testing. Overall, therefore, there is likely to be little misclassification of HIV results.

Meta-analyses are vulnerable to bias, due to failure to identify all eligible articles and also because research yielding statistically significant results is more likely to be submitted and published. Therefore, even if all published studies have been identified, these may be only a subset of the studies actually carried out. However, this seems unlikely to cause substantial bias in the present review because studies of risk factors for HIV almost always examine a range of behavioural and biological risk factors, and it is unlikely that the published studies are biased in terms of finding circumcision as a risk factor. However, it is possible that if no association with circumcision was found, this was sometimes not mentioned in the paper. The results of the funnel plot suggest that publication bias was not a problem in this meta-analysis, although the existence of such bias cannot be excluded. It was decided to exclude conference abstracts from the review as it is probable that these are more likely to report significant than non-significant findings, and hence the inclusion of abstracts might increase bias.

The 28 studies reviewed were from just eight countries, including nine studies from Kenya (seven of these from Nairobi) and seven from Mwanza Region, Tanzania. As expected, there were few studies from West Africa where circumcision in most areas is almost universal, although 39% of men in the study in Ziguinchor region, SW Senegal were uncircumcised [29].

To our knowledge, there have been two further studies on the effect of circumcision on HIV infection published since our literature review. A cohort study of trucking company employees in Kenya [52] found a very similar effect to our summary measure for high risk populations (adjusted RR = 0.25, CI 0.1-0.5). A recently published study of HIV discordant couples from Rakai, Uganda [53] found that the rate of female-male transmission was significantly higher among couples with uncircumcised male partners compared with couples with a circumcised male partner (rates of 16.7 per 100 person-years compared with 0 per 100 person-years; p<0.001).

In conclusion, the data from observational studies provide compelling evidence of a substantial protective effect of male circumcision against HIV infection in sub-Saharan Africa, especially in populations at high risk of HIV/STDs. The continuing rapid spread of HIV infection, especially in eastern and southern Africa, suggests that the potential public health benefit of introducing safe services for male circumcision on a wider scale should be explored. However, there are many concerns around such an introduction, including the possibility that men may increase their risky sexual behaviour if they think circumcision confers a high degree of protection, as well as the risk of bleeding or infection, cost, and issues of cultural identity. Studies are therefore needed to examine the acceptability, feasibility and safety of introducing male circumcision as an HIV/STD prevention strategy in high prevalence areas where men do not traditionally circumcise. Results from one such study among the Luo people in Nyanza province, western Kenya, who are traditionally non-circumcising, showed that 60% of men would prefer to be circumcised, and 62% of women would prefer a circumcised partner [54]. In areas where male circumcision is acceptable to the local community, randomised controlled trials of male circumcision performed by trained health workers as part of a package intervention incorporating safe sex education are needed. Such trials would overcome the inherent limitations of observational studies, and allow for reliable empirical evidence on the overall impact of introduction of male circumcision on HIV incidence.

We thank UNAIDS for funding this study, Michel CaraNl and Stephen Sharp for advice, and Simin Bahrainipur for administrative assistance. Helen Weiss and Maria Quigley are funded by the UK Medical Research Council.

1. Buve A for the Study Group on Heterogeneity of HIV Epidemics in African Cities: HIV/AIDS in Africa: Why So Severe, Why So Heterogeneous? 7th Conference on Retroviruses and Opportunistic Infections, San Francisco, CA, USA, February 2000 [Abstract S28].

2. Fink AJ: Circumcision and heterosexual transmission of HIV infection to men [letter]. New Engl J Med 1987, 316(24):1546-7.

3. Moses S BR, Ronald AR. Male circumcision: assessment of health benefits and risk. Sex Trans Inf 1998, 74:368-373.

4. Fleming ST, Wasserheit JN: From epidemiological synergy to public health policy and practice: the contribution of other sexually transmitted diseases to sexual transmission of HIV infection. Sex Trans Inf 1999, 75:13-17.

5. Moses S, Bradley JE, Nagelkerke NJ, Ronald AR, Ndinya Achola JO, Plummer FA: Geographical patterns of male circumcision practices in Africa: association with HIV seroprevalence. Int J Epidemiol 1990, 19(3):693-7.

6. Aral SO, Holmes KK: Social and behavioral determinants of the epidemiology of STDs: Industrialized and developing countries. In Sexually Transmitted Diseases. Edited by Holmes KK, Mardh PA, Sparling PF et al. McGraw-Hill; 1999:

7. Hussain LA, Lehner T: Comparative investigation of Langerhans' cells and potential receptors for HIV in oral, genitourinary and rectal epithelia. Immunology 1995, 85(3):475-84.

