AIDS has now been documented as the leading cause of adult death in Côte d'Ivoire, Zaïre and Uganda , and HIV-related deaths have substantially increased mortality rates in many parts of sub-Saharan Africa. The HIV epidemic in sub-Saharan Africa is spread predominantly through heterosexual contact . Groups that are at particularly high risk of HIV infection have been identified as commercial sex workers, those attending sexually transmitted disease (STD) clinics, and truck drivers .
The transmission dynamics of HIV infection outside these high-risk groups are not clear. Studies that have been conducted in rural African populations have identified sociodemographic risk factors for HIV infection [4–12]. This information is useful for the planning of HIV intervention programmes, but additional information on sexual attitudes and behaviour is needed to design appropriate educational campaigns. Most previous studies [4–8,10–12] have obtained some information on sexual behaviour: multiple sex partners, history of sex with commercial sex workers, and history of STD were identified as risk factors for HIV. However, no studies to date have sought detailed information on types of sexual partnerships or patterns of casual sex in such populations. Such information might be used to focus HIV interventions on those at greatest risk of infection.
In 1991, implementation of an STD intervention programme began in 12 rural communities of Mwanza Region, Tanzania. The impact of this intervention programme was assessed in a community randomized trial [13–15]. Patients with HIV infection and other STD at baseline were identified from the cohort of subjects participating in the trial. These individuals, together with a random subsample of the cohort, were selected for detailed questioning about sexual attitudes and practices. The random subsample formed the basis for a survey of sexual behaviour in this population. In addition, case–control analyses were carried out by comparing cases of HIV or STD with controls drawn appropriately from the random subsample. The results of the sexual behaviour survey and the nested case–control study for STD will be reported separately. We report here the results of a case–control study of the association of HIV prevalence with sexual behaviour patterns and other risk factors in this rural Tanzanian population.
The demographic characteristics of the study population have been described previously . Implementation of a programme of improved STD treatment services began in the Mwanza Region in December 1991. A baseline survey was conducted prior to implementation and has been described previously . In brief, a cohort of approximately 1000 adults aged 15–54 years was selected from each of 12 rural communities using random cluster sampling. This cohort of 12 537 adults was enrolled, interviewed and examined between November 1991 and December 1992, prior to implementation of the intervention in each community. A sample of venous blood was taken from consenting adults and sera were tested for HIV antibodies by enzyme-linked immunosorbent assay (ELISA; Vironostika HIV MIXT Microelisa, Organon Technika, Boxtel, The Netherlands). All positive samples underwent confirmatory testing with a methodologically independent ELISA (Wellcozyme HIV 1+2 GACELISA, Murex Diagnostics, Dartford, Kent, UK). In the case of discrepant or indeterminate ELISA results, a confirmatory Western blot was performed (HIV-1 Westernblot, Epitope, Beaverton, Oregon, USA). Individuals wishing to know their HIV test result were referred for pre- and post-test counselling. HIV tests were performed on 12 500 (99.7%) of those enrolled and these individuals formed the sampling frame for a nested case–control study of HIV infection. Overall, 219 men (3.7% of the total) and 291 women (4.4%) were HIV-positive, and were therefore eligible for selection as cases. The controls were drawn from a simple random sample of one in eight persons in the sampling frame. All HIV-negative persons in this random sample were eligible as controls.
An attempt was made to revisit all eligible cases and controls. After obtaining informed consent, cases and controls were interviewed about sexual and non-sexual risk factors. A structured questionnaire was used to ascertain sociodemographic details and information on blood transfusions, injections, sexual attitudes, sexual practices, and perception of risk. In the questionnaire, sexual partners were classified as ‘marital’ (spouses/constant partners), ‘regular’ (partners for more than a few weeks), or ‘casual’ (partners for no more than a few weeks, including ‘one-off’ partners and commercial sex workers). Men and women were asked whether they ever got bruises on their sexual organs because of sexual intercourse. Women were asked whether they ever practised ‘dry sex’, involving the insertion of herbs or other substances into the vagina to reduce vaginal secretions. The questionnaire was designed in Kiswahili, back-translated into English and pre-tested in a pilot study in February 1993. Non-medical personnel, most of whom were teachers, community development workers or cultural officers from Mwanza Region were carefully selected and trained in interview techniques. Interviewers were unaware of the participants' HIV status. Very few participants requested their HIV test result and most were unaware of their HIV status. The study took place in May and June 1993, 6–17 months after the baseline survey of the 12 study communities.
