Incidence of Cervical Squamous Intraepithelial Lesions in HIV-Infected Women FREE

Invasive cervical cancer and its precursor lesions are the most important gynecologic manifestations of human immunodeficiency virus (HIV) infection.¹ Women with HIV infection are about 5 times more likely than uninfected women to have cervical squamous intraepithelial lesions (SILs), the precursors to invasive cervical cancer.^{2 - 5} Invasive cervical cancer is found about 3 times more frequently in HIV-infected vs uninfected women in New York City.⁶ Also, young HIV-infected black and Hispanic women in the United States may have a 4- to 7-fold increased risk of invasive cervical cancer.⁷ However, most studies of this association have been cross-sectional in design, and the underlying causes for the association remain unclear.^{3 ,8 - 10} Prospective studies are needed to better define the relative importance of risk factors in the development of cervical SILs.

To increase our understanding of cervical disease in HIV-infected women, we began a prospective cohort study in 1991, the New York Cervical Disease Study, designed to compare the incidence of SILs in a well-defined cohort of HIV-infected and uninfected women and to assess the role of risk factors in the pathogenesis of these lesions.

From October 1, 1991, to June 30, 1996, a total of 925 women participated in the New York Cervical Disease Study. Most HIV-infected women were recruited from HIV–acquired immunodeficiency syndrome (AIDS) clinics, and most uninfected women were recruited from a methadone treatment program or sexually transmitted disease (STD) clinics. There were 221 women recruited from the St Michael's Medical Center (Newark, NJ) HIV-AIDS clinic, 98 from the Columbia-Presbyterian Medical Center (New York, NY) HIV-AIDS clinic, 142 from the Beth Israel Medical Center (New York) methadone maintenance treatment program, 243 from a New York City Department of Health STD clinic in the south Bronx, 39 from an STD clinic in Newark, and 182 from a clinic established in New York City to study heterosexual HIV transmission. At the STD clinics and the methadone maintenance program, patients were recruited by clinic nurses and distributed fliers. At the HIV-AIDS clinics and the clinic established for studying heterosexual HIV transmission, participants were recruited directly by their primary health care providers. Women were recruited without regard to risk factors for cervical disease or clinical HIV status unless too ill to participate in gynecological examinations. Because participants were not recruited directly by study personnel, the number of women approached for the study and an estimate of the percentage of women electing to participate are not available. Full enrollment details for this study are published.^{3 ,11} Written informed consent was obtained from all women, and the institutional review boards of participating institutions approved the study. A total of 272 (29%) of the 925 women enrolled in the cohort were excluded from the current analysis: 96 women (79 HIV-infected and 17 uninfected) had SILs at enrollment, 140 women (56 HIV-infected, 78 uninfected, and 6 of unknown serostatus) were examined only once or had missing HIV data, 20 women had a hysterectomy, 6 women HIV-seroconverted during the study, and 10 had no initial interview.

At enrollment, subjects were interviewed and had a gynecologic examination including Papanicolaou (Pap) test, cervicovaginal lavage for human papillomavirus (HPV) DNA testing, photograph of the cervix, and colposcopy. Biopsies were done on colposcopic abnormalities thought to represent SIL. Thirty HIV-infected and 25 uninfected women had minor colposcopic abnormalities considered too trivial for biopsy, contraindications (eg, low platelet count), or refused. Follow-up gynecologic examinations were scheduled at 6-month intervals and included an interview, Pap test, and cervicovaginal lavage. All women had colposcopic assessment including cervical biopsy, if indicated, at every examination. The HIV-seronegative women were retested annually for HIV. The CD4 cell levels were assessed at follow-up visits in HIV-infected women with prior CD4 cell count of more than 200 × 10⁶/L. A total of 3549 examinations of 653 women were performed, of which 255 (7%) were excluded because the examination occurred after SIL diagnosis (n = 213) or data were missing (n = 42).

