The risk of cervical cancer associated with specific types of human papillomavirus: A case–control study in a UK population

Authors

  • Ned G. Powell,

    Corresponding author
    1. HPV Research Group, Department of Obstetrics and Gynaecology, School of Medicine, Cardiff University, Cardiff, United Kingdom
    • HPV Research Group, Department of Obstetrics and Gynaecology, School of Medicine, Cardiff University, Cardiff, CF14 4XN, United Kingdom
    Search for more papers by this author
    • Tel.: 0044-(0)2920-744742, Fax: 0044-(0)2920-744399

  • Sam J. Hibbitts,

    1. HPV Research Group, Department of Obstetrics and Gynaecology, School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author
  • Adam M. Boyde,

    1. Department of Pathology, University Hospital of Wales, Heath Park, Cardiff, United Kingdom
    Search for more papers by this author
  • Robert G. Newcombe,

    1. Department of Primary Care and Public Health, School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author
  • Amanda J. Tristram,

    1. HPV Research Group, Department of Obstetrics and Gynaecology, School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author
  • Alison N. Fiander

    1. HPV Research Group, Department of Obstetrics and Gynaecology, School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author

Abstract

Mounting evidence supports incorporation of HPV testing into cervical screening; however, the optimal test format and target population have yet to be confirmed. Assessment of the potential benefits of type-specific testing requires estimation of the risk associated with infection with individual HPV types. However, the risk posed by individual HPV types may be population specific and influenced by cervical screening practice. The existing data on HPV type-specific risk is derived largely from unscreened populations. Our study addressed the lack of data on HPV type-specific risk in cytologically screened populations using a case–control study of 262 invasive cervical cancers diagnosed in Wales between 2000 and 2006, and 8,428 controls who attended for cytological screening in 2004. The analysis showed that the odds ratios (ORs) for infection with HPV 16 and 18 are considerable; 2770 (95% CI 1050–7320) and 950 (95% CI 330–2740), respectively, and that the OR for other oncogenic types are in general considerably less (ranging, where quantified, from 20.2 to 386 in the same population). The effect of age on OR associated with particular HPV types was also assessed; this indicated that infection with a high-risk HPV in women older than 40 years was associated with an approximately 30-fold increased risk of invasive cervical cancer relative to women younger than 40 years. These results indicate that there is significant prognostic information associated with knowledge of HPV type.

Human papillomavirus (HPV) is present in the vast majority of invasive cervical cancers and is considered the primary cause of this disease.1 In the United Kingdom, vaccination to prevent infection with two oncogenic types of HPV (16 and 18) commenced in September 2008, with vaccine being given to 12- to 13-year-old girls born between September 1995 and August 1996. A 2-year catch up program for women aged up to 18 years is also underway. In clinical trials, this vaccine has shown 100% efficacy for prevention of both persistent infection and high-grade cervical intraepithelial neoplasia associated with HPV 16 and 18 in HPV naïve women.2

It is likely that cervical screening in the United Kingdom will change substantially in the near future, with two main factors driving this change. The first is the growing body of evidence showing that HPV testing has greater sensitivity than cytological screening for cervical cancer and precancer.3–5 The second is the expected reduction in the positive predictive value of cytological screening after introduction of vaccination; this is inevitable as the number of false positives will remain the same while the number of true positives will decline. However, there remain many questions regarding how best to use HPV testing. These include whether genotyping for specific HPV types is worthwhile, and what is the significance of a positive HPV test result in different age groups.

The objectives of our study were to investigate the potential benefit of HPV genotyping by assessing the risk of cervical cancer associated with infection with individual HPV types, and to determine the effect of age on estimates of type-specific risk, in a cytologically well-screened UK population.

It was necessary to perform these analyses on a population specific basis as a recent analysis of contemporary archival cervical cancers from the Welsh (UK) population indicated that HPV 16 and 18 were present in 82% of cancers; which is a higher proportion than previously reported.6 Estimates of the risks associated with infection with specific HPV types are currently based mainly on case–control studies from populations with smaller proportions of HPV 16/18 associated disease,7 which suggests that such estimates might not accurately represent the risks associated with HPV infection in UK populations.