In the analysis, each sex was considered separately. Risk factors for HIV have been broadly categorized as either sociodemographic or ‘proximate’ factors, the latter including factors likely to be directly related to HIV transmission. Odds ratios (OR), adjusted for age-group (15–19, 20–24, 25–29, 30–34, 35–44 and 45–54 years) and community of residence, were obtained for all risk factors using logistic regression. For most of the sociodemographic factors, no adjustment was made for additional confounders, because the aim was to describe the relative risk in different classes of individual, rather than to isolate the specific effect of a particular variable. OR for proximate factors and those sociodemographic factors relating to an individual's spouse were adjusted for confounders. All risk factors found to be significant after adjusting for age-group and community of residence were fitted in a logistic model. A final model included only those factors which remained significant after adjusting for all other factors in the model. Significance was assessed using the likelihood ratio test. Logistic regression was performed in Egret (Statistics and Epidemiology Research Corporation, Seattle, Washington, USA). Adjusted OR were used to estimate the population attributable fraction for each exposure . Each population-attributable fraction estimates the proportion of HIV infections that would have been prevented had the risk for all subjects been as low as in the baseline category while all other factors were kept constant. Although each population-attributable fraction is adjusted for confounders, it is not possible to estimate the joint population-attributable fraction for several exposures by summing several population-attributable fraction.
The participation rate among those sought for this case–control study was 66% for cases and 75% for controls (Table 1). Non-participation was largely due to the person not being found (20% had moved, travelled, died or could not be traced). Only 0.5% of those found refused to be interviewed. Fifty-four persons were excluded because of doubts about identity. The age distribution of cases and controls is also shown in Table 1. Since no male cases were under 20 years of age, the remainder of the analysis of men was restricted to the 149 cases and 394 controls aged 20–54 years. The analysis of women was conducted on all 189 cases and 574 controls.
Sociodemographic risk factors in men and women
OR for sociodemographic risk factors are shown in Table 2. There was no significant association between education and HIV infection in men. In women, the prevalence of HIV infection increased significantly with level of education, the small group of women with secondary education or higher having four times the prevalence of those with no education. Men and women in manual work, office work or business had a significantly higher prevalence of HIV infection than farmers, as did men in ‘other’ occupations (mostly fishermen). Those who had lived elsewhere in the past 5 years had a higher prevalence of HIV infection, especially men who had lived in another large town [OR, 2.13; 95% confidence interval (CI), 1.11–4.07], women who had lived in Mwanza town (OR, 3.52; 95% CI, 1.62–6.27), and women who had lived in another large town (OR, 3.19; 95% CI, 1.62–6.27). Travel was much less common among women than men but was more strongly associated with HIV. In particular, travel to another large town in the past year was significantly associated with HIV infection in women and gave rise to a population-attributable fraction of 9.7%. The prevalence of HIV infection was somewhat higher among Moslems, although in men this was not statistically significant. Moslems in this study shared a number of characteristics, some of which may have put them at higher risk of infection. Among the controls, Moslem women were more likely than other women to have lived in a large town in the past 5 years (14 versus 3%; P = 0.011) and to report five or more lifetime partners (30 versus 18%; P = 0.13), but their spouses were not more likely to have more than one wife (18 versus 25%; P = 0.47). Moslem men were more likely than other men to be circumcised (79 versus 29%; P < 0.001), to have lived in roadside settlements in the past 5 years (36 versus 7%; P = 0.003), to have travelled to Mwanza in the past year (64 versus 37%; P = 0.036), to have secondary or higher education (14 versus 5%; P = 0.054), and to report 10 or more lifetime partners (79 versus 56%; P = 0.096), but were not significantly more likely to have more than one wife (20 versus 14%; P = 0.62).