Cervical cytology and biopsy specimens were interpreted by persons masked to HIV and clinical status. Pap tests were evaluated at 1 of 3 cytology laboratories. A single gynecologic pathologist interpreted cervical biopsies. Biopsy results were reported using the low-grade and high-grade SIL terminology.¹² Seventeen (4%) of 428 biopsies in the study were diagnosed as having borderline changes when the pathologist could not distinguish between inflammation and/or repair and a SIL. For our analysis, women with borderline changes were categorized with women who did not have SILs; in an analysis of the data classifying these women as having SILs, none of the major study conclusions were changed. Subjects were tested for HIV with standard screening and confirmatory methods. For HIV-infected women, CD4 cell counts were obtained from clinic records or specifically for this study. In several clinics, patients with CD4 cell counts of less than 200 × 10⁶/L did not have further testing and, for this analysis, the counts are considered to have remained less than 200 × 10⁶/L. Of 66 women with these low CD4 cell counts, 17 had no additional measurements, 30 had 1, 13 had 2 or more, and 6 had 3 or more. Only CD4 cell counts from within 6 months of the clinical examination were included.

Cervicovaginal lavages with 10 mL of phosphate-buffered saline (pH, 7.4) were obtained at each visit. The lavage sample was analyzed for HPV DNA using a strategy based on a 2-stage polymerase chain reaction (PCR).^{11 ,13} First, all samples were tested for HPV DNA using PCR and L1 "consensus" primers that identify the common anogenital HPV types. If a sample was found to contain HPV DNA, the type(s) of HPV present in the sample was determined using restriction fragment-length polymorphism typing (RFLP).¹⁴ The RFLP method identifies the following types: 5, 6, 8, 11, 16, 18, 31, 33, 35, 39, 41, 43, 44, 45, 47, 51, 53, and 57.¹⁴ The HPV types not corresponding to 1 of these 18 types were classified as "novel." In 7% of the HPV DNA–positive samples L1 PCR products could not be typed because too many types of HPV were present. To overcome potential limitations associated with RFLP typing of L1 PCR products, all samples found to be HPV DNA positive using the L1 consensus primers were also subsequently retested using PCR and E6 type-specific primers for HPV types 16 and 18. Type-specific PCR for HPV types 16 and 18 detected considerably more infections with these types than did the L1 PCR; 66 (46%) of 145 HPV-16 infections and 103 (79%) of 131 HPV-18 infections were detected only with type-specific E6 primers. We did not perform DNA sequencing to determine whether the same molecular variants of a given HPV type were detected at the visits. The HPV DNA–negative samples were tested with "control" Ki-ras gene primers to ensure that the sample DNA was suitable for PCR analysis.¹⁵ If the Ki-ras gene did not amplify, samples were considered inadequate for analysis and excluded.

Data analysis was performed with SAS statistical software, version 6.12 (SAS Institute, Cary, NC). Statistical associations between HIV status and categorical variables were tested by Pearson χ² test. The incidence of SILs by HIV status, HPV DNA status, and CD4 cell count at entry were estimated using the Kaplan-Meier method, and the statistical significance was assessed using the log-rank test. Univariate and multivariate analyses of associations between incident SILs and possible risk factors were performed using Cox proportional hazards models. Variables that were significant in univariate analysis and those that were risk factors in previous studies were included in a number of different models for the multivariate analyses. The assumption of proportionality was tested and met in the multivariate analyses. Antiretroviral therapy was included as a time-dependent variable in a separate Cox proportional hazards model restricted to HIV-infected women.

At study initiation in 1991, the sample size calculations for the original enrollment were based on our cross-sectional objectives. We calculated that 400 infected and 400 uninfected women would have to be enrolled to investigate high-risk women, assuming that SIL prevalence would be about 8% in the uninfected women and a sample size of 800 women (400 infected and 400 uninfected) would enable us to detect a relative risk (RR) of 2.25 or greater of SILs in infected women with 95% confidence and a power of 95%. For the prospective study, we continued to follow up women who did not have evidence of SIL at entry.

