Immunodeficiency and the risk of cervical intra-epithelial neoplasia 2/3 and cervical cancer: A nested case-control study in the Swiss HIV Cohort Study

HIV-infected women are at increased risk of cervical intra-epithelial neoplasia (CIN) and invasive cervical cancer (ICC), but it has been difficult to disentangle the influences of heavy exposure to HPV infection, inadequate screening, and immunodeficiency. A case-control study including 364 CIN2/3 and 20 ICC cases matched to 1,147 controls was nested in the Swiss HIV Cohort Study (1985-2013). CIN2/3 risk was significantly associated with low CD4+ cell counts, whether measured as nadir (odds ratio (OR) per 100-cell/µL decrease=1.15, 95% CI: 1.08, 1.22), or at CIN2/3 diagnosis (1.10, 95% CI: 1.04, 1.16). An association was evident even for nadir CD4+ 200-349 versus ≥350 cells/ µ L (OR=1.57, 95% CI: 1.09, 2.25). After adjustment for nadir CD4+, a protective effect of >2-year cART use was seen against CIN2/3 (OR versus never cART use=0.64, 95% CI: 0.42, 0.98). Despite low study power, similar associations were seen for ICC, notably with nadir CD4+ (OR for 50 versus >350 cells/µL= 11.10, 95% CI: 1.24, 100). HPV16-L1 antibodies were significantly associated with CIN2/3, but HPV16-E6 antibodies were nearly exclusively detected in ICC. In conclusion, worsening immunodeficiency, even at only moderately decreased CD4+ cell counts (200-349 CD4+ cells/µL), is a significant risk factor for CIN2/3 and cervical cancer.


INTRODUCTION
There is a substantial excess of invasive cervical cancer (ICC) among women infected with HIV (WHIV). 1 This excess has tended to be largest in settings where WHIV have access to combined antiretroviral therapy (cART), but are not well reached by cervical screening. [2][3][4] Studies have also shown a consistent excess of cervical intraepithelial neoplasia grade 2/3 (CIN2/3) in WHIV. 5,6 Sexual behaviors that favor acquisition of HIV are the same that favor the acquisition of high-risk human papillomavirus (HPV) types, which leads to a higher prevalence of HPV in WHIV. 7 In addition, however, HIV-related immunodeficiency is known to unfavorably influence HPV natural history. CD4+ cell count is negatively associated with the prevalence, [8][9][10][11] persistence, 12 and cumulative incidence 13 of HPV infection, as well as with cytological abnormalities 11 and CIN2/3 6,14,15 among WHIV.
The effects of HIV-related immunodeficiency on ICC have been more difficult to establish. Firstly, ICC has been classified as an acquired immunodeficiency syndrome (AIDS)-defining cancer since 1993, which created problems for early linkage studies between people with AIDS and cancer registries. Many prevalent cervical cancers diagnosed at AIDS onset had to be excluded from estimations of relative risk. 16 Secondly, screening can substantially curb the ICC burden. 17 Only in recent years are large studies of WHIV revealing that CD4+ cell count is indeed negatively related to ICC risk. [18][19][20] The dose response relationship between the level of immunodeficiency and the risk of CIN2/3 and ICC remains unclear, however, as does the effect of cART-induced improvements in immunity on the evolution of cervical neoplasia. Thus, we undertook a case-control study nested within the Swiss HIV Cohort Study (SHCS), a nationwide study of HIV-positive persons, to characterize the influence of immunodeficiency and cART use on the development of CIN2/3 and ICC among WHIV.

The Swiss HIV Cohort Study
The SHCS is nationwide prospective cohort that has been enrolling HIV-infected persons aged 16 years or older since 1988 from all five Swiss University hospitals (Basel, Bern, Geneva, Lausanne and Zurich), and two cantonal hospitals (St Gallen and Ticino). 21 Since 1995, interested private physicians have also been enrolling patients. Twenty-eight percent of participants in the SHCS are women. The database used for the current study History of Papanicolaou (Pap) smear was self-reported by SHCS participants at all enrollment and follow-up SHCS visits after April 2000. Pap smear history was censored at the SHCS visit preceding the reference date (see definition below), which should exclude the smear that led to diagnosis of CIN2/3 or ICC.
Written informed consent was obtained from all SHCS participants. The present study was approved by the local ethical committees of the seven SHCS sites and of the International Agency for Research on Cancer.

