Is HIV Infection a Cofactor for Cervical Squamous Cell Neoplasia?¹

More than one-third of the 8 million persons infected with the HIV virus worldwide are women (1), and AIDS is rapidly becoming a leading cause of death in women, especially in developing countries (2). Likewise, worldwide, cervical cancer is a leading cancer cause of death in women, particularly in developing countries (3).

Numerous studies of HIV infection in association with CN³(4) prompted the CDC to add invasive cervical cancer to their revised definition of AIDS-related conditions (5). However, the meaning of observed associations between HIV and CN is not clear. Several researchers have suggested that the relationship is not etiological because, in the United States where data are most complete, there has yet to be an increase in the incidence of cervical cancer parallel to the increasing rates of HIV infection in women (6). Others have suggested that the association between HIV and CN is due to uncontrolled confounding by factors such as HPV, or other sexual and reproductive variables (7). Alternatively, the relationship between HIV and CN may be causal, either directly via depressed cellular immunity, or as a cofactor with HPV (8, 9).

Because there are few controlled studies testing the relationships between HIV, HPV, and CN, we pooled data from published sources to explore these relationships. We hypothesized that the relationship between HPV and neoplasia would be modified by HIV, with HIV-positive women demonstrating higher odds of disease in the presence of HPV than HIV-negative women. Moreover, we postulated that among HIV-infected women, women with higher levels of immunosuppression would have higher odds of neoplasia in the face of HPV infection than women with lower levels of immunosuppression.

A meta-analysis was conducted pooling data from published sources to address the research objectives. Meta-analysis is a technique that can be used to summarize results of studies performed in diverse settings with differing methodologies (10, 11, 12, 13). Such analyses are useful when no one study has sufficient power to address a particular question; meta-analysis is also useful to answer questions that were not posed in the original studies. All of these issues were considered in our choice of meta-analysis. However, given the post-hoc nature of these analyses, the results should be considered as hypothesis-generating (12).

Multiple sources were used to identify all potential research for inclusion in the analysis. First, a Medline search was conducted for all English language articles published between January 1986 and March 1998. The earlier date of 1986 was selected because AIDS was not well described or reported in women before this time period. The following key words were used for the search: Pap smear, cervical cytology, cervical cancer, CN, cervical intraepithelial neoplasia, SIL, or invasive cervical cancer, and AIDS, HIV, and HPV. Second, to ensure complete ascertainment (14), the search was supplemented by a review of Current Contents (1997–98) and bibliographies of all pertinent original and review articles.

Third, given the rapidly evolving database in this subject area, all meeting abstracts of the Annual International Conferences on AIDS and Annual Papillomavirus Workshops were reviewed for titles containing the same key words. For identified abstracts, all author names were searched on Medline and authors and titles were searched in Current Contents to identify potential published journal sources of these data. Finally, for articles that appeared eligible, but did not contain data in a format that permitted abstraction and reanalysis, the authors were contacted to provide the raw data.

“Methods” and “Results” sections were reviewed for eligibility by three investigators (J. S. M., P. K., L. E.) using a standardized abstraction form (15, 16). Reviewers were blinded to author names, journal, and the discussion sections. Study inclusion criteria included having data available on HIV, HPV, and CN among nonpregnant women; if pregnancy status was not included, it was assumed that the subjects were not pregnant. Additional inclusion criteria included having appropriate comparison groups, laboratory measurement of HIV infection and degree of HIV immunosuppression, HPV infection and types, and an adequate Pap smear classification schema. If data, or data subsets, had been published in more than forum or time period, the largest data set that contained the highest quality of information was selected for the final sample. Review articles, case reports, and animal studies were excluded from consideration.

HIV infection was defined serologically (from ELISA and Western blot tests) or clinically, using standardized criteria (1, 17, 18). All HIV viral types were considered (i.e., HIV-1 and HIV-2). Cellular immune suppression among women with HIV infection was defined either clinically by CDC class (class III and IV, symptomatic versus class II, asymptomatic; Refs. 1, 17, 18), or by laboratory measures (CD4+ count, <500 versus ≥500; 19). HPV positivity and types were measured by Southern blot, PCR, or dot-blot hybridization. Studies diagnosing HPV by the clinical presence of condyloma (on physical examination or colposcopy), cytological interpretation of a Pap smear, histological interpretation of a biopsy specimen, or filter in situ hybridization were only considered if the remainder of the study was of high quality. CN was defined as low-grade SILs or more severe (Bethesda system) or the equivalent; histological diagnoses were not available from the majority of studies. There were insufficient numbers of cases to stratify analyses by level of neoplasia (i.e., low versus high-grade SIL) or HPV type. Adenocarcinomas of the cervix were excluded.