The 328 HIV-infected and 325 uninfected women were similar in age, race, income, age at first sexual intercourse, and current cigarette smoking (Table 1). However, HIV-infected women were significantly more likely to report history of prostitution, intravenous drug use, genital herpes, or genital warts (P<.05 for each). Of the 653 women included in the analysis, 200 had 2 to 3 follow-up visits, 267 had 4 to 6 visits, and 186 had 7 to 10 visits. Mean length of follow-up for HIV-infected women was 29.5 months vs 33.3 months for uninfected women (t test, P<.001). At enrollment, the mean CD4 cell count for the HIV-infected women was 429 × 10⁶/L (66 [23.7%] had a count of <200 × 10⁶/L; 119 [42.7%], 200-500 × 10⁶/L; 94 [33.7%], >500 × 10⁶/L; and 49 were missing baseline values). This study was conducted prior to the introduction of protease inhibitors. Of the 328 HIV-infected women, 178 (54%) took 1 or more antiretroviral medications for 6 or more consecutive months during the study. Of 178 women, 136 (76%) took only 1 antiretroviral drug. Zidovudine alone was taken by 118 women, didanosine by 15, zalcitabine by 2, and stavudine by 1. Also, 39 women (22%) took 2 antiretroviral drugs in combination for 6 or more consecutive months, and 3 (2%) took 3 drugs in combination. Women taking 2 drugs took zidovudine in combination with didanosine (28 women), zalcitabine (8), lamivudine (2), or stavudine (1). All 3 women who took 3 drugs took zidovudine, didanosine, and zalcitabine.

At enrollment, HPV DNA was detected in 143 (54%) of 264 HIV-infected women and 85 (32%) of 265 uninfected women (Table 2). High-, intermediate-, and low-risk, novel, and multiple HPV types were more often detected in HIV-infected women vs uninfected women and in both groups, HPV DNA detection was highly variable over time. Of women with at least 2 visits during a 3- to 12-month period, 155 (61%) of 256 HIV-infected and 58 (23%) of 257 uninfected women had HPV DNA detected at 2 or more visits (ie, were persistently HPV DNA–positive, P<.001) (Table 2). Of the 213 women persistently positive for HPV, 4 had visits less than 5 months apart, (intervals ranging from 3.5 to 4.8 months).

Sixty-seven (20%) of 328 HIV-infected and 16 (5%) of 325 uninfected women developed biopsy-confirmed SILs during the study (Figure 1, A). Incidence of SILs was 8.3 and 1.8 cases per 100 person-years of follow-up in HIV-infected and uninfected women, respectively (P<.001). In HIV-infected women, 61 (91%) of 67 incident SILs were low grade and 6 (9%) were high grade vs uninfected women, in whom 12 (75%) of 16 incident cases were low grade and 4 (25%) were high grade (P = .08). No invasive cervical cancers were identified. In HIV-infected women, the association of CD4 cell level with incident SILs was not significant (P = .08) (Figure 1, B).

A Pap test diagnosis at enrollment of atypical squamous cells of undetermined significance was significantly associated with incident SILs. Twenty-seven HIV-infected women received this diagnosis at study entry, but had no colposcopic evidence of a SIL. At 54 months, 39% (95% confidence interval [CI], 20%-59%) of these women had developed biopsy-confirmed SILs vs 25% (95% CI, 19%-31%) of women with a Pap test diagnosed as within normal limits at enrollment (Figure 1, C).

Women who were HPV DNA positive at entry were significantly more likely to develop incident biopsy-confirmed SILs. By 54 months of follow-up, SILs had developed in 10% (95% CI, 5%-15%) of women who were HPV DNA negative at enrollment vs 24% (95% CI, 17%-30%) of women who were HPV positive for a type other than 16 or 18, and 48% (95% CI, 31%-65%) of women who were HPV type 16 or 18 positive. The association between HPV infection and incident SILs was more pronounced in HIV-infected women (Figure 2). At 54 months of follow-up, 7% (95% CI, 1%-13%) of HIV-uninfected and 15% (95% CI, 6%-23%) of HIV-infected women who were HPV DNA negative at enrollment had developed SILs. In comparison, 12% (95% CI, 0%-35%) of HIV-uninfected and 54% (95% CI, 36%-72%) of HIV-infected women who were HPV type 16 or 18 positive had developed SILs by 54 months (Figure 2).

The HIV-infected women were 4.5 times more likely than uninfected women to develop SILs (Table 3). In univariate analysis, other factors significantly associated with incident SILs were HPV DNA status, younger age, and young age at first sexual intercourse. Although sociodemographic features such as race, age, and intravenous drug use history varied across enrollment sites, site was not significantly associated with incident SILs when biological variables such as HPV DNA status, HIV status, and CD4 cell count were controlled for.

