HIV Infection as a Risk Factor for Cervical Cancer and Cervical Intraepithelial Neoplasia in Senegal

Cervical cancer is the second most common cause of cancer mortality in women worldwide. Its incidence has decreased in developed countries, where cervical screening is widespread. However, it remains common in less developed countries, where access to screening is limited, and is the leading cause of cancer death among women in Africa (1-4). Exposures related to sexual and reproductive behavior play an important role in the etiology of cervical cancer. Human papillomavirus (HPV), a sexually transmitted infection, is clearly established as a necessary agent in the development of cervical cancer (2, 3, 5-10). Other established risk factors include smoking, increasing parity, early age at first intercourse, multiple sexual partners, and infection with other sexually transmitted diseases (3, 5-7, 11). Hormonal contraceptives also seem to increase the risk of cervical cancer in most populations studied. A recent meta-analysis by the International Collaboration of Epidemiological Studies of Cervical Cancer included 24 studies conducted worldwide (12). It found elevated risk of cervical carcinoma associated with both oral and injectable contraceptives, increasing with duration of use. Some studies of populations in sub-Saharan Africa, however, have not shown increased risk (9, 12).

The role of HIV in the etiology of cervical cancer is also unclear, especially in Africa. Immunosuppression is a risk factor for HPV infection and/or detection, and there is consistent evidence that HIV-positive women have higher prevalence of HPV infection, more persistent infection, and resulting higher rates of preinvasive cervical lesions (1, 2, 6, 7, 13-15). However, the evidence is mixed for an increased risk of invasive cervical cancer (ICC) associated with HIV (1, 13). In the United States and Europe, large studies have shown 5- to 8-fold increases in the risk of ICC associated with HIV infection or acquired immunodeficiency syndrome (16-18). In contrast, in a study of South African women between 1998 and 2001, Moodley and coauthors (19) found increased risk associated with HIV for low-grade squamous intraepithelial lesions and high-grade squamous intraepithelial lesions (HSIL), but no significant increase in the risk of ICC [odds ratio (OR), 1.17; 95% confidence interval (CI), 0.75-1.85]. Several authors have suggested that in Africa, where treatment for HIV is limited, the life span of HIV-positive women may be too short for ICC to develop (1, 2, 13, 16). However, in another recent study from South Africa, HIV-1 infection was associated with increased risk of cervical cancer (OR, 1.6; 95% CI, 1.3-2.0; ref. 20). In a cross-sectional study conducted in Senegal of commercial sex workers and women presenting to an outpatient infectious disease clinic between 1994 and 1998, HIV was associated with both HSIL and ICC (ORs of 3.7 and 6.7, respectively, for HIV-1 infection, and 7.1 and 16.0 for HIV-2 infection; ref. 8). However, that study was limited by a small number of invasive cancers and lack of histologic confirmation of disease status. Other studies have also suggested higher risks associated with HIV-2 than with HIV-1 (2, 21). The present study investigates the roles of HIV-1 and HIV-2 infection as risk factors for biopsy-proven cervical intraepithelial neoplasia and cancer in a West African population.

The study setting, subjects, and data collection methods have been previously described (22, 23). Briefly, data were collected in and near Dakar, Senegal between January 1998 and August 2000. A survey of the prevalence of cervical lesions was conducted in two groups of women age 35 y and older: a larger group presenting to an outpatient clinic in Pikine, a suburb of Dakar, for primary care; and a smaller group presenting to Dantec Hospital at the University of Dakar for nononcology care. Approximately 75% of eligible women were approached, all of whom agreed to participate. Subjects were excluded if they were pregnant or did not have an intact cervix. An additional group of subjects included women referred to the University of Dakar from clinics throughout the country with symptoms suggesting ICC. To improve comparability between case and control populations, this analysis was restricted to subjects residing in the capital city of Dakar.

Study subjects were interviewed to obtain demographic and reproductive information, general physical and gynecologic exams were conducted, and blood samples were collected for HIV serology. The study protocol specified collection of samples for cervical cytology and HPV testing from all participants; all women with abnormal cytology as well as a subset of women with normal cytology, both with and without high-risk HPV infection, were referred for colposcopy and biopsy. Liquid-based cytology slides and biopsy tissue specimens were prepared in Senegal and sent to Seattle for evaluation by the study cytopathologist and pathologist. The present analysis included only subjects who had biopsy results. With exclusions as defined above, the total number of subjects available for analysis was 984. Institutional review boards at both the University of Dakar and the University of Washington approved the study procedures.