Current and past marital status was associated with HIV infection (Table 2). Men and women who were divorced or widowed were three times more likely to be HIV-positive than those currently married, and in women the HIV prevalence was higher in those never married than in those currently married. The higher prevalence in divorced or widowed women than among currently married women may be due partly to a greater number of lifetime partners and casual partners (28 versus 19% reported five or more lifetime partners, and 28 versus 7% reported a casual partner in the past year). Men who were divorced or widowed reported similar numbers of lifetime partners to married men, but more casual partners in the past year. For example, 33% of divorced or widowed men and 11% of those currently married reported three or more casual partners in the past year. Among those currently married, a substantial proportion (34% of male controls, 28% of female controls) had previous marriages that ended in divorce or widowhood, and these individuals had a higher HIV prevalence. There was no association between number of wives and HIV prevalence.
OR for marital status within different age-groups are shown in Table 3. In men, there was a significant interaction between marital status and age (P = 0.017). In men aged 30–54 years, a fourfold increased prevalence was found in those never married and in those divorced or widowed, compared with those currently married. In men aged 20–29 years, there was a higher prevalence in those divorced or widowed, but the group of men who had never married had only half the prevalence of those currently married. There was no interaction between marital status and age in women (P = 0.44), although, as for men, the highest prevalence was found among those aged 30–54 years who had never married.
Analysis of the spouse-related factors suggested that some married men and women may be at increased risk of HIV infection through their spouses (Table 2). First, among men there was a significant trend with age of current spouse (the younger the spouse, the higher the prevalence of HIV infection), which remained significant after adjustment for confounders (P = 0.020). Second, those men whose spouses had previously been widowed or divorced had an increased HIV prevalence although after adjustment for confounders, this excess was not statistically significant (P = 0.145). Among married women, there was no association between age of current spouse and HIV infection. However, those women whose spouse had previously been widowed or divorced had a higher HIV prevalence, although this effect was smaller and not significant after adjustment for confounders (OR, 1.46; 95% CI, 0.89–2.42). Furthermore, those women married to men employed in manual work, office work or business had a twofold increased HIV prevalence, even after adjustment for confounders (OR, 2.20; 95% CI, 1.22–3.95).
Proximate risk factors for men
OR for proximate risk factors among men are shown in Table 4. Only 2% of cases and controls had received a blood transfusion during the past 5 years; therefore, transfusions are clearly not a major source of infection in this population. There was a significant upward trend between HIV infection and the number of injections received in the past year, although this trend was no longer significant after adjustment for confounders. A substantial proportion of men (62% of controls) reported skin incisions or tattoos, but these men did not have a higher prevalence of HIV.
Most men (63% of cases, 65% of controls) had their first sexual intercourse aged 15–19 years, and 90% of men had their first sexual intercourse before they were aged 20 years. Age at first sexual intercourse was similar in cases (mean, 15.9 years) and controls (mean, 16.1 years) and was not associated with HIV infection (P = 0.90). There was a very wide variation in the reported number of lifetime partners, with 34% of controls reporting 20 or more partners and 18% reporting fewer than five. All men reported at least one sex partner. There was a highly significant trend for increasing HIV prevalence with increasing partners (P = 0.002), but the trend weakened after adjustment for confounders (P = 0.059). The population-attributable fraction for reporting more than one lifetime sex partner was 66% and the population-attributable fraction for reporting more than four lifetime sex partners was 18%.