For studies eligible for inclusion, the following data were abstracted: the study setting; country/city; design, number, and age of subjects; source of control population; mode of HIV transmission; type of HIV virus, if specified; sexual and reproductive risk factors; sexually transmitted disease history or concurrence; degree of HIV immunosuppression; and numbers of subjects in each exposure and disease subcategory.

The concordance between investigators was 100% for final sample inclusion/exclusion status, and 95% for specific data items among the studies in the final study sample. Discrepancies were resolved by discussion, review of the methods and/or results, and consensus.

Overall and strata-specific ORs and 95% CIs were calculated individually for each study 2 × 2 table using standard techniques (20). Because many studies had data tables with zero cell counts, 0.5 was added to each cell of all studies for calculation of an estimated OR and CI. The MH test for homogeneity of studies was conducted separately for the overall and sub-strata. In strata with zero cells, this test could not be calculated because we could not calculate variance. In these situations, the individual studies were examined for consistency. Rejection of the null hypothesis of heterogeneity implies that the effects are homogeneous across studies. If there was homogeneity of effects, the MH method was then used for calculating a summary OR based on the raw data, adding 0.5 to zero cells. This method weighs studies by the inverse of the within-study variance to provide a pooled estimate of the OR; thus the sample size of a study determines its contribution to the summary OR (20, 21). The MH summary OR for pooled data are still valid when data are heterogeneous, although the result is less powerful than in the condition of homogeneity (20). Interaction was considered to occur if the substratum ORs were unequal (21). In addition, to quantify the independent effects of HPV and HIV, and to formally test for interaction, we used a generalized estimation equation logistic regression model (22). This method takes the same form as a standard logistic regression, but adjusts parameter estimates for observation intercorrelation attributable to study observation clustering. The model-dependent variable was CN, and the independent variables were HPV infection, HIV infection (positive or negative), HIV immunosuppression (suppressed HIV positive, nonsuppressed HIV positive, and HIV negative), and interactions between HPV and HIV.

There were 78 original research articles identified by the search; thirty (38.5%) of these met the criteria for study inclusion. The most common reason for exclusion included lack of one or more comparison groups (n = 39; Refs. 6, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 33, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58); the remainder were excluded due to a focus on non-HIV-related immunosuppression (n = 3; Refs. 59, 60, 61), pregnant women (n = 1; Ref.62) or contained data reported more completely in another publication (n = 4; Ref. 63, 64, 65, 66), or has substantial design flaws threatening validity (n = 1; Ref.67). Among the 30 studies potentially eligible for inclusion, 11 included data that could be abstracted for analysis ((7, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77); the remaining 19 study authors were contacted to provide raw data for analysis (26, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94). We received a response from 8 of 19 authors (78, 79, 80, 81, 82, 90, 91, 92); of these, four provided data in a manner that was amenable to analysis (81, 82, 91, 92); others provided data that did not correspond to the data in the original report (i.e., included updated and unpublished data). We also contacted and received additional data from two of authors whose original studies included some raw data (72,70). One study (90) conducted additional analyses on specimens collected from an earlier study (89). Thus, we present results for a total of 15 studies (Table 1; Refs. 7, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 81, 82, 91, 92); the remaining 15 potentially eligible studies were excluded. The excluded studies were very similar to those included in design, year of publication, mean population age, sample size, settings, methods, and results [i.e., magnitude of results and proportion with null results (data not shown)].

Thirteen studies (7, 68, 69, 70, 71, 72, 74, 75, 76, 77, 81, 91, 92) contained sufficient data for inclusion in analyses of the relationship between HIV, HPV, and SIL; two only included data for HIV-positive women (73, 82). The individual studies all demonstrated a significant association between HPV and SIL among HIV-infected and -noninfected women (Table 2). Together, the pooled data demonstrated that women who were HPV positive had odds of having SIL that were 8.1 times higher (95% CI, 6.5–10.1) than the odds for HPV-negative women. Stratifying by HIV status, the odds of SIL among HIV-positive women were generally higher than the odds for HIV-negative women. The pooled summary OR for the association between HPV and CN were 8.8 (95% CI, 6.3–12.5) and 5.0 (95% CI, 3.7–6.8), respectively, for HIV-positive and -negative women. These data suggest that HIV interacts with HPV to increase the odds of neoplasia; however, because the CIs overlap minimally (upper limit for HIV− 6.8 versus lower limit for HIV+ 6.3), we can rule out a lack of interaction effect.