This study is a case-control analysis of three outcomes: ICC, carcinoma in situ (CIN) 2-3/CIS (an earlier term for CIN 3), and CIN 1, as identified by biopsy. The prevalence of adenocarcinoma in our population was low and our ICC cases include only squamous cell carcinomas. For comparison with previous studies, the cytologic (Pap smear) result low-grade squamous intraepithelial lesion is correlated, although imperfectly, with the histologic (biopsy) diagnosis CIN 1, and HSIL with CIN 2-3. Subjects with negative histology and cytology served as controls for all three outcomes. Of 984 subjects enrolled, 150 had histology ICC, 92 had CIN 2-3, 70 had CIN 1, and 590 had negative cervical histology; the remaining 82 had other or missing histology and were excluded from analyses. To reduce misclassification, we further excluded any of the 590 control subjects with positive cytology (i.e., for whom the data showed positive results from a Pap smear despite negative biopsy results), reducing the number of controls to 515 and the total number of subjects to 827. HPV status was not used to define controls.

The primary exposure analyzed was HIV infection, with one analysis pooling subjects with HIV-1 or HIV-2 infection, and also separate analyses of HIV-1 and HIV-2 as exposures. As previously described (24), a two-test sequence was used to determine HIV status. Serum samples were first tested for the presence of either HIV-1 or HIV-2 antibodies (HIV-1/2 EIA; Sanofi Diagnostics Pasteur). Positive samples then underwent a second immunoassay to distinguish HIV-1 and HIV-2 antibodies (Multispot; Genetic Systems).

Multiple logistic regression analysis was used to estimate ORs and 95% CIs for each exposure (either HIV-1 or HIV-2, HIV-1 only, HIV-2 only), with separate models estimated for each outcome. Potential confounding factors evaluated included age, age at first sexual intercourse, lifetime number of sexual partners, use of oral and injectable contraceptives, smoking, parity, and level of education. Women with values of age at first intercourse <10 y were assigned missing values for this variable. Age, age at first intercourse, and parity were evaluated as continuous variables, and all others as dichotomous categorical variables. The number of sexual partners was classified as zero or one, versus two or more. We defined use of contraceptives as current or past use versus no use. Women were considered to be exposed to smoke if they were current or past smokers themselves, or if their husbands were known to be smokers; otherwise, they were defined as unexposed. Subjects' education was classified as none versus some (primary, secondary, or university). Potential confounding variables were assessed individually and considered to be confounders if they changed the estimated OR by >10%. These variables were then added sequentially to a multivariate model to determine whether each was still confounding when others were included. Of those evaluated, factors found to confound the relationships between HIV and ICC or CIN were education level and the use of oral or injectable contraceptives, and all final models were adjusted for these variables.

Table 1 shows characteristics of subjects that (based on prior research) were likely to be associated with cervical pathology. ICC cases were older than controls, with 32.2% of cases 55 years of age or older compared with 5.8% of controls. ICC cases were also younger at first sexual intercourse, with 24.0% of cases <15 years of age versus 15.3% of controls. Controls were better educated than ICC cases, with 40.2% having some education compared with 11.3% of cases. Both lifetime sexual partners and smoking history were similar for ICC cases and controls. Use of hormonal contraceptives was lower among women with ICC than among control women. Prevalence of exposure to smoking was low across all groups. Parity differed between ICC cases and controls, but without a clear pattern. For CIN 2-3 and CIN 1, there were no associations of any size between any of the three outcomes and the characteristics presented in Table 1. Table 1 also presents numbers of cases and controls by clinical site. Most ICC cases were recruited at Dantec Hospital, and most control subjects at Pikine outpatient clinic. Proportions of CIN 2-3 and CIN 1 cases recruited at the two sites were comparable with those of controls.

Of the 827 subjects analyzed, a total of 28 were infected with HIV, including 13 infected with HIV-1, 15 with HIV-2, and none with both viruses. Among the 515 controls without cervical neoplasia, 6 (1.2%) were infected with HIV-1 and/or HIV-2, compared with 10 (6.7%) of 150 cases of ICC, 10 (10.9%) of 92 cases of CIN 2-3, and 2 (2.9%) of 70 cases of CIN 1. After adjusting for confounding by education and hormonal contraceptive use, we found large increases in the risks of both ICC and CIN 2-3 associated with HIV infection (Tables 2 and 3), and a smaller increase in the risk of CIN 1 (Table 4). ORs were 6.5 for ICC (95% CI 2.1-19.8), and 10.4 for CIN 2-3 (95% CI 3.6-29.6). Few (n = 2) cases of CIN 1 were infected with HIV, and little can be concluded from the modest elevation in the risk of CIN 1 that we observed (OR, 2.8; 95% CI, 0.5-14.3). In separate analyses of HIV-1 and HIV-2, we found large increases in the risk of both ICC and CIN 2-3 associated with both HIV types, and smaller increases in the risk of CIN 1. Risk estimates were higher for HIV-1 than for HIV-2.

