The HIV epidemic in South Africa is among the most rapidly growing in the world. In 1990, the prevalence of HIV infection among women attending antenatal clinics was less than 1%. In 2000, antenatal prevalence had reached 24.5% nationwide, with provincial prevalences ranging from a low of 8.7% in the Western Cape to a high of 36.2% in KwaZulu-Natal . In South Africa, as in most sub-Saharan countries, a range of prevention tools has been available, including the promotion of condom use, safer sex and the treatment of sexually transmitted diseases (STD), but these have failed to control the epidemic.
A growing body of evidence suggests that male circumcision (MC) is associated with a reduced risk of HIV infection in Africa [2,3]. Although its precise role in reducing HIV transmission has not been determined and little is known about the biological mechanism by which it may protect against HIV infection, MC is currently being considered as an intervention . Randomized controlled trials are still needed to assess its potential impact in reducing HIV transmission . Should a protective effect be shown, it remains crucial to evaluate the acceptability of this surgery. The feasibility of such an intervention has been doubted, and an assessment of the acceptability of MC in different areas of sub-Saharan Africa is necessary . A qualitative study recently conducted in Kenya in a traditionally non-circumcising ethnic group  suggested that acceptability may be high and that the primary barriers to acceptance are cultural identification, a fear of pain and bleeding, and the cost.
The purpose of this study was to measure the prevalence and associated factors of MC in a South African township, and to assess its acceptability as a tool for HIV prevention.
The survey took place during 14 days in August and September 2001. It was designed to collect data about 500 men aged 19–29 years and 300 women aged 14–25 years living near Westonaria town in South Africa, in the townships of Bekkersdal, Mohlakeng and Poortje. Westonaria is approximately 60 km west of Johannesburg. Bekkersdal is adjacent to Westonaria town, and Mohlakeng is about 5 km to the north; Poortje is approximately 20 km south of Bekkersdal. Administratively, Bekkersdal is governed by the Westonaria Local Council, Mohlakeng by the Randfontein Local Council, and Poortje by the Lenasia South Local Council.
Households were selected by a two-stage random sampling technique. Index houses were randomly selected from a map obtained from the local municipal offices. All men aged 19–29 years and all women aged 14–25 years who slept in the selected households the night before the study team visit were eligible for inclusion in the study. If eligible participants were not at home, the study team visited two more times before abandoning the household. All eligible individuals provided written informed consent, after presentation and explanation of the study and the consent form, in the respondent's language. They were then transported to a local facility for the interviews, the collection of biological material and clinical examination. Results from blood serum were collected for the study of the epidemiology of HIV infection but are not analysed in the present paper. The HIV prevalence in our sample was 11% among the men and 30% among the women.
The questionnaire used in this study was based on a UNAIDS questionnaire  adapted to the local situation and to the age of the subjects in our sample. The interviewers completed the questionnaire during a private interview in the preferred language of the interviewee. To protect confidentiality and enhance the validity of the self-reports, interviews were conducted through an intercom system, with the interviewer and the interviewee in different rooms. Entrances and exits were designed to prevent them from meeting, either before or after the interview. We collected data about:
Household facilities; the number of resident and occasional members of the household; age; marital status and the number of children; educational level; ethnicity; religion; duration of residence in the area, and alcohol consumption.
Age at first sexual intercourse; the number of lifetime partners; condom use; non-consensual sex (and the perpetrator's characteristics); commercial sex; characteristics of the seven most recent non-spousal partners (ethnic group, age, marital status, duration of residence in the area, duration of relationship, the number of episodes of sexual intercourse, condom use and the circumcision status of partners).
Genital hygiene; genital hygiene before and after sexual intercourse; injections, transfusions and hospitalization histories; circumcision status including age at, place of, and reasons for circumcision.
Perceptions and beliefs about male circumcision
A previous qualitative study conducted in a nearby area enabled us to design standardized questions relating to MC . All main statements and beliefs recorded during the qualitative study were included as questions in the present survey; subjects were asked if they agreed or disagreed with each (binary format).