To address these objectives, the risks posed by infection with specific HPV types were determined using a case–control analysis of HPV infection in recently diagnosed cervical cancers and in women who underwent cervical screening in Wales (UK).

Material and Methods

Study design

Data derived during two previous investigations, one examining HPV types associated with contemporary cervical cancers and the other assessing HPV prevalence in women attending for cervical screening, were combined in a case–control analysis to determine the risks associated with infection by specific types of HPV. Both studies were conducted with appropriate Ethics Committee approvals. To identify HPV types associated with cervical cancers, archival material (tumor biopsy or surgical excision sample) was collected from 453 cases diagnosed in Wales between January 1, 2000 and September 1, 2006.6 HPV prevalence was also investigated in 10,000 consecutive women (nonprobability convenience sample) who underwent cervical screening in South Wales in 2004.8 These studies used the same HPV typing methodology and were conducted in the same laboratory.

DNA was extracted from liquid-based cytology samples by digestion with proteinase K, and from formalin-fixed paraffin embedded material using Qiagen DNeasy Blood and Tissue kit without xylene washes.9 The quality of DNA extracted was assessed by PCR for the human β-globin gene. HPV typing was performed by PCR using GP5+/6+ primers followed by enzyme immunoassay,10 as previously described.11 HPV typing was performed in two stages: the first stage used a cocktail of probes for 14 high-risk (HR) HPV types, PCR was then repeated on positive samples and the amplified DNA typed with individual HR probes. This assay detects DNA from 14 HR HPV types: 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66 and 68.

The investigation of HPV prevalence was not powered to detect differences in specified parameters, rather it was intended to be sufficient to provide a robust overview of HPV infection in the population investigated. For the archival sample, study size was a compromise between collecting a substantial number of cases and retaining relevance to the contemporary population. The number of tumors investigated was sufficient to ensure that the amplitude of the 95% confidence intervals would not exceed 6% when the percentage attributable to a specific type was less than 20%, and would not exceed 5% when the percentage attributable to a specific type was less than 10%.

Statistical methods

The cancer and control series available differed greatly in age: the mean and range for age at last birthday were 51.8 (22–97) years for 280 cases and 38.3 (17–80) for 9,305 control women. Accordingly, the main analysis is restricted to 262 cases and 8,428 controls in the age range 22–80, with mean ages 49.4 and 40.1 years. The rationale for excluding younger and older women was to restrict the analysis to the window of overlap between the cases and controls, as a first step toward controlling for age. Crude odds ratios were calculated to characterize the association of cancer with each HPV type separately, disregarding the presence of coinfections. Age-adjusted odds ratios were then calculated by logistic regression, adjusting for age as a categorical covariate using approximate 5-year age groups (age groups: 22–25, 26–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74 and 75–80). Confirmatory alternative analyses were constructed by assembling an individually age-matched series comprising 226 case–control pairs. These analyses gave similar results but were much less powerful, with much wider confidence intervals.

Results

Description of cases

The cases were derived from a series of 453 cervical cancers diagnosed in hospitals across Wales between January 1, 2000 and September 1, 2006. DNA was successfully extracted from 297 blocks, as indicated by PCR for the human β-globin gene. Five cases were excluded as tumor was not observed in both sections flanking those taken for DNA analysis. A further 12 cases were excluded as duplicate submissions (i.e., the same case submitted by different hospitals). This left a total of 280 cases. Pathology was reported as: squamous cell carcinoma for 222 cases, adenocarcinoma for 35 cases, adenosquamous carcinoma for 12 cases, neuroendocrine/small cell for 5 cases, carcinosarcoma for 2 cases, adenoid basal for 2 cases and undifferentiated carcinoma for 2 cases. DNA from HR HPV types was identified in all but seven cases (comprising four squamous cell carcinomas, two adenosquamous carcinoma and 1 small cell/neuroendocrine carcinoma). DNA from more than one HR HPV type was identified in 33 cases. For calculation of OR, cases were restricted to 262 women aged 22–80 years. A detailed analysis of HPV infection in these cases has been published.6