Development of SILs was significantly associated with HPV DNA status (Table 3). Only 5 women who never tested positive for HPV DNA developed incident SILs but 3 of these women had only 1 HPV DNA test before developing a SIL. Previously, we showed that HPV infections are more likely to be persistent in HIV-infected vs uninfected women.¹¹ To assess the impact of persistence of HPV infections on incidence of SILs, several different multivariate analyses were performed. In one, we compared women with transient HPV infections (defined as detection of HPV DNA at any visit, but in women not fulfilling the criteria of a persistent infection) with women with persistent HPV infections (defined as HPV DNA detected at 2 or more visits within a 3- to 12-month period) (Table 3). This analysis included HIV status, age, cigarette smoking, and transient and persistent HPV infections. Persistence of HPV infection for types other than 16 or 18 and persistence for HPV types 16 or 18 were significantly associated with incident SILs (adjusted RRs of 7.6 and 11.6, respectively) in this analysis. To determine the existence of a statistical interaction between HIV and HPV infections, an interaction term containing HIV infection and HPV DNA positivity at baseline was introduced into a multivariate analysis that included HIV status, age, cigarette smoking, and HPV DNA status. The interaction term had an adjusted RR of 2.7 (95% CI, 1.5-4.8; P<.001), indicating a significant statistical interaction between HIV and HPV.

In an analysis of HIV-infected women only, we assessed the importance of transient HPV infections, persistent HPV infections with types other than 16 or 18, and persistent HPV infections with types 16 or 18 (Table 4). Transient HPV infections (adjusted RR, 7.4; 95% CI, 1.0-57.4), persistent infections with types other than 16 or 18 (adjusted RR, 8.9; 95% CI, 1.2-66.2), and persistent infections with HPV type 16 or 18 (adjusted RR, 11.0; 95% CI, 1.4-88.7) were all significantly associated with SIL development.

To evaluate the impact of antiretroviral therapy on the incidence of SILs, the multivariate model for HIV-infected women included antiretroviral therapy as a time-dependent variable, and controlled for CD4 cell count at entry, current cigarette smoking, age, and HPV DNA status (transiently infected with any type of HPV, persistently infected with an HPV type other than 16 or 18, and persistently infected with HPV type 16 or 18). Thirty (45%) of 67 HIV-infected women who developed SILs had been taking antiretroviral therapy for at least 6 months at the time the lesion was detected. Controlling for CD4 cell count and HPV DNA status, the incidence of SILs in HIV-infected women taking antiretroviral therapy was not significantly different from that of women not receiving antiretroviral therapy (Table 4).

In 1993, the Centers for Disease Control and Prevention designated invasive cervical cancer as an AIDS-defining illness.¹⁶ Our study provides the first comprehensive analysis of the incidence of SILs, the precursor to invasive cervical cancer, in this population. The incidence of biopsy-confirmed SILs among HIV-infected women was 8.3 per 100 person-years of follow-up vs 1.8 per 100 person-years in uninfected women (P<.001). In HIV-infected women, almost all (91%) of the incident SILs were low grade, and no invasive cervical cancers were identified during more than 800 person-years of follow-up.

The incidence of SILs in HIV-infected women in our study is considerably lower than that reported by a French collaborative group¹⁷ in which SILs were cytologically identified in 21% of 160 HIV-infected women with a normal Pap test result at enrollment during 1 year of follow-up. The lower incidence observed in our study may reflect, in part, the fact that all of the women we enrolled had both a Pap test and a colposcopic examination at entry, reducing the risk for misclassification of disease status. Also, in our study all incident SILs were biopsy-confirmed. Differences between the 2 studies do not appear to be attributable to differences in the use of antiretroviral therapies. In our study, the incidence of SILs in HIV-infected women receiving 1- and 2-drug antiretroviral therapy was not significantly different from that in women who received no therapy. However, it should be stressed that both studies were done prior to widespread use of protease inhibitors.