As part of the clinical examination, the clinical observers were asked to classify the male foreskin into one of the four categories proposed by Wynder and Licklider : (A) the foreskin completely covers the glans; (B) the foreskin covers one-half of the glans; (C) foreskin beyond sulcus but can be extended to cover one half of the glans without compressing it; and (D) foreskin completely absent.
The data from the questionnaires were entered into a database (Microsoft Access, Redmond, USA) twice, by different individuals. The two entries were compared and discrepancies were corrected. Extensive checking for inconsistencies followed. The files were then imported into the Statistical Package for Social Sciences (SPSS 11.0 for Windows, Chicago, USA) and prepared for statistical analysis.
Analysis and statistical methods
The proportion of men who reported being circumcised was computed among those aged 24–29 years, because a significant proportion of the younger men may still be circumcised in the future. Because of the censored nature of these data, we used Kaplan–Meier survival analysis to compute the median and interquartile ages at circumcision, with the approximation that no circumcisions occur after the age of 29 years.
We compared self-reports of circumcision and the foreskin classifications made during clinical examination. The analyses that follow are based on self-reports only, because questions related to MC were asked only of those who reported being circumcised, and because we remained unsure whether clinical foreskin classification is an appropriate method to ascertain circumcision status. We repeated the analyses according to foreskin classifications to ensure that the conclusions were the same, but do not present these results in detail.
A multivariate model was used to assess the factors associated with reported male circumcision. We considered for inclusion in the analysis variables relating to background (except age), sexual behaviour, and health behaviour (see description of questionnaire above). Because circumcision status was a censored variable, we fitted a Cox regression model for the sexually active male respondents. The time variable was age, the event circumcision, and the time of the event the reported age at circumcision. Data for men not circumcised at the time of the survey were censored. Adjusted proportional hazard ratios were used to estimate the relative risk of circumcision associated with each of these variables.
Finally, we tabulated the attitudes and beliefs about circumcision among men and women separately. We compared those of self-reported circumcised and non-circumcised men with a Chi-square test or Fisher's exact test when necessary. These comparisons were further restricted to those aged more than 23 years, to verify that the age structure of the sample did not influence our conclusions (detailed results not shown).
The Ethics Committee of the University of Pretoria, Pretoria Academic Hospital, approved this project on 26 April 2001 (Protocol number 97/2001).
The field workers were able to interview 78% (482/617) of the eligible male and 64% (302/472) of the eligible female occupants of the sample households.
The median age was 21 years for men [interquartile range (IQR) 20–24] and 19 for women (IQR 17–23). Of the 784 participants, 92 resided in Poortje, 328 in Bekkersdal, and 364 in Mohlakeng; 164 were Sotho, 304 Tswana, 191 Xhosa, 68 Zulu, and 48 participants came from other ethnic groups. More than half (54%) lived in squatter settlements. Birthplace distribution was as follows: 45% born in the study area; 21.6% in another urban area in South Africa; 28.6% in a rural area in South Africa; and 3.6% in another country. Only 27.8% of men and 40.1% of women were currently enrolled in a school. Few members of the sample were married at the time of the survey: 5.6% of the men and 16.9% of the women. The other respondents were either in a committed relationship (59.3%) or single (30.5%). Twenty-one percent of the men and 42.7% of the women had ever had a child. Most had only one child at the time of the survey. Half the sample (49.7%) reported that they were Christian, 26.4% were Muslim and 23.9% another religion, including traditional.
Male circumcision practices
Of the 482 men aged 19–29 years who participated in the survey, 108 reported that they were circumcised (22.4%). Because a significant proportion of the younger men will be circumcised after the survey, we restricted the analysis to the 143 men aged 24–29 years: 44 reported being circumcised (30.8%). Survival analysis indicated that the median age for circumcision was 17 years, and the interquartile range 16–18. This assumed that very few circumcisions are performed on men aged more than 29 years. None of the 108 circumcised men of our sample reported an age at circumcision above 26 years old.
These 108 self-reported circumcised men reported being circumcised in three different settings: two-thirds at initiation schools (64.8%), less than one-third at a hospital or a clinic (28.7%), and the rest by a general practitioner (6.5%). The variations in circumcision rate by ethnic group involved mainly the numbers of circumcisions in traditional settings: the proportion of those circumcised in clinical settings was relatively constant (7%) in the three main ethnic groups (Table 1).