Ascertainment of incident cases
Cases of CIN2/3 and of ICC are routinely recorded in the SHCS. Additional cases were identified through record linkage with eight cantonal cancer registries. 22,23 Prevalent CIN2/3 and ICC cases were defined as women diagnosed before, or within one month of enrolment into the SHCS and were excluded from analyses. 6 Selection of controls Three control women never diagnosed with CIN2/3 or ICC were matched to each CIN2/3 or ICC case using incidence density sampling. 24 Controls had at least the same length of follow-up as the matched case, and could serve as controls for only one case.
Matching criteria were: 1) SHCS centre; 2) HIV-transmission category (injecting drug users [IDU], heterosexual/other); 3) age at enrollment in SHCS (as close as possible, up to a maximum of nine years difference); 4) calendar year of enrollment in SHCS (as close as possible, up to a maximum of nine years difference). For each case a list of possible controls was drawn up and three controls were selected at random.

Definitions
For each case-control pair, the reference date was defined as the date corresponding to the same length of follow-up as the matched case had at the time of the diagnosis of CIN2/3 or ICC (thus cases and controls were also matched for age and year of reference date). The nadir CD4+ cell count was defined as the lowest ever reported CD4+ cell count prior to the reference date.
cART use was defined as a combination of at least three antiretroviral drugs, including a protease inhibitor or a non-nucleoside reverse transcriptase inhibitor or three nucleosides, including abacavir. Only persons who had used cART for more than one month prior to the reference date were classified as ever users, and cART users were additionally stratified by duration of use from date of cART initiation to reference date (≤2 years versus >2 years). This duration may include periods of interruptions, but these occur in only a small fraction of SHCS participants on cART (5-10%). 25 History of AIDS was defined as a history of any AIDS-defining illness prior to the reference date. Of note, for ICC and corresponding controls, AIDS-defining events were 7 counted only if diagnosed at least 3 months before the reference date, to avoid that ICC itself was the first AIDS-defining illness.
We considered CIN2/3 and ICC to be different biological entities and chose not to combine in any analyses.

HPV serology
For a sub-set of CIN2/3 and ICC cases with available serum samples (79 and 13, respectively), as well as 248 corresponding controls, HPV antibodies were tested in serum samples taken closest in time prior to the reference date. HPV antibodies were tested at the German Cancer Research Center (DKFZ) in Heidelberg, Germany, using multiplex bead-based technology (xMAP, Luminex Corporation, Austin, Texas) 26 including antigens for L1, and E6 proteins of HPV16. Antigens were categorized as antibody positive or negative by applying previously defined antigen-specific cut-off values. 27

Statistical Analysis
Odds ratios (OR) and corresponding 95% confidence intervals (CI) for possible risk factors were computed separately for CIN2/3 and ICC by univariate conditional logistic regression conditioned on matching variables. For CD4+ cell count measures, OR were also calculated per 100cells/µl decrease considering CD4+ cell count as a continuous variable. The OR for cART use adjusted for nadir CD4+ cell count was additionally computed. In analyses stratified by ever versus never cART use, however, OR for nadir CD4+ cell count were computed by unconditional logistic regression with adjustment for matching variables (SHCS centre; HIV-transmission category; age at enrollment; year at enrollment), due to the need to break case:control pairs (cART use was not a matching criteria). Missing data were considered as a separate category in the above models. For smoking and Pap smear history that were unavailable prior to April 2000 and April 2001 8 respectively, conditioning on year at reference date meant that most missing data clustered in the same case:control pairs which were effectively dropped from the estimates for these two variables). Chi-square for trend was calculated for the categories listed in the tables. P values of all statistical tests were two-sided.
Median CD4+ cell counts were calculated separately for never and ever cART users in yearly periods prior to the reference date, restricted to cases and controls who (1) were under active follow-up and (2) had a valid CD4+ cell count, in each yearly period. If more than one measurement was available during any one time period, that closest to the reference date was used. Matching was not retained in the presentation of median CD4+ cell counts. 9