Description of controls

The control dataset was derived from 10,000 women who attended for cervical screening in South Wales in 2004. Amplifiable DNA was not obtained from 688 samples, as indicated by PCR for the human β-globin gene. The age of the woman was not known for seven samples. The final number of controls included was 9,305. Mean age was 38.3 years. Cytology results for the 9,305 were: normal 8,519 (91.55%), borderline 435 (4.67%), mild dyskaryosis 129 (1.39%), moderate dyskaryosis 43 (0.46%), severe dyskaryosis 47 (0.51%), severe dyskaryosis/invasive 3 (0.03%), glandular neoplasia 3 (0.03%) and inadequate 126 (1.35%). HR HPV DNA was present in 1,043 samples, and 603 samples contained DNA from more than one HPV type. A detailed analysis of HPV infection in these samples has been published previously.8

HPV type-specific risk

The distribution of HPV types in cases and controls as a percentage of the total number of HR HPV infections is shown in Figure 1. From Figure 1, it is apparent that the distribution of HPV types in cervical cancers is not simply a reflection of their distribution in the control population. Figure 1 reinforces the observation that HPV 16 and 18 are the most common HPV types present in cervical cancers and also suggests that this is due to greater oncogenic potential for HPV 16 and 18, as well as to greater prevalence in the screening population for HPV 16.

Figure 1.

HPV type distribution in cases and controls. Showing the proportion of high-risk HPV infections attributable to individual types (as a proportion of the total number of infections in cases or controls; cases, n = 308, controls, n = 2037).

Table 1 shows the results of the case–control analysis to quantify the oncogenic potential of individual HPV types in this population. Eighteen cases aged over 80 years and 877 controls aged less than 22 years were excluded. HPV 51 and 68 were absent from the case series, but for all other types, significant positive associations were found. In most analyses, adjustment for age caused a significant increase in OR (e.g., for HPV 16 from 774 to 2,770) resulting from the decline in HPV prevalence and increased incidence of cervical cancer with increasing age. It can be seen that the strongest relationships between HPV and invasive cancer were observed for infection with HPV 16, 18, 45, 33 and 31; with point estimates for OR ranging from 2,770 to 115.

Table 1. Association of cervical cancer with HR HPV types
inline image

There were significant differences in the prevalence of multiple HPV infections, and in the apparent relationships between HPV types, between cases and controls. Data on multiple infections are shown in Table 2. Among women who were HR HPV positive, multiple infections were significantly more common in controls than in cases (χ2: p < 0.0001).

Table 2. Multiplicity of infections in 280 cases and 9,305 control women
inline image

The relationship between HPV 16 and/or 18 with other HPV types was assessed and the strength of association quantified as OR. The relationship between HPV types 16 and 18 together and the presence of any other type was positive and generally significantly in the control series (OR range: 4.41 [HPV 59] to 72.1 [HPV 31]), but there was a strong inverse relationship in cancer specimens (OR range: 0 [HPV 35, 56, 59 and 66] to 0.105 [HPV 39 and 52]). Further analysis indicated that this effect was primarily attributable to HPV 16, which in cases was more likely to be present as a single infection.

To investigate the effect of age on the relationship between a positive HPV test and cancer, the odds ratios for women 22–40 and 41–80 years were compared. This age division was chosen as it ensured near equal distribution of the referent group between the subgroups. ORs were calculated by logistic regression including adjustment for age as a categorical covariate within subgroups, for HPV 16 and 18 separately and together, and for any HR HPV. The results of this analysis are shown in Table 3. It is clear that the strength of the relationship between cancer and HPV infection increases markedly with age. The odds ratios are significantly increased (by approximately 30-fold) in women older than 40 years relative to women younger than 40 years.