Only 18 respondents reported being circumcised locally, and all in clinical settings. Overall, 62 men said they were circumcised in a rural area of South Africa, all but five in traditional settings. The reasons given for circumcision were tradition (63.6%), medical (15%), and partner request (13.1%).
Most of the circumcised men reported that the procedure was painful – very (42.6%) or mildly (34.3%). Only 18.5% said it was not painful. The adverse events most frequently reported were bleeding (reported by 26.3%), penile injury (6.3%), and local infections (4.2%). Reported pain and adverse outcomes were significantly less frequent when circumcision was performed in clinical settings. Pain was reported by 23 out of 38 respondents (60.5%) circumcised in a clinical setting compared with 60 of the 70 men circumcised in traditional settings (85.7%; P = 0.004, Chi-square test). Adverse outcomes were reported by eight of the 32 respondents (25%) circumcised in a clinical setting (six did not answer the question) compared with 30 of the 63 men (47.6%) circumcised in traditional settings (seven did not answer the question) (P = 0.03, chi-square test).
Of the 97 men who remembered, 93.8% reported a healing period of 4 weeks or less. The mean healing time was 3 weeks. A small but significant proportion of respondents (4.7%) reported having sex during this healing period. The median healing time did not differ according to the setting.
Self-reported circumcision status and clinical classification of foreskin
The 473 male participants who answered the question on circumcision status and who participated in the clinical examination had their foreskin classified during the clinical examination: 80% into type A, 3% into type B, 4% into type C and 13% into type D. During the interview, 12% of those in category A, 20% in B, 70% in C and 72% in D reported being circumcised. We compared clinical classifications A and D with self-reports. Only 14% of these answers were discordant (63/438). The kappa value for agreement between self-reports and clinical observations was moderate (0.494), but significantly higher than 0 (P < 10−4). Because some questions were asked only of those who reported being circumcised, the rest of the analysis is based on self-reported circumcision status. The same analysis was repeated with categories A and D to confirm the findings based on self-report.
Factors associated with male circumcision
We used a Cox regression model of sexually active male participants to seek associations between self-reported circumcision status and sociodemographic and behavioural variables (Table 2). We found a strong association between the reported number of lifetime partners and circumcision status. The proportion of circumcised men was significantly higher among those who reported more than four lifetime non-spousal partners than among those who reported between one and four (adjusted odds ratio 1.75). Duration in the area was also associated with circumcision status: the proportion of circumcised men decreased significantly with duration. The proportion of circumcised men who reported being born in the area was only 9.6%, compared with 50.9% among those in the area for less than one year.
Attitudes and beliefs about circumcision
A high proportion of non-circumcised men (50.8%) said that they would be circumcised in the future (Table 3). Moreover, 72.5% of the non-circumcised men reported they would want to be circumcised if MC were proved to be protective against HIV/STD. Similar proportions of circumcised men and non-circumcised men wanted their children circumcised in that case, but we noted that the pain caused by circumcision was considered to be bearable more often for adults than for children.
Only 9% of the circumcised men and 7% of the non-circumcised men reported that circumcised men do not need to use condoms, and 29 and 22%, respectively, reported that MC protects against AIDS and STD. Moreover, 30 and 18%, respectively, believed that circumcised men can safely have sex with many women; this difference was significant (P = 0.006).
Overall, 30% of respondents believed that infection or even death could result from circumcision, and this proportion was much higher when the question asked specifically about traditional circumcision.
Approximately half of the non-circumcised men agreed that MC proves manhood and earns respect from peers, and that most women prefer circumcised men. This was confirmed by the women's reports: 47% said that women prefer circumcised men. Moreover, 25% personally preferred sex with circumcised men, compared with 9% with non-circumcised men. Another 36% reported that it was all the same to them, and 30% said that they did not know.
Not surprisingly, circumcised men were significantly more likely than non-circumcised men to agree about the advantages of circumcision. Opinions about its drawbacks were not symmetrical; circumcised men were not significantly more likely to disagree about the drawbacks than non-circumcised men (MC decreases pleasure during sex, is old-fashioned and is expensive).
These comparisons were repeated for the sample older than 23 years. The significance of some associations disappeared, but the trends remained unchanged.