RESULTS
A total of 508 CIN2/3 and 40 ICC cases were identified in SHCS participants, of whom 437 and 32 were identified from the SHCS database, and 71 and 8 additional cases were identified through record linkage with eight population-based Swiss Cantonal Cancer Registries. 22,23 One hundred and forty CIN2/3 and 16 ICC occurring before (or within one month of) SHCS enrollment, as well as four CIN2/3 and four ICC diagnosed more than six months after the last SHCS follow-up date, were excluded, leaving 364 and 20 eligible incident cases occurring during active SHCS follow-up, respectively. Table 1 shows the distribution of the CIN2/3 and ICC cases, as well as their respective controls, by matching variables. For five CIN2/3 cases, only two valid controls were available. A majority of both CIN2/3 and ICC cases were diagnosed after the introduction of cART in 1996 (72% and 65% respectively), and were under active follow-up in the SHCS for less than five years prior to diagnosis (66% and 60%, respectively).
There were no significant associations with history of Pap smear, neither for CIN2/3 nor ICC, but this information was available only for case:control pairs whose reference date was after April 2001 (the date at which this information started to be collected in the SHCS). Nevertheless, findings did suggest frequent lack of screening among ICC cases (37.5%) ( Table 2). There was no association of smoking history with CIN2/3 and, although 75% of ICC cases were current or former smokers compared to 51% of controls, this difference did not reach statistical significance. Likewise, no associations with a history of AIDS were observed for either CIN2/3 or ICC ( Table 2).
Associations of CIN2/3 and ICC with various markers of immunodeficiency are also shown in Table 2 About half of CIN2/3 and ICC cases had started cART before the reference date (Table 3), but there were no significant differences in cART use between cases and controls for either disease. Upon stratification by duration of cART use, however, women with ≤2 years use showed elevated CIN2/3 risks compared to never users, whilst those with >2 years use showed non-significantly reduced risks. After additional adjustment for nadir CD4+ cell count, the OR for >2-years cART use was associated with a significant reduction in CIN2/3 risk (OR=0.64, 95% CI: 0.42, 0.98). A similar association, albeit nonsignificant, was seen for ICC risk (e.g. OR for ≥2years versus never=0.34, 95% CI: 0.05, 2.26) ( Table 3). Table 4 shows associations of nadir CD4+ cell count with CIN2/3 and ICC, stratified by cART use (never/ever). Nadir CD4+ cell count was significantly associated with CIN2/3 risk both among ever users (OR per 100 cell/µL decrease = 1.12, 95% CI: 1.03, 1.21) and never users (1.17, 95% CI: 1.05, 1.30). Nadir CD4+ cell count was strongly associated with ICC risk in never users (OR for <50 versus ≥350 cells/µL= 12.90, 95% CI: 0.38, 436, P for trend=0.004), but the corresponding associations did not reach statistical significance

International Journal of Cancer
This article is protected by copyright. All rights reserved.

11
in ever users (OR for <50 versus ≥350 cells/µL= 6.41, 95% CI: 0.43, 95.1, P for trend=0.26). Figure 1 shows median CD4+ cell counts in yearly periods prior to the reference date in CIN2/3 cases and controls, separately for never and ever cART users. As expected, mean CD4+ cell counts among both cases and controls declined over time among never cART users ( Figure 1A) but were stable in ever cART users ( Figure 1B). Median CD4+ cell count was lower in cases compared to that in controls at most time points prior to the reference date, and the size of this difference did not appear to vary by time period, neither for never or ever users. Indeed, the association between CIN2/3 risk and more historical measures of CD4+ count was similar to that for CD4+ cell count at reference date (e.g.

International Journal of Cancer
This article is protected by copyright. All rights reserved.