Because the sensitivity and specificity of Pap tests are similar in HIV-infected and uninfected women, many clinicians use cytology to screen for cervical disease.^{3 ,9 ,18} However, others have recommended that colposcopy be used to screen for cervical disease in HIV-infected women.^{19 - 20} The lack of consensus about how to screen HIV-infected women results from several factors, including the paucity of data about the incidence and natural history of cervical cancer precursors in HIV-infected women. Our finding that most incident SILs in HIV-infected women are low grade, coupled with the fact that no invasive cervical cancers were identified in HIV-infected women during more than 800 person-years of colposcopic follow-up, suggests that current recommendations of the US Public Health Service and Infectious Diseases Society of America, which recommend yearly Pap screening of HIV-infected women after obtaining 2 normal test results, are adequate.^{21 - 22}

Our results also provide new data about the role of HPV infections in the pathogenesis of SILs. Cohort studies investigating the role of HPV in the development of SILs in women in the general population have reported somewhat conflicting results. In a study of women from an STD clinic, Koutsky et al²³ found that the 2-year cumulative incidence of high-grade SILs was greater than 45% in cytologically negative women with HPV type 16 or 18 detected using a dot-blot method. Infection with multiple types of HPV and persistent detection of HPV DNA were significant independent risk factors in that study. In contrast, Rozendaal et al²⁴ reported that only 9% of Dutch women with no cytologic evidence of SILs and who had a high-risk HPV type (eg, 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, or 66) detected using a PCR assay developed high-grade SILs (cervical intraepithelial neoplasia grade 2 or 3, average length of follow-up, 40 months). Ho et al²⁵ reported that, although persistent HPV infections were significant risk factors for incident SILs in a college population, most (29 of 31) incident SILs were low rather than high grade. In our cohort, the incidence of SILs (of any grade) at 54 months was almost 5 times greater in women who were HPV type 16 or 18 positive at entry vs women who were HPV DNA negative. Persistence of HPV infection conferred an increased risk, and most (88%) incident SILs presented as low-grade lesions. This confirms the importance of persistence of HPV infection.²⁶ In our results, the difference in risk magnitude observed for persistent HPV between the results of the univariate and multivariate analyses is likely due to controlling for HPV in the multivariate analysis.

The incidence of SILs was 4.5 times higher in HIV-infected than uninfected women. In part, the increased incidence of SILs reflects the fact that HPV infections are significantly more common, more likely to be persistent, and more likely to include multiple types of HPV in HIV-infected vs uninfected women.^{11 ,13 ,27 - 29} However, even after controlling for the presence and persistence of HPV infections, a significantly higher incidence was observed in HIV-infected women. Previously, we have shown in cross-sectional studies that the relative frequency of latent HPV infections, defined as detection of HPV DNA in the absence of a clinical lesion, and clinically expressed HPV infections, such as genital warts or SILs, is altered by HIV-induced immunosuppression.¹³ In women in the general population, and in HIV-infected women with relatively high CD4 cell counts, latent HPV infections are significantly more common than are clinically expressed HPV infections. However, with increasing degrees of immunosuppression, equivalent numbers of latent and clinically expressed HPV infections are detected.³⁰ The current finding of an increased incidence of SILs in HIV-infected women, even after controlling for HPV status, supports the concept that alterations in viral-host interactions occur in HIV-infected women and affect the natural history of HPV infections.

This study has several limitations, especially with respect to evaluating the impact of antiretroviral therapy on incidence of SILs. Although the use of 1 or 2 antiretroviral drugs in combination was not found to have a significant effect on the incidence of SILs, none of the HIV-infected women in this study were receiving protease inhibitors. Our study was conducted prior to the widespread use of highly active antiretroviral therapy. Triple-drug therapy that includes a protease inhibitor has a significant impact on the incidence of other HIV-associated diseases and may have a similar effect on SILs. Another limitation is that information about antiretroviral drug use, as well as history of STDs, was obtained through interviews rather than through record reviews or drug usage monitoring diaries. Self-reported data may have introduced recall bias in the analysis.

Our findings have important implications for clinicians caring for HIV-infected women. The use of potent antiretroviral agents and effective prophylactic regimens against opportunistic pathogens is having a profound impact on the survival of HIV-infected individuals.^{21 ,31 - 32} Of the HIV-infected women enrolled into the New York Cervical Disease Study, 28% had a biopsy-confirmed SIL at their enrollment visit.³ The current findings indicate that an additional 26% will develop a SIL after 54 months of follow-up. This suggests that as HIV-infected women live longer, most may develop a cervical cancer precursor sometime during the course of their HIV infection. Therefore, it is imperative that these women participate in cervical cancer screening programs.