The prevalence of MC in the study area was relatively low (30.8% of men aged 24–29 years) compared with many populations in sub-Saharan Africa, where nearly 100% of the men may have been circumcised. The overall perception of MC was nonetheless positive. Adverse outcomes were considered more likely for traditional than for modern medical circumcision. A significant proportion of respondents felt that MC protected them from HIV: this was translated in turn into unsafe sex practices, with circumcised men more likely to report a high number of lifetime partners.
Although the vast majority of men in the study area are still not circumcised, there is some evidence that a preference for circumcision may be high. Our data suggest that the proportion of circumcised men will increase: 22% of the male sample is circumcised but more than half of those who are not circumcised reported that they will be. This would yield an overall rate of MC greater than 60%, compared with our rough estimate that 30% of the men will eventually be circumcised. Should the protective effect of MC in HIV transmission be demonstrated, the proportion of non-circumcised men who would like to be circumcised will reach 72.5%, with an overall circumcision rate of 78%.
These self-reports indicate that the acceptability of MC may be quite high. This is particularly surprising because the adverse effects of MC are well known. Adverse medical outcomes of MC, mainly haemorrhage and sepsis, have also been reported in several studies from sub-Saharan Africa [11–14].
MC may be performed in clinical settings by health professionals or by religious or traditional practitioners. Our data confirm the general feeling that MC is safer in a clinical setting (all reported MC performed in the area were performed in a clinical setting). Traditional circumcision may include practices such as using the same knife for each man during a circumcision ceremony, thereby increasing the risk of transmitting HIV. This confirms previous reports that complications appear most often when MC is performed in non-medical settings [15–17].
Our results suggest that the perception of safety is the principal issue in implementing a prevention method based on MC. A false sense of security accompanies circumcision, in view of the findings that a significant proportion of men and an even higher proportion of circumcised men said that circumcised men can safely have sex with many women. Even more worrisome, circumcised men were more likely to report a high number of lifetime partners. This indicates that this feeling of security is transformed into dangerous practices.
We found a moderate concordance, as measured by the kappa value (49%), between observed foreskin status and self-reported circumcision status. The proportion of discordant self-reports was 14%. Whereas the classification proposed by Wynder and Licklider  is objective, we do not think that it is an appropriate method for ascertaining MC status. The position of the prepuce on the glans may also depend on other factors such as slight erection or temperature. In future studies, it may be useful to look for a scar even if some circumcision practised in childhood leaves only a barely noticeable scar. Because of the discrepancies between self-reported circumcision status and clinical classification, we also assessed factors associated with MC and perception and attitudes towards MC for clinical classifications A and D; our results did not change significantly.
The association between MC and a higher number of lifetime partners may be partly explained by circumcisions for medical reasons, including STD. In this scenario, MC would be partly a consequence of a high number of partners. Only 15% of our respondents, however, said that they were circumcised for medical reasons. We re-computed our analysis excluding them without changing the results: both duration in the area (P < 10−3) and the number of lifetime partners (P = 0.02) remained significantly associated with circumcision status.
Before MC may be considered as a prevention tool for HIV infection in sub-Saharan Africa, similar investigations are needed in other geographical areas. In addition, our study did not provide information on a number of issues. The cost-effectiveness of such an intervention must still be assessed. Even if MC is effective in reducing HIV transmissibility, poorly implemented MC programmes are not likely to have an impact on the HIV epidemic. Messages provided during trials or interventions must be tested to achieve an appropriate balance between an invitation to this protocol and warning messages explaining that MC protects only partly (if at all), and that condom use is still needed. Finally, the impact of MC on the HIV epidemic at a population level must be modelled to optimize intervention efforts.
Our results are very optimistic on the potential feasibility of implementing a prevention programme that includes MC, but we strongly suggest that interventions including MC as a tool for preventing HIV transmission should carefully address the false sense of security that MC may provide.
The authors are indebted to Michel Caraël for his useful comments on a previous draft of the paper.
Sponsorship: Funding was received from the Agence Nationale de Recherche contre le SIDA (AO2000), Ensemble Contre le SIDA, and the Institut National de la Santé et de la Recherche Médicale.