DISCUSSION
Our carefully matched case-control study within the SHCS was able to confirm the role of immunodeficiency in the aetiology of CIN2/3 14,15 and ICC 18-20 among WHIV.
Furthermore, we showed that increases in CIN2/3 and ICC risk are already evident even at moderate levels of immunosuppression (200-349 cells/µL), similar to our recent findings in the SHCS for anal cancer, which is also HPV-related. 28 Low CD4+ cell counts at diagnosis were significantly associated with risk of CIN2/3, but nadir CD4+ cell count appeared a more discriminant CD4+ measure of both CIN2/3 and ICC.
We were able to demonstrate a protective effect of cART use on the risk of CIN2/3, but exclusively if the use had lasted 2 years or more. Whereas early studies failed to show a significant effect of cART use on HPV-related outcomes, others have since reported significant decreases in prevalence of HPV infection, 15,29 incidence of cytological lesions, 30 or carcinoma in situ/ICC 4 in medium/long-term users. In our study, the protective effect of cART use on CIN2/3 was strengthened by adjusting for nadir CD4+ cell count, which partially adjusts for the negative confounding of CD4+ cell count by indication to treat.
Indeed, cART tends to be initiated only when CD4+ cell count falls below a certain level, so that non-cART users had higher average nadir CD4+ cell counts than cART users. In our present study, mean nadir CD4+ cell count in cART users was 163 in CIN2/3 and 207 in corresponding controls, versus 293 and 363, respectively among non-cART users.
Previous findings of lack of influence of cART may thus be due to lack of accounting for duration of treatment and CD4+ cell count at cART initiation.
As more than half of cases and controls had never used cART at the reference date, we were able to show the effect of nadir CD4+ cell counts on CIN2/3 both in ever and never users of cART. Two previous studies on this topic showed that CD4+ cell counts at diagnosis were significantly associated with CIN2/3 risk, but in non-cART users only. 14,15 These findings suggest that cART use can only partially eliminate the negative influence of

International Journal of Cancer
This article is protected by copyright. All rights reserved. 13 a low nadir on CIN2/3, which differs from other AIDS-defining malignancies like immunoblastic non-Hodgkin lymphoma 31 and Kaposi sarcoma 32 in which cART can rapidly lower cancer risk. Hence the benefits of cART on ICC may be limited by the nonreversibility of certain pre-invasive stages.
Cervical cancer screening is a pre-requisite for diagnosis of CIN2/3, and its treatment has undoubtedly curbed incidence rates of ICC in the SHCS. Even in the SHCS, however, screening has been sub-optimal with 20-30% of WHIV in the control groups reporting no Our limited data on anti-HPV16 L1 seroprevalence demonstrates the high burden of HPV in SHCS women. Thirty percent of women in the control group were seropositive for anti-HPV16 L1, which compares to only 10% among mixed-sex controls from a large European-wide population-based study using the same serological assay. 33 Furthermore, whilst anti-HPV16 L1 is considered the best marker of cumulative exposure to HPV16 infection, the reported seroprevalence is certainly an under-estimate as a substantial fraction of HPV infections are known not to elicit L1 seroconversion. 34 Antibodies to HPV16 E6 were seen in ICC, but were almost absent, not only in controls, but also in CIN2/3, confirming them as a marker specific to late stages of HPV malignant transformation in the anogenital tract. 33 Nevertheless, only 15% of ICC cases were seropositive for antibodies against HPV16 E6, which compares to 32% seen in serum samples taken close in time to ICC in the HIV-negative population. 35 14 An association, albeit non-significant, between smoking and ICC risk in WHIV is consistent with the sexual behavior-adjusted findings from a large pooled analysis of ICC in HIV-uninfected women 36 and with the significant association seen for anal cancer in the SHCS 28  Furthermore, the completeness of recording of CIN2/3 in the SHCS is known to differ over time and between cantons. This is one reason that a nested case:control design with careful matching is probably more appropriate than an incidence rate approach. Although ICC cases were few and gave us limited statistical power to achieve significant associations, previous data are very scant. Indeed, the historically sub-optimal level of cervical screening and treatment in the SHCS meant that the observed rate of ICC was higher than in well-screened cohort studies of WHIV. 6 In conclusion, our data suggest that worsening immunodeficiency, even at only moderately decreased CD4+ cell counts (200-349 CD4+ cells/µL), is a significant risk factor for CIN2/3 and cervical cancer. Thus, the earliness at which HIV infection is

International Journal of Cancer
This article is protected by copyright. All rights reserved.
2 Abbreviations: cART, combined antiretroviral therapy; CI, confidence interval; CIN, cervical intra-epithelial neoplasia; ICC, invasive cervical cancer; OR, odds ratio. a conditioned upon matching variables; b conditioned on matching variables and adjusted for nadir CD4 categories in Table 2.

International Journal of Cancer
This article is protected by copyright. All rights reserved.