We examined the association between HIV infection and reported number of casual and non-casual partners in the past year. Before adjustment for confounders, there was no association between the number of sexual partners in the past year and HIV infection. After adjustment for confounders, there was a significant negative trend (P = 0.011), those men reporting no partners having the highest prevalence and those reporting five or more partners having the lowest prevalence. A strong negative trend was also observed between the number of casual partners in the past year and HIV infection. Since the number of sexual partners in the past year was strongly correlated with lifetime partners, these effects were also examined without adjustment for lifetime partners. However, similar negative trends were observed even after adjustment for age, community, marital status, occupation, history of STD, and perceived risk of STD (data not shown). Those reporting sexual contact with commercial sex workers had a higher HIV prevalence, although numbers reporting such contacts were very small (4% of cases, 2% of controls) and the association was not significant. Casual sex with other types of partner was more common (in controls, 38% reported short-term partners and 9% reported other casual partners) but was not associated with HIV infection. Sex during travel away from home, or at dances, weddings or other traditional events is assumed to contribute to the risk of HIV infection. In controls, 13% reported casual sex during travel and 18% reported casual sex at dances or other events over the past year. However, neither of these were associated with HIV (OR, 0.70; 95% CI, 0.35–1.38 for travel; OR, 0.73; 95% CI, 0.40–1.31 for dances and other events). Bruising during sex was reported by 25% of controls and 38% of cases, but after adjustment for confounders, this effect was very weak (OR, 1.07) and not significant (P = 0.80). The main confounders were lifetime sex partners, history of genital ulcer or discharge, and perceived risk of STD.
There was no significant association between reported condom use and HIV infection, either before or after adjustment for confounders. Those reporting a history of genital ulcer or discharge during the past year had a higher HIV prevalence, even after allowing for confounders; the estimated fraction of HIV infections attributable to this effect was 15%.
The effect of circumcision on HIV prevalence was examined in a number of ways. Comparing circumcised men with non-circumcised men, and adjusting for confounders, circumcision showed a protective effect, although this was not statistically significant (Table 4). The main confounder of the effect of circumcision was occupation: 64% of non-farmers and only 26% of farmers were circumcised. Comparing men who had been circumcised before and after 15 years of age with non-circumcised men, and adjusting for confounders, showed a significant (P = 0.027) association with HIV: circumcision after age 15 years was associated with a lower HIV prevalence (OR, 0.48; 95% CI, 0.25–0.90), but circumcision before age 15 years was associated with a higher prevalence (OR, 1.34; 95% CI, 0.64–3.00). Circumcision was strongly associated with religion: among the controls, 64% of Moslems and 5% of non-Moslems were circumcised at birth, whereas 21% of Moslems and 71% of non-Moslems were never circumcised. However, excluding Moslems made little difference to the significance of the association with HIV (P = 0.005) or to the OR: after adjusting for confounders, circumcision after age 15 years was associated with a lower HIV prevalence (OR, 0.37; 95% CI, 0.18–0.74), but circumcision before age 15 years resulted in a higher prevalence (OR, 1.50; 95% CI, 0.57–3.90). The numbers were too small (n = 26) to examine the effect of circumcision in the Moslems.
The perception of being at risk of STD was significantly associated with HIV prevalence, but not the perception of being at risk of AIDS. Perceived risk of AIDS was associated with perceived risk of STD, but although men admitted to a high risk of STD (18% of controls), few men (4% of controls) would admit to being at high risk of AIDS. There was no association between knowing anyone with AIDS and HIV prevalence. Perceived faithfulness of spouse was significantly associated with HIV, those men reporting that their spouse was not faithful having the highest prevalence, although this effect was reduced after adjustment for confounders.
Proximate risk factors for women
OR for proximate risk factors among women are shown in Table 5. More women (4% of controls) than men had received a blood transfusion in the past 5 years, and this was associated with a twofold increased prevalence of HIV infection (OR, 2.40), but only a small population-attributable fraction (4%). There was a significant upward trend in HIV prevalence with increased number of injections, even after adjustment for confounders. Skin incisions or tattoos were associated with an increased HIV prevalence, although this effect was not significant after adjustment for confounders.
Most women (64% of cases, 66% of controls) had their first sexual intercourse aged 15–19 years, and almost all (92% of cases, 89% of controls) had their first sexual intercourse before they were 20 years. Age at first sexual intercourse was not associated with HIV infection (P = 0.79). The number of lifetime partners reported by women was lower and less varied than for men. However, most women reported more than one partner and a substantial proportion (18% of controls) reported at least five lifetime partners. Increasing life-time partners were significantly associated with increasing HIV prevalence, even after adjustment for confounders (P < 0.001 for trend), with a sevenfold (3.59/0.49) increased prevalence in those women reporting 10 or more lifetime partners. The population-attributable fraction for reporting more than one lifetime partner was 50%, and the population-attributable fraction for reporting more than four lifetime partners was 17%. Only one case and 16 controls reported no sex partners ever. This case had received a blood transfusion in the past 5 years, but had received no injections in the past year and had no skin incisions or tattoos. It is likely that the blood transfusion was the mode of HIV transmission in this case.