8. Soto-Ramirez LE, Renjifo B, McLane MF et al: HIV-1 Langerhans' cell tropism associated with heterosexual transmission of HIV. Science 1996, 271(5253):1291-3.

9. Bongaarts J, Reining P, Way P, Conant F: The relationship between male circumcision and HIV infection in African populations. AIDS 1989, 3(6):373-7.

10. Moses S, Plummer FA, Bradley JE, Ndinya Achola JO, Nagelkerke NJ, Ronald AR. The association between lack of male circumcision and risk for HIV infection: a review of the epidemiological data: Sex Trans Dis 1994, 21(4):201-10.

11. de Vincenzi I, Mertens T: Male circumcision: a role in HIV prevention? AIDS 1994, 8(2):153-60.

12. Van Howe RS: Circumcision and HIV infection: review of the literature and meta-analysis. Int J STD AIDS 1999, 10(1):8-16.

13. Moses S, Nagelkerke NJD, Blanchard J: Analysis of the scientific literature on male circumcision and risk for HIV infection. Int J STD AIDS 1999, 10: 626-628.

14. O'Farrell N, Egger M: Circumcision in men and the prevention of HIV infection: a 'meta-analysis' revisited. Int J STD AIDS 2000, 11(3):137-42.

15. Pepin J, Quigley M, Todd J et al: Association between HIV-2 infection and genital ulcer diseases among male sexually transmitted disease patients in The Gambia. AIDS 1992, 6(5):489-93.

16. Harrison LH, da Silva AP, Gayle HD et al: Risk factors for HIV-2 infection in Guinea-Bissau. J Acquir Immune Defic Syndr 1991, 4(11):1155-60.

17. Hudson CP, Hennis AJ, Kataaha P et al: Risk factors for the spread of AIDS in rural Africa evidence from a comparative seroepidemiological survey of AIDS, hepatitis B and syphilis in southwestern Uganda. AIDS 1988, 2(4):255-60.

18. Gomo E, Chibatamoto PP, Chandiwana SK, Sabeta CT: Risk factors for HIV infection in a rural cohort in Zimbabwe: a pilot study. Cent-Afr-J-Med 1997, 43(12):350-4.

19. Kane F, Alary M, Ndoye I et al: Temporary expatriation is related to HIV-1 infection in rural Senegal. AIDS 1993, 7(9):1261-5.

20. Mbugua GG, Muthami LN, Mutura CW et al: Epidemiology of HIV infection among long distance truck drivers in Kenya. East-Afr-Med-J 1995, 72(8):515-8.

21. Urassa M, Todd J, Boerma JT, Hayes R, Isingo R: Male circumcision and susceptibility to HIV infection among men in Tanzania [corrected and republished article originally printed in AIDS 1997 Jan;11(1)73-9]. AIDS 1997, 11(3):73-80.

22. 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(2):237-48.

23. 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(12):1521-8.

24. Barongo LR, Borgdorff MW, Newell JN et al: Intake of a cohort study of urban factory workers in northwest Tanzania. Risk factors for HIV-1 infection. Trop Geogr Med 1994, 46(3):157-62.

25. 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(8):927-34.

26. Wawer MJ, Sewankambo NK, Serwadda D et al: Control of sexually transmitted diseases for AIDS prevention in Uganda: a randomised community trial. Rakai Project Study Group. Lancet 1999, 353(9152):525-35.

27. Nasio JM, Nagelkerke NJ, Mwatha A, Moses S, Ndinya Achola JO, Plummer FA: Genital ulcer disease among STD clinic attenders in Nairobi: association with HIV-1 and circumcision status. Int J STD AIDS 1996, 7(6):410-4.

28. Kelly R, Kiwanuka N, Serwadda D et al: Age of male circumcision and risk of prevalent HIV infection in rural Uganda. AIDS 1999, 13(3):399-405.

29. Pison G, Le Guenno B, Lagarde E, Enel C, Seck C: Seasonal migration a risk factor for HIV infection in rural Senegal. J Acquir Immune Defic Syndr 1993, 6(2):196-200.

30. Serwadda D, Wawer MJ, Musgrave SD, Sewankambo NK, Kaplan JE, Gray RH: HIV risk factors in three geographic strata of rural Rakai District, Uganda. AIDS 1992, 6(9):983-9.

31. Carael M, Van de Perre PH, Lepage PH et al: Human immunodeficiency virus transmission among heterosexual couples in Central Africa. AIDS 1988, 2(3):201-5.