In a population in Dakar, Senegal, we studied associations between HIV infection and cervical neoplasia. We observed increases in the risk of ICC and CIN 2-3, but not of CIN 1, associated with HIV-1 or HIV-2 infection when participants with either HIV type were analyzed together. Separate analyses of infection with HIV-1 and HIV-2 showed similar results, with increased risk of ICC and CIN 2-3 but not CIN 1 associated with infection with either HIV type.

This study is among the first to show that the strong association between HIV and cervical cancer risk shown in Europe (18) and the United States (17) can also be found in sub-Saharan Africa. Our results are consistent with those of Hawes and coauthors (8) from another Senegalese population, in which risks of both HSIL and ICC were increased among women with HIV infection; however, they contrast with those of Moodley et al. (19) from South Africa, which showed no increased risk of ICC associated with HIV. A partial explanation for this difference could be the relatively high prevalence of infection with HIV-2 versus HIV-1 in our Senegalese population. HIV-2 was first described in West Africa, and its prevalence is highest there. Although both HIV-1 and HIV-2 can cause acquired immunodeficiency syndrome, HIV-2 has a longer latent period and progression to advanced disease is slower (25). Women infected with HIV-2 would be expected to live longer, and those also infected with oncogenic HPV to be more likely to progress to ICC. Studies in Côte d'Ivoire (2, 21) found an association between HIV-2 infection, but not HIV-1 infection, and ICC, although no associations were found in multivariate analysis and small numbers of HIV-infected ICC patients limited their ability to detect associations. Other studies distinguishing HIV-1 and HIV-2 infection have found increased ICC risk associated with both exposures, but higher risk with HIV-2 (8, 20). Our results are consistent with increased risk associated with both HIV types, although they do not support earlier findings of higher risks associated with HIV-2 compared with HIV-1. This could result from a limitation of our analysis, that the prevalence of HIV infection is relatively low in Senegal compared with other countries in sub-Saharan Africa; the number of HIV-positive subjects in this data set is relatively small, particularly when stratified by HIV type. The CIs around our risk estimates were therefore wide; however, the risk increases were large and, for ICC and CIN 2-3, unlikely to have occurred by chance.

Another limitation of our data are that in assessing and controlling for confounding, we did not have information on history of sexually transmitted diseases; however, we did have data on all other established risk factors for cervical cancer, including age, age of sexual debut, number of lifetime sex partners, parity, use of contraception, cigarette smoking, and education level, and assessed confounding by these variables. Another potential confounder is the clinical site where patients were recruited, either an outpatient clinic (Pikine) or Dantec Hospital at the University of Dakar. Those recruited at the University included subjects referred there from throughout the country for suspicion of ICC, although we restricted subjects in this analysis to those residing in Dakar. As a result, there are relatively few ICC cases from Pikine (n = 19), and few controls from Dantec (n = 31). We also found that none of the control participants from the Dantec site were among the relatively few in this analysis who were HIV positive, so we were unable to adjust for confounding by site. As discussed above, we did assess and adjust for a number of factors that may have varied by site and confounded the relationship between HIV status and cervical cancer. However, we cannot exclude residual confounding from other unmeasured factors associated with study site.

There is some debate in the literature on the definition of controls in a study of cervical neoplasia. Because HPV infection is considered necessary to the development of cervical neoplasia, cases are generally assumed to be HPV positive. Some authors argue that only HPV-positive controls should be analyzed, because only women who have been infected with HPV are at risk (7). One issue is how a given exposure affects the risk of development of cervical cancer from HPV infection. It may change the risk of exposure to and infection with HPV, or it may change the course of existing HPV infection and the likelihood of progression to cervical neoplasia. HIV would more likely have the latter effect, and the use of HPV-positive controls would be appropriate because the exposure would have no effect in those not infected with HPV. However, another issue is the measure of HPV available. HPV infection measured at the time of the study may be a transient infection, rather than the persistent infection associated with cervical cancer (3). In this study, HPV infection was determined when subjects entered the study, but information on earlier infection and persistence was not available; therefore, in this analysis, HPV status was not used to define controls.

Our findings support the association of HIV infection with advanced as well as early-stage cervical neoplasia in an African population, even one with a relatively low prevalence of HIV infection. Associations were strong for infection with both HIV-2, prevalent in West Africa, and HIV-1. These findings are notable given the high rates of HIV infection and low rates of cervical screening in Africa, which together make cervical cancer an important public health problem there.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.