Table 3. Age-adjusted odds ratios for relationship of cancer to high-risk HPV type, compared between age groups 22–40 and 41–80 years
inline image

The relationship of the mean number of HPV types to age was considered; there was a positive correlation in cases (r = +0.13, 95% CI +0.01 to +0.25, p = 0.03), but an inverse relationship in controls (r = −0.09, 95% CI −0.16 to −0.02, p = 0.009). To further investigate the relationship between age and HPV type, the mean age of women with a particular type was compared with mean age of women without that type. This analysis indicated that HPV 16, 18 and 45 tended to occur in younger women. To explore this effect in more detail, women were divided into four groups: (a) positive for HPV 16, 18 or 45 only; (b) positive for other HPV types only; (c) positive for HPV types included in groups (a) and (b); and (d) HPV negative women (Table 4). For cases, the mean ages of women in groups (a) (n = 223, mean = 48.0) and (b) (n = 27, mean = 61.6) were significantly different (t-test: p < 0.001). Multiple infections were also significantly more common in cases in group (b) (mean, 1.41) than in group (a) (mean, 1.06) (p < 0.001).

Table 4. Summary statistics for age for four groups defined by HPV types present
inline image

Discussion

This is the first study to assess the risks associated with infection by specific HR HPV types in a UK population. The key findings of this investigation were: the substantial variation in risk of invasive cervical cancer associated with different HPV types; that the association between HPV infection and cancer is stronger in older women; and that the mean age of diagnosis for women with cancers associated with HPV 16/18/45 was significantly less than for cancers associated with other HPV types. Point estimates of OR varied from 2,770 for HPV 16 to 20.2 for HPV 39. HPV 16 and 18 clearly conferred much greater risk than the other HPV types, whereas types 45, 31 and 33 appear to form a second tier of oncogenic types. However, the data on age at presentation suggests that HPV 45 might be more appropriately grouped with 16 and 18 rather than 31 and 33 years. This leads to the conclusion that there is significant prognostic information associated with knowledge of HPV genotype. This should be considered in addition to the other potential benefits of type-specific testing of HPV positive women, which include assisting identification of type-specific persistence and facilitating monitoring of the long-term effect of HPV 16 and 18 vaccination.12

The limitations of our study relate primarily to the use of a screening population as a source of controls. The use of a screening population is relevant as it constitutes a self-selected population; despite this, it is likely to give a reasonable representation of the distribution of HPV types among women in Wales. Screening coverage in Wales for April 1, 2004 to March 31, 2005 was 75.6% (adequate result in the last 5 years).13 Screening uptake has been shown to correlate with socioeconomic status (SES)14; however, neither HPV prevalence nor type distribution appear to correlate with SES in the study population.8 Hence, there is no reason to suspect bias due to lack of inclusion of women of lower SES. Some studies have observed a positive correlation between screening uptake and age,15 which would suggest that older women may be overrepresented in our sample relative to an ideal screening population, but due to the adjustment for age this would not effect estimates of OR.

The anonymization procedure for controls did not allow for information on subsequent diagnoses to be linked to the sample used for HPV testing. Consequently, it is not possible to absolutely guarantee that the control samples did not include some women subsequently diagnosed with cancer. However, in Wales, in the year 2004/2005, 206,000 women attended for screening13 while 168 cervical cancers were diagnosed.16 If all these diagnoses were made as a result of a positive screening test (which is highly unlikely), then this would result in inclusion of (168/206,000) × 9,305 cervical cancers in 9,305 controls. This equates to 7.6 of 9,305 controls being possibly misclassified, which would not introduce an appreciable error. Finally, these studies relate to a relatively stable population under specific local conditions, hence caution should be exercised in generalizing these results to other populations.

An additional methodological consideration was the use of HPV negative cases as the referent group. This is standard practice but is somewhat artificial, as it is possible that “HPV negative” cases merely lack the region of HPV DNA targeted by the assay used in our study. This group does however constitute an invariant referent group and so allows meaningful comparison of the relative oncogenicity of different HPV types; however, it is likely to result in less meaningful assessments of the absolute levels of risk.