Six articles (7, 72, 73, 81, 91, 92) contained sufficient data among HIV-positive women to evaluate the role of immunosuppression on the odds of neoplasia. When data were stratified by level of immunosuppression, the OR was higher for suppressed than nonsuppressed women in four of six studies; point estimates for the two studies with nonsignificant results (i.e., CI includes 1) are more difficult to interpret (Table 3).

Table 4 summarizes the multivariate results. HPV was the strongest predictor of neoplasia, considering HPV, HIV, and immunosuppression among HIV-infected women. As suggested by the univariate summary ORs, there was a significant interaction between HPV and HIV infections; immunosuppressed women were also more likely to have neoplasia.

The results of this meta-analysis suggest that HIV is a cofactor in the association between HPV and CN; this effect seems to vary by level of immune function. These hypotheses are biologically plausible and are supported by the general consistency of findings across studies and similar types of neoplasia (i.e., anal) and the dose-response-like relationship for level of immunosuppression.

Although the precise biological mechanism underlying the interaction between HIV and HPV has yet to be determined, there are two potential mechanisms for the interaction: (a) HIV and its resultant effects on immune function affect the susceptibility to, severity of, and/or potential oncogenicity of HPV; and/or (b) there is a molecular interaction between HIV and HPV.

Biological mechanisms through which cellular immunosuppression could facilitate the oncogenic effects of HPV include prolonging the length of time of an HPV infection (95, 96, 97), increasing HPV viral load (96, 98), allowing for more rapid HPV replication (65), persistence (56), or progression (65), or impacting on Langerhans’ cells (90). Enhanced oncogenicity of HPV in HIV-infected women is also supported by clinical observations of more rapidly progressive disease (37, 99) and higher rates of recurrence among HIV-positive compared with HIV-negative women with HPV (40, 100, 101). Indirect evidence also links cellular immune responses and HPV. For example, there are increases in both HPV and CN in transplant patients, pregnant women, aged women, and women with AIDS (4, 6, 37, 61, 62, 63, 102, 103, 104, 105, 106, 107). Cellular immune suppression also seems to increase the risk of anal neoplasia among HIV-positive men with anal HPV infection, where risk of neoplasia increases with increasing level of immunosuppression (108, 109). Preliminary data also suggest a molecular interaction between HIV and HPV, where the HIV-1 tat protein transactivates HPV and effects expression of HPV- E2 and the E6 and E7 oncoproteins (110, 111).

Our pooled results suggesting an interaction between HIV and HPV are consistent with the results of other investigations (65, 78). For example, Klein et al.(65) noted odds of neoplasia of 43.6 (95% CI, 2.3–818.7) and 20 (95% CI, 2.2–180.7) among HIV-infected women with moderate or severe immunosuppression, compared with odds of 3.9 (95% CI, 1.2–12.6) among HIV-negative women (65). In our study, 11 of 13 studies (85%) noted higher odds of neoplasia in HIV-positive women compared with HIV-negative women with HPV infection; the remaining two studies had wide CIs and nonsignificant results. Last, all but two of the studies that included data on immune function suggest that there is a dose-response relationship, with women having the lowest CD4+ counts having the highest risk of disease. In one study where the relationship was not significant for a cut-point of a CD4+ count of 500, there was an increased odds of disease in women with CD4+ counts below 200 compared with those with higher counts (73). In the other inconsistent study (91), nonsignificant results for suppression may be a result two factors. First, data from HIV-1- and HIV-2-infected women were combined, where CD4 counts for HIV-2-infected women were more similar to those in HIV-negative women, biasing results to the null. Second, there were only 15 women with CD4 counts of <500, and clinically apparent immunosuppression was rare in this group. Further study of HIV-1 and -2 infections in relation to HPV and neoplasia risk will be important (91).

Our conclusions must be considered in light of several caveats, including the post-hoc nature of the analysis, the inability to assess the role of uncontrolled confounding, power, inconsistent studies, possible misclassification, inability to separate types of HPV and degree of neoplasia (i.e., preinvasive and invasive), and lack of data on temporality of effects.

A meta-analysis such as this, where the original data were collected for other purposes, should be considered preliminary; prospective data collection will be important to confirm our hypothesis. Also, meta-analyses are prone to “publication bias,” where studies with positive effects are more likely to be published than those with no association. Such bias may overestimate neoplasia risks, but should not have affected our conclusions regarding interactions.

Pooled ORs for the association of neoplasia and HPV showed clinically meaningful differences between HIV-positive and -negative women, and suggested an interaction of HIV and HPV, with minimal overlap in CIs; this interaction was significant in multivariate analysis (P = .01). As more quality data are published, it will be important to extend our analyses to confirm and explain interaction effects.

