Our study showed that women with cervical cancer were three times more likely to be infected with HIV-1 than women without cervical cancer. Furthermore, the risk of cervical cancer increased with age, higher parity, and among those who had ever smoked. Our estimate for the association between HIV-1 and cervical cancer of OR = 2.9 is consistent with the risk or odds ratio range of 1.6 to 2.4 reported in previous studies. Few studies have established an association between HIV and invasive cancer of the cervix in sub-Saharan Africa [1
]. To explain this situation, there are a number of plausible reasons.
First, though cancer of the cervix is the commonest malignancy in women in sub-Saharan Africa, it is rare when compared to many other diseases. Furthermore, invasive cancer of the cervix is less frequent than pre invasive cervical cancer. Second, it has been suggested that the combination of cancer of the cervix and HIV is usually lethal and many women with these two diseases die before they seek health care [35
]. However, the evidence for this view is lacking, perhaps because of the difficulty of identifying such cases. In contrast, Wabinga and colleagues in an analysis of population based data linked to the cancer registry in Uganda showed that HIV did not influence survival of patients with invasive cervical cancer [37
Third, methodological problems related to studies on invasive cervical cancer may have contributed to the situation. A review of the literature suggests that studies that have established an association between HIV and cancer of the cervix, in general, share certain characteristics. The common characteristics include the use of a population rather than a hospital setting [27
]; use of the cohort or cross sectional study design as opposed to the case control design [26
]; having large samples (usually thousands of participants) [26
]; and the use of population controls [27
]. It would therefore appear that methodological issues including the challenges of ensuring adequate power of studies (particularly since the association between HIV and cancer of the cervix is reported to be weak to modest), of minimizing bias, and of controlling for confounding may have contributed to the scarcity and negative findings of studies on invasive cervical cancer and HIV.
Confounding, in particular, is of special interest given the role of HPV in the aetiology of invasive cervical cancer. It has been established that HPV is the cause of invasive cervical cancer and it is believed that HPV negative cases of invasive cervical cancer should not exist if high quality methods for HPV detection are used [12
]. Therefore, in a study to assess the association between HIV and invasive cervical cancer, it would be impossible to control for HPV because there would be no HPV negative cases with invasive cervical cancer
. The implications of this are that HPV cannot be assessed as a confounder of the association between HIV and invasive cervical cancer. Instead, HIV should be assessed as a confounder for the relationship between HPV and invasive cervical cancer. However, even this approach appears to be futile because one of the prerequisites for such an assessment would be to establish an association between HIV and invasive cervical cancer among the HPV negative cases of invasive cervical cancer who theoretically do not exist
Data on control for HPV among studies that have looked at HIV and invasive cervical cancer is very limited and also conflicting. Ter Meulen [31
] and Moodley [28
] controlled for HPV and found no association between HIV and cervical cancer. However, Hawes and colleagues found an association between HIV and cervical cancer after controlling for HPV [26
]. Although all the studies were hospital based, the study by Hawes and colleagues was cross sectional (perhaps providing the same source population for cases and controls) and in addition had a much larger sample size.
In our study, we did not measure HPV and therefore we did not control for it in the analysis. It is possible that if we had measured it with currently available methods for detection of HPV and then adjusted for it, our estimate may have been weakened or even been nullified. This could perhaps be one of the reasons why Ter Meulen and Moodley were not able to demonstrate an association between HIV-1 and cervical cancer. Whether HPV should be adjusted for in the association between HIV-1 and cervical cancer or its precursors is an important epidemiological issue that requires further study. Our estimate was confounded by age and parity although the effect appears to be small. Although education was a confounder, it was not necessary to control for it because age and parity formed a minimum confounder group. Previous studies have reported that age and parity are independent risk factors for cervical cancer and are also associated with HIV-1 infection [9
]. Consistent with previous studies, smoking was significantly associated with cervical cancer but was not a confounder because it was not associated with HIV-1 infection.