32. Sassan-Morokro M, Greenberg AE, Coulibaly IM et al: High rates of sexual contact with female sex workers, sexually transmitted diseases, and condom neglect among HIV-infected and uninfected men with tuberculosis in Abidjan, Cote d'Ivoire. J Acquir Immune Defic Syndr Hum Retrovirol 1996, 11(2):183-7.

33. Tyndall MW, Ronald AR, Agoki E et al: Increased risk of infection with human immunodeficiency virus type 1 among uncircumcised men presenting with genital ulcer disease in Kenya. Clin Infect Dis 1996, 23(3):449-53.

34. Bwayo J, Plummer F, Omari M et al: Human immunodeficiency virus infection in long-distance truck drivers in east Africa. Arch Intern Med 1994, 154(12):1391-6.

35. Gilks CF, Otieno LS, Brindle RJ et al: The presentation and outcome of HIV-related disease in Nairobi. Q J Med 1992, 82(297):25-32.

36. Diallo MO, Ackah AN, Lafontaine MF et al: HIV-1 and HIV-2 infections in men attending sexually transmitted disease clinics in Abidjan, Cote d'Ivoire. AIDS 1992, 6(6):581-5.

37. Hira SK, Kamanga J, Macuacua R, Mwansa N, Cruess DF, Perine PL: Genital ulcers and male circumcision as risk factors for acquiring HIV-1 in Zambia [letter]. J Infect Dis 1990, 161(3):584-5.

38. Cameron DW, Simonsen JN, D'Costa LJ et al: Female to male transmission of human immunodeficiency virus type 1: risk factors for seroconversion in men. Lancet 1989, 2(8660):403-7.

39. Greenblatt RM, Lukehart SA, Plummer FA et al: Genital ulceration as a risk factor for human immunodeficiency virus infection. AIDS 1988, 2(1):47-50.

40. Simonsen JN, Cameron DW, Gakinya MN et al: Human immunodeficiency virus infection among men with sexually transmitted diseases. Experience from a center in Africa. N Engl J Med 1988, 319(5):274-8.

41. Lankoande S, Meda N, Sangare L et al: HIV infection in truck drivers in Burkina Faso a seroprevalence survey. Med Trop Mars 1998, 58(1):41-6.

42. Seed J, Allen S, Mertens T et al: Male circumcision, sexually transmitted disease, and risk of HIV. J Acquir Immune Defic Syndr Hum Retrovirol 1995, 8(1):83-90.

43. Van de Perre P, Le Polain B, Carael M, Nzaramba D, Zissis G, Butzler JP: HIV antibodies in a remote rural area in Rwanda, Central Africa: an analysis of potential risk factors for HIV seropositivity. AIDS 1987, 1(4):213-5.

44. StataCorp. Stata Statistical Software: Release 6.0 1999: College Station, TX: Stata Corporation.

45. Sharp S, Sterne J. sbe16: Meta-analysis. Stata Technical Bulletin 1997; 38:9-14.

46. DerSimonian R, Laird N: Meta-analysis in clinical trials. Controlled Clin Trials 1986, 7(3):177-88.

47. Egger M, Smith GD, Phillips AN: Meta-analysis: principles and procedures. BMJ 1997, 315(7121):1533-7.

48. Begg CB, Mazumdar M: Operating characteristics of a rank correlation test for publication bias. Biometrics 1994, 50(4):1088-101.

49. Bailey RC, Neema S, Othieno R: Sexual behaviours and other HIV risk factors in circumcised and uncircumcised men in Uganda. J Acquir Immune Defic Syndr Hum Retrovirol 1999, 22(3):294-301.

50. Halperin DT, Bailey RC: Male circumcision and HIV infection 10 years and counting. Lancet 1999, 354:1813-15.

51. Gray R, Wawer MJ, Sewakambo et al. HIV incidence associated with male circumcision in a population-based cohort, and HIV acquisition/transmission associated with circumcision and viral load in discordant couples: Rakai, Uganda. XIII International AIDS Conference, Durban, S. Africa, July 2000; [Abstract MoOrC193].

52. Lavreys L, Rakwar JP, Thompson ML et al: Effect of circumcision on incidence of HIV-1 and other sexually transmitted diseases: A prospective cohort study of trucking company employees in Kenya. J Infect Dis 1999, 180:330-336.

53. Quinn TC, Wawer MJ, Sewankambo N et al: Viral Load and Heterosexual Transmission of Human Immunodeficiency Virus Type 1. N Engl J Med 2000, 342(13):921-929.

54. Bailey RC, Muga R Poulussen R: Trial intervention introducing male circumcision to reduce HIV/STD infections in Nyanza province, Kenya: Baseline results. XIIIth International AIDS conference, Durban, S. Africa, July 2000 [Abstract MoOrC196].