The few women who reported more than one sexual partner in the past year (9% of controls) or any casual partner in the past year (11% of controls) had a much higher HIV prevalence. These effects remained even after adjustment for age-group, community, travel, religion and marital status (data not shown). However, when further adjustment was made for lifetime partners, these effects were weaker and no longer significant (Table 5). Few women reported that they had casual sex during travel away from home (3% of controls), or at dances, weddings or other traditional events (4% of controls), and this was not associated with HIV prevalence. Bruising during sex was reported by 10% of controls and was associated with a small and non-significant increased HIV prevalence. Only 1% of women reported the practice of ‘dry sex’ ever, which was too few to examine its association with HIV.
Fewer women than men reported that they had ever used a condom and no association was found between condom use and HIV prevalence. Fewer women than men reported an STD in the past year and the association with HIV was not significant after adjustment for confounders. Very few women placed themselves in the category of being at high risk of STD (9% of controls) or AIDS (3% of controls). Risk perception was not significantly associated with HIV infection. However, there was a higher HIV prevalence among those women who had known anyone with AIDS, but this effect was weaker and not significant after adjustment for confounders. More women than men stated that their spouses were definitely ‘not faithful’ (22% compared with 5%), but this was not associated with HIV prevalence.
Compared with most previous studies of risk factors for HIV infection in representative samples of rural African populations [4–12], our study has provided more detailed information on casual and non-casual sexual behaviour patterns. Controls were drawn from the same population as cases and were not matched on age or any other factor. The unmatched design employed here has several advantages. First, the same controls were used for case–control studies of several outcomes (HIV, syphilis and other STD), so for given resources each study yielded a greater power. Since cases and controls were not matched on community of residence, the amount of fieldwork was similar in each of the 12 communities, rather than being proportional to the community-specific prevalence of HIV. The selection of individuals rather than matched pairs was logistically simpler. The confounding effects of age, community of residence and other factors were controlled for in the analysis, rather than in the design. The controls also provided a survey of sexual behaviour patterns in a representative sample of this population.
The findings of this study, however, are subject to the same limitations as all case–control studies. First, the results may be biased if the controls do not represent those who would have been selected as cases, had they satisfied the case definition. In our study, cases and controls were drawn from the same source population and the participation rates were broadly similar in cases and controls. Bias may also result if cases and controls differ in their abilities to recall or report. Information on past and current sexual behaviour is particularly prone to recall and reporting bias. In this population, most people are not aware of their HIV status, and most participants of this study chose not to be told their HIV status. Furthermore, interviewers were not aware of the participant's HIV status. Recall and reporting bias are therefore likely to be non-differential and so OR will tend to be underestimated. A further problem arising from case–control studies is the difficulty in determining temporal sequences of events. Our study is based on prevalent HIV cases. Therefore, information pertaining to the past year may not be relevant to an individual's HIV status, or may even represent the effects of HIV infection rather than its causes. This problem may be addressed by using incident HIV cases. We are currently conducting a case–control study based on incident cases of HIV infection in these communities.
Our study identified particular groups of men and women with a high HIV prevalence. First, the prevalence of HIV was higher in men and women who were divorced or widowed than in those currently married. In those aged 30–54 years, marriage was associated with the lowest HIV prevalence, the highest prevalence being in those never married. In those under 30 years, however, married men had a higher HIV prevalence than never-married men. Our findings are consistent with those from rural Uganda , except that we did not observe an increased prevalence of HIV infection among young married women. At younger ages, marriage may be a marker of recent sexual activity. However, being divorced, widowed or never married may reflect greater exposure to HIV-infected partners, as suggested by the higher number of casual sexual partners reported in these groups: in male controls (female controls), the proportions reporting three or more (one or more) casual partners in the past year were 33% (25%) in the widowed or divorced, 26% (28%) in the never married, and 11% (7%) in the currently married. Alternatively, it may be that HIV infection was the cause of divorce or widowhood.