Although the mechanism by which HIV increases risk of cervical cancer is not completely understood, studies suggest that HIV-induced immunosuppression leads to an inability to control the expression of HPV and the production of HPV oncoproteins E6 and E7 [40
]. According to Hawes and colleagues, this risk appears to be associated with increased HPV persistence that may result from immunosuppression related to HIV. Furthermore the risk is greater in women with CD4 counts less than 200 cells per microliter and in those with high plasma HIV RNA levels [19
]. Studies have shown that HIV-1 infection is associated with an increased rate of HPV infection, mainly restricted to HR-HPV types which are the cause of invasive cancer of the cervix [16
Women with higher parity were more likely to have cervical cancer than women of lower parity, with the risk of cervical cancer increasing by 1.1 times for each additional birth. The findings of our study confirm previous work that high-parity women are at high risk for cervical cancer [9
]. It has been suggested that parity is a marker of the oestrogen-hormonal environment throughout the fertile years of women as well as a marker of repeated cervical trauma among highly parous women. It is not known whether hormones intervene in cervical carcinogenesis but oestradiol has been reported to induce immortalization of HPV infected cells [42
Consistent with previous findings [11
], our study showed that women who reported that they had ever smoked were six times more likely to have cervical cancer than women who had no history of smoking. Studies suggest that cancer causing chemicals (benzopyrene) from cigarette smoke damage the cervix and make it vulnerable to HPV infection. Women who smoke and also have HR-HPV genital infection are twice as likely to have pre-cancerous cells or to get cervical cancer.
Our findings have important implications for women with HIV or cervical cancer in Tanzania and other sub-Saharan countries which have high burdens of HIV and cervical cancer. According to Chirenje and colleagues, 95% of health care facilities in East, Central and Southern Africa have the infrastructure for cervical screening but the major bottlenecks to screening and treatment of cervical cancer are lack of policy guidelines, infrequent supply of basic materials, and lack of suitably qualified staff [7
]. Ngwalle and colleagues have rightly pointed out the urgent need to develop a national policy and to introduce systematic screening for cervical cancer in Tanzania [43
]. Such a task, important and urgent as it is, is no easy undertaking for a resource-constrained country like Tanzania.
However, the findings of our study suggest that a feasible alternative for Tanzania could be adoption of a high-risk approach that targets HIV positive women for cervical cancer screening. The resources for the cervical screening would initially come from the HIV/AIDS programmes. As has been pointed out, antiretroviral therapy (ART) for HIV provides a golden opportunity to improve cervical screening through the sharing of the ART resources and the frequent check-ups of women on ART that can also be used to screen for cervical cancer [44
]. This approach would enable the Tanzania Ministry of Health and Social Welfare to develop the capacity for cervical screening and treatment in a gradual and feasible manner, while mobilizing additional resources and the population that would be essential for a national cervical cancer screening and treatment programme.
Our study had a number of potential limitations that may have distorted our estimates. Selection bias among the cases may have occurred in our study. The cases were selected from the only cancer treatment centre in Tanzania and thus were more likely to be representative, with regard to the exposure (HIV infection), of cervical cancers in the population than if they were selected from hospitals. However, a very small proportion of all cases of cervical cancer are seen in hospitals, and even fewer are referred to ORCI. The cases seen in ORCI are likely to be those who have access to health services (geographical, economic); women with advanced cancer; and possibly the survivors of HIV and cervical cancer.
In our study, the controls were mostly attendants or visitors of the cases. Attendants are usually female relatives who provide care for the patient while in hospital. The attendants or visitors usually came from the same community as the cases. Thus the controls in our study were "community" controls, derived from the same source population as the cases, and therefore representative of the non cervical cancer cases with regard to the exposure (HIV infection). To validate this, we found that the frequency of HIV-1 in the controls was similar to that in the general population. However selection bias may have occurred since the controls were significantly different from the cases in age, parity and education. Selection bias among the cases and controls could have caused an under-estimate of our findings.
Measurement bias may have occurred among the 20% of the controls who were not screened for cervical cancer leading to under-estimation. Similarly, women with pre-invasive disease could also have been included among the controls and caused under-estimation. Recall bias with possible differential misclassification and over or under-estimation of effect could have occurred in the assessment of age of first sexual coitus. Since there was no blinding, interviewer bias could also have occurred in the assessment of risk factors. However, we took precautions by training the interviewers and standardizing the instruments. We did not measure HPV and therefore did not control for it although as discussed earlier this could have been problematic.