Several variables may confound or bias the relationships being explored, such as route of HIV transmission, type of HIV (91), anti-HIV medication use, other sexual risks, country, and HIV prevalence in the general population. Unfortunately, raw data were not presented in such a manner to allow for control of these variables in our analyses.

Some data from three studies were contradictory or inconclusive. There are several possible explanations of these apparently discordant findings, including true differences, lack of power, chance, selection bias, confounding or differential misclassification, a short duration of HIV infection relative to the natural history of cervical carcinogenesis, competing mortality, and/or difficulty accurately diagnosing cervical cancer and its precursors in HIV-infected women. For example, in the study by Johnson et al.(73), although more than three-quarters of the HIV-infected women who were HPV positive had HPV types of high to moderate oncogenic potential, only one woman was diagnosed with high-grade SIL. The mean time from infection to evaluation was 24 months; thus, many of these women may still develop neoplasia. In two studies, failure to detect strata-specific differences between HIV-positive and -negative women may have been due to the facts that in one study 93% of the HIV positives were asymptomatic (CDC stage II; Ref. 68), and the other study had a small sample, sparse data, and limited power to detect effects (70).

Misclassification of subjects can effect risk estimates. Two-thirds of the included studies (six of nine) relied on a cytological diagnosis of CN; only three studies included histopathological results. If women with neoplasia are falsely labeled as normal, estimates will underestimate the true effect, if misclassification is random with respect to HPV or HIV infection status. If there is some systematic misclassification of women (differential misclassification), results may produce artificially high or low estimates of risk. It is difficult to ascertain the potential for differential misclassification in the studies included in this analysis (28, 83, 85, 112, 113), although there is no reason to believe that any misclassification would be not differential with respect to HIV status, the variable of interest for this study.

It would have been interesting to further stratify our results by HPV type (i.e., low versus high oncogenicity), numbers of HPV types, and viral load because these factors seem to be related to the risk of neoplasia (63, 65, 73, 92, 114, 115). Finally, our data were all cross-sectional and do not allow us to determine the temporality of effects. For instance, does having HPV and/or CN facilitate transmission of, and infection with, the HIV virus; or does having HIV immunosuppression enhance the effects of HPV in producing CN? The precise role of HPV in transmission of HIV is not yet known (9, 116, 117, 118, 119). This is an important area for future research.

If HIV is further confirmed as a cofactor with HPV in cervical carcinogenesis, the roles or modifying influences of different HIV viral types and other potential cocarcinogens will need to be evaluated in well-controlled, prospective research (91, 120, 121). Examples of factors which have been suggested to modulate cervical cancer risk include repeated and/or concurrent exposure to sexually transmitted antigens, or cervical trauma (120, 121), cigarette-related carcinogens (122, 123), contraceptives (123, 124), and nutritional deficiencies (125), including B-carotene (126), folate (127), and vitamin C (128, 129). As more women are included in clinical trials and receive primary preventive care for HIV infection, the observed relationships between HIV, HPV, and CN will need to be reevaluated, controlling for use of AZT or other drugs and access to care. Finally, if the role of cellular immunity is confirmed for HPV-mediated carcinogenesis, a critical question that must be addressed is whether less extreme immune depression, such as that seen with normal aging or chronic disease, also increases the risk of cervical cancer.

The elucidation of the relationships between immune function, HPV, and cervical carcinogenesis will be critical to efforts to preventive efforts, including HIV and HPV vaccinations. Such efforts will be key to decreasing the high burden of disease and deaths attributable to the dual epidemics of cervical cancer and HIV infection in medically under-served women worldwide.

We acknowledge the contributions of the authors of the original research included in this meta-analysis; we specifically thank the following authors who provided raw data based on original reports: M. Conti, S. Cu-uvin, G. Gentile, C. Langley (and C. Critchlow), R. Matorras, J. Smith, A. Spinillo, G. Van Doornum, and A. Williams. We also thank Marianne Fahs, MPH, Ph.D., Ruby Senie, Ph.D., Herbert Peterson, M.D., and Karen Garibaldi for contributions to an earlier version of this work; Sten Vermund, M.D., Ph.D., Thomas Wright, M.D., and Alfred Neuget, M.D., Ph.D., for review and comments on earlier versions of the manuscript; Colin Begg, Ph.D., for review of the statistical analyses; and Kathy Summers and Ronda Jones for manuscript preparation.

This manuscript is dedicated to the memory of our colleague Rachel G. Fruchter, Ph.D., MPH, a tireless advocate for women’s health care.