Several previous studies have investigated HPV type-specific risk, including the seminal study by Munoz et al.7 on epidemiological classification of HPV types. The data presented differ from previous analyses in a number of respects. These differences probably arise from differences in study population, and in methodology (i.e., use of an age-matched population). In the study by Munoz et al., subjects were drawn mainly from populations in Africa, South America and Asia, which did not benefit from cervical cytological screening and may have a higher frequency of immune suppression. The absence of screening could contribute to differences in the prevalence of HPV 16 and 18, as screening may identify and treat more slowly developing lesions with less tendency to progress. This may effectively enrich for more aggressive, faster developing interval cancers, which are more likely to be associated with HPV 16 and 18.17, 18 It is certainly the case that the distribution of HPV types identified in invasive cancers differs from that observed in high-grade cervical intraepithelial neoplasia,19 with non-HPV 16, 18 and 45 types present in a greater proportion in precancerous lesions. Screening may also alter the proportion of squamous cell carcinoma relative to adenocarcinoma. Cytological screening is known to be relatively insensitive for adenocarcinoma,20 partly due to the difficulty of sampling cells from within the endocervix where this disease develops21 (although high-quality screening does reduce the incidence of adenocarcinoma22). As HPV 16 and 18 are somewhat more common in adenocarcinoma (79.7% in SCC compared to 91.5% in ADC, in this population6), a change in the ratio of squamous to adenocarcinoma could potentially contribute to an increase in the proportion of cervical cancers attributable to HPV 16 and 18 infections.

The other main finding of our study was that the mean age of women with cancers associated with HPV 16/18/45 was significantly less than the mean age of women with invasive cancers attributable to other HR HPV types. A similar observation has recently been made in an American study, but only for types 16 and 18.23 There are several possible explanations for this observation, including lower levels of oncogene expression in the less carcinogenetic types, and possible greater dependence on long-term exposure to carcinogens in the immediate microenvironment.23 The earlier development of HPV 16/18/45 associated cancers may have significant implications for cervical screening in populations vaccinated against HPV 16 and 18. In populations where vaccine uptake in the target group (12- to 13-year-old girls) is high, and ultimately confers significant reduction in the prevalence of HPV 16 and 18, this would lead to a decrease in the number of early occurring cancers. It might then become possible to raise the age at which cervical screening starts (women are invited to attend cervical screening from the age of 20 in Wales and Scotland, but from 25 in England). The age at which screening should commence in the United Kingdom has recently been debated24 and HPV vaccination could eventually reduce the need to screen women in the 20- to 25-year age bracket.

The observation that multiple infections were significantly more common in cases with non-HPV 16/18/45 infections relative to cases positive for HPV 16/18/45 is interesting, and might suggest that women who are unable to clear infections with non-HPV 16/18/45 types and who subsequently develop cancer, may be less able to clear HPV infections in general. This could arise from a specific immune defect or generally weaker immunity.

The main implication of this data is that there is appreciable prognostic value in knowledge of HPV type, and in genotyping for HPV 16 and 18 in particular. There is mounting evidence that separate detection of HPV 16 and 18 in cytologically normal women might be useful to stratify risk25 and HPV testing has been suggested for primary screening to prevent cervical cancer in women aged 30 years or more.26 The finding of relatively modest OR for non-HPV 16/18 HPV types is relevant, and could be explained in part by the influence of cytological screening. This would be consistent with the high ORs reported for non-HPV 16/18 types in unscreened populations by Munoz et al.7 and the low estimates of risk associated with non HPV 16/18 in a prospective study in a very well-screened population25 (which followed 20,810 women for 10 years and showed cumulative incidence rates for CIN3+ of 17.2%, 13.6% and 3% for women positive for HPV 16, 18 or non-HPV 16/18 HR HPV, respectively). It is worth noting that several of the commercial HPV tests marketed for high-throughput screening offer the facility to identify DNA from a pool of HR HPV types, and then reflex test to determine specifically whether HPV 16/18 DNA is present.

Acknowledgements

We thank the assistance of Cervical Screening Wales, in particular, Dr. Hilary Fielder and Ms. Helen Beer, in assessment of HPV infection in the screening population.

Ancillary