An important finding from this study is that some married men and women may have been infected with HIV by their spouse. Women married to men in occupations other than farming had a significantly higher prevalence, as did men married to younger women. Our data do not indicate whether the spouse was the source of HIV infection in the index case. However, these associations persisted even after adjustment for the sexual behaviour of the index case. Furthermore, it is notable that many of the cases whose own sexual behaviour would be categorized as ‘low risk’ had spouses falling into ‘high risk’ categories: in women, 56% of cases reporting one or fewer lifetime partners, 33% of cases reporting two or more lifetime partners, and 14% of controls were married to men in manual work, office work or business; in men, 58% of cases reporting four or fewer lifetime partners, 39% of cases reporting five or more lifetime partners, and 37% of controls were married to women aged 24 years or younger. Thus it is possible that a husband in an occupation other than farming, with a risky lifestyle, is the source of HIV infection in some women. This is consistent with the belief that many African women are at increased risk of HIV infection through their spouse [1,17–20]. Likewise, in some men, it is possible that the source of HIV infection may be their younger wife. This is consistent with the high HIV prevalence among young women in this and other studies [4,7–11,14]. It is also consistent with the strong association between HIV and reported sexual partners among the women in this and other studies [7,8,18], and with evidence suggesting that some African women are driven to exchange sex for means of subsistence .
In these rural communities, most men and women have primary education or less, most men and women are farmers, and women tend not to travel, either to Mwanza or other large towns (Table 3). The groups who appear to be at particularly high risk of HIV infection are those not currently employed in farming, and the small group of women who travel or have secondary or higher education. These high-risk groups have been identified in other studies [5,7,8,11] and may be markers of more modern lifestyles or higher disposable incomes, which may be associated with risky behaviour. In contrast to other studies [11,12], we found the highest HIV prevalence in the small group of Moslems; in women, this excess risk was statistically significant. The Moslems in this study comprised only a few individuals and may not be representative of the wider population of Moslems. It is also worth noting that the higher prevalence in Moslems could not be attributed to polygamy, since in our study polygamy was not strongly associated with being a Moslem or HIV prevalence. A possible explanation for the increased prevalence, which is consistent with the findings of another study , is that Moslems were more likely to have lived in roadside settlements and large towns in the past 5 years, where the risk of HIV infection is greatest [7,14].
Blood transfusions were associated with a twofold increased in risk among women, although only a small population-attributable fraction (4%). The population-attributable fraction associated with at least one injection in the past year was 24% in men and 21% in women. These are likely to overestimate the true population-attributable fraction, since injections in the past year may be an effect of HIV-related illness rather than a route of HIV transmission. No data were collected on history of illnesses, so it was not possible to evaluate this association further. Skin incisions and tattoos were not significantly associated with HIV in men or women and are therefore unlikely to be a major route of HIV transmission.
Particular patterns of reported sexual behaviour were associated with an increased prevalence of HIV infection. As in other studies [4,5,7,8,12,18,19], there was an increased prevalence in those reporting higher numbers of lifetime partners. This association was particularly strong among women, in whom there was a sevenfold increased prevalence associated with 10 or more lifetime partners. The population-attributable fraction for women, associated with 10 or more lifetime partners (9.8%), is consistent with the finding of 9.4% among family planning clinic attenders in Nairobi . In our study, the population-attributable fraction for reporting two or more lifetime sex partners was 66% in men and 50% in women; these figures are similar to those found in rural Uganda (53% in men and 69% in women) . Given the limitations of recall and reporting, and the fact that some partnerships may have occurred long before the HIV epidemic, it is perhaps surprising that the association is so strong. Lifetime sexual history may be acting as a proxy for sexual behaviour during the past few years. In women, reported sexual behaviour in the past year was associated with an increased HIV prevalence, except when it was adjusted for lifetime partners. In men, however, reported sexual behaviour in the past year was not associated with HIV prevalence. This applied irrespective of whether we analysed total number of partners, or number of partners of any type, including short-term partners, commercial sex workers and other casual partners. There was even a significant negative trend of HIV prevalence with increasing partners and increasing casual partners in the past year. An obvious explanation for these findings is that sexual behaviour in men changed as a result of their HIV infection. Since awareness of HIV status is low in this population, changes in behaviour are more likely to be due to HIV-related illness or to the loss of a regular partner due to HIV/AIDS.
There was no increase in HIV prevalence associated with reported sex during travel away from home, or at traditional events. Few men reported sex with commercial sex workers in the past year, although there is a substantial amount of casual sex, and we cannot exclude the possibility that this sometimes involves an element of exchange. However, outside the towns and truck-stops, there are no clearly defined groups of commercial sex workers who could form the basis for targeted intervention programmes. Few women reported the practice of ‘dry sex’, so it was not possible to estimate its association with HIV infection, as suggested by some other studies . Bruising during sex was reported by 25% of male controls and 10% of female controls, but was not associated with HIV prevalence, in contrast to a study in Jamaica . Interpretation of the association between HIV and history of genital ulcer or discharge in the past year is subject to the same limitations as in other case–control or cross-sectional studies [6–8,11,12]: first, it is possible that this association reflects the confounding effect of some aspect of sexual behaviour for which we have not fully adjusted; second, it is difficult to determine the temporal sequence of these factors.
Many studies have reported a lower risk of HIV and other STD in circumcised men, although some other studies have given conflicting results . Our findings clearly demonstrate the importance of adjusting for confounders. On unadjusted analysis, there was no association between HIV prevalence and circumcision. However, circumcision was strongly associated with several other risk factors, including age, community, religion, marital status and current occupation. Most notably, 64% of non-farmers and only 26% of farmers were circumcised, and 78% of Moslems and only 29% of non-Moslems were circumcised. After adjusting for confounders, circumcision had a modest but non-significant protective effect. The substantial change in the OR after adjustment suggests that there may be residual confounding, for which we have been unable to adjust .
Our data suggest that circumcision at age 15 years or older was protective, whereas circumcision at a younger age was associated with an increased HIV prevalence. We cannot explain the greater apparent protective effect among men circumcised at older ages, which was surprising given that some of these men may have been circumcised after onset of exposure to HIV or other STD. Our results contrast with findings in rural Uganda  where Moslems, who are usually circumcised in childhood, had a lower prevalence of HIV infection. Although our study population included a suitable mix of circumcised and non-circumcised men, facilitating analysis of the effects of circumcision, the circumcised men formed a heterogeneous group. Men may have been circumcised for different reasons, which may have been differentially associated with their HIV risk.
The main conclusions of this study are, first, that most HIV infections in these rural communities occur through sexual transmission, although some may be due to non-sterile injections. The population-attributable fraction associated with blood transfusions or other routes of infection is small. Interventions to change sexual behaviour patterns are clearly a major priority. Some men and women are at high risk of HIV infection through large numbers of sex partners. Intervention strategies, therefore, should aim to reduce partner change and to promote condom use. However, some men and women with a low-risk profile, most notably those with few sex partners, may be at risk through their spouse or regular partner. This suggests that interventions should extend beyond the high-risk groups. The practice of dry sex and bruising during sex do not seem to play an important role in HIV transmission, whereas the role of male circumcision is unclear. Although a substantial proportion of the population are engaged in high-risk behaviour, commercial sex, as observed in urban surroundings, seems to play a negligible role in HIV transmission. Our results confirm marked heterogeneity in the risk of HIV infection in these rural communities, indicating the scope for risk-reduction strategies.
We wish to thank the Principal Secretary, Ministry of Health, the manager of the National AIDS Control Programme, and the Director General of the National Institute for Medical Research, Tanzania for permission to carry out and publish the results of this study. We thank the Regional Medical Officer, Mwanza, the Director of the National Institute for Medical Research, Mwanza, the Director of the Bugando Medical Centre, Mwanza, the Director General of the African Medical and Research Foundation, and regional, district, ward, and community leaders for their support. We are grateful to the study population, particularly those who gave their time to respond to the detailed questionnaire, and to the field team.