HPV detection predicts CIN3 but not lesser abnormality
The relationship between HR-HPV infection at entry and low-grade abnormality at the next smear was weaker in this study (OR 1.4, CI 0.8–2.5) than in several other cohorts (Liaw et al, 1999
),(Kjaer et al, 2002
), (Schlecht et al, 2001
). This difference is at least partly due to the longer screening intervals in Manchester, which gave most transient HPV infections and their associated lesions time to disappear. In the Portland cohort with a policy of annual screening, the prevalence of HPV infection at entry among women who subsequently developed ASCUS or LSIL declined from 60% for screening intervals of 17 months or less to 24% for intervals of 3 years or longer (Liaw et al, 1999
). Cytologically normal women with HPV infection at entry also had a higher risk of dysplasia in smears taken after a shorter interval than in later smears in other cohorts (Schlecht et al, 2001
; Kjaer et al, 2002
). In addition, our lesser abnormality category included some mildly abnormal smear reports that would have been excluded by the rigorous pathological review conducted in these three cohorts. The combined prevalence of HR-HPV at diagnosis was 64% for ASCUS or worse smears in Portland (Sherman et al, 2003
) and only 44% for all abnormal smears in our study (). Any relationship between HPV infection and later risk of low-grade lesions was further reduced in Manchester by the exclusion from our lesser abnormality category of smears that were followed within 2 years by histological diagnosis of CIN3+, most of which contained HPV.
Sensitivity of HPV testing in primary screening
Our sample collection protocol was not designed for PCR analysis and was unsatisfactory by modern standards. β
-Globin could be amplified in only 89% of samples, due probably to the inhibitory effect of the wooden spatula tip, which remained in the sample at room temperature for several weeks before decanting and freezing. Among those that were β
-globin positive, however, these sampling and assay methods detected HR-HPV in 82% (257 out of 314) of the CIN3+ cases diagnosed at an abnormal entry smear (). This is similar to the average sensitivity of 84% for HSIL detection by HPV assay reported in a recent meta-analysis of published studies (Arbyn et al, 2004
). A second spatula was taken before histological diagnosis from 21 of the 57 women with prevalent CIN3+ whose initial spatula was HR-HPV negative, and 13 (62%) were HR-HPV positive. These statistics suggest that about 95% of CIN3s were HPV positive, but this rate would presumably have been higher with the lower detection threshold of modern PCR methods. Regular HPV testing of sufficient sensitivity would thus detect HR-HPV in almost all CIN3 cases (Nobbenhuis et al, 1999
; Cuzick et al, 2003
). The right-hand part of shows that the prevalence of HR-HPV at entry among cytologically normal women in whom CIN3+ was diagnosed at the next routine smear was 50% (42 out of 84). More sensitive HPV testing would thus detect the majority of CIN3+ cases several years before they become cytologically abnormal. In the Portland study the prevalence of HR-HPV at entry in cytologically normal women was 73% in those who developed CIN3+ within 45 months and 38% in those diagnosed later (Sherman et al, 2003
). Women with incipient CIN3 usually shed HPV persistently for several years before becoming cytologically abnormal (Nobbenhuis et al, 1999
), so cases detected by HPV screening would still be HPV positive on retesting, and would be referred for colposcopy irrespective of the rescreening interval.
The sensitivity of our HPV assay must depend on viral load, and would be lower in women with normal cytology. Using MY-Gold instead of MY-Taq polymerase increased the HR-HPV detection rate by 54% in women with normal cytology, but only by 2% in women with HSIL or cancer (Castle et al, 2002
). Our HPV prevalence estimates () may thus be substantially too low at each age.
Lifetime HPV and CIN3 rates
The relationship between HPV prevalence based on a single sample and lifelong risk of infection is not known. Persistent or recurrent infection is more likely to be detected than a transient episode due to the length-biased sampling effect. The cumulative HPV rate was higher than the cross-sectional rate by a factor of 1.6 among American students sampled weekly on 10 occasions (Wheeler et al, 1996
), and by about two-fold among Brazilian women sampled four times at 4-monthly intervals (Schlecht et al, 2001
). The proportion of younger women in our cohort who will acquire an HR-HPV infection during their lifetime is thus likely to be at least twice the prevalence of 18% seen in women aged 20–24 (). At current rates 9% of the cohort will eventually develop CIN3 (), although this may be an underestimate, as both CIN3 and HPV rates appear to have increased over the following decade. More than 25
000 women were recruited to the ARTISTIC trial of primary HPV screening between 2001 and 2003 through many of the same clinics in and around Manchester. HR-HPV prevalence by Hybrid Capture II was 33% at age 20–29, 15% at 30–39, 9% at 40–49 and 6% at 50–69 (Kitchener et al, 2004
), more than double the rate we observed in the same population about 12 years earlier at each age (). This large increase seems unlikely to be due entirely to differences in HPV detection sensitivity. The CIN3 incidence rate in England and Wales at age 20–24 doubled from 1.2 per thousand in 1990 to 2.4 per thousand in 2000, and the lower prevalence of cervical abnormality in the retrospective cohort compared with the prospective cohort (: OR 0.88, P
<0.05) is also consistent with a continuing increase in HPV prevalence during the 1990s.
Transiently detectable HPV infection is generally regarded as innocuous, but our long-term follow-up of 232 women with HR-HPV at entry shows a cumulative CIN3+ diagnosis rate of 16% after 10 years (). A lower CIN3 rate following HPV detection was seen in the Portland study, where the cumulative risk of CIN3 or cancer diagnosis among women with HR-HPV infection and normal cytology at entry was only about 6% after 10 years' follow-up. The risk in the Portland cohort was even lower among those who were cytologically abnormal (ASCUS or worse) at entry after the initial lesion had regressed or been treated (Sherman et al, 2003
). Only one CIN3 was diagnosed in the Portland study in smears taken between 21 months and 10 years after entry among 417 HR-HPV positive women with an abnormal entry smear compared with 38 cases among 2562 women with HR-HPV infection but normal cytology at entry (OR 0.16, P
=0.03). This reduction in risk supports the long-standing belief that even biopsy of HPV lesions may reduce the subsequent risk of CIN3 (Koss et al, 1963
). Cervical ablation substantially reduced the risk of subsequent CIN or cancer when this was a routine procedure in cytologically normal women (Vonka et al, 1984
). This protective effect, combined with an annual screening policy under which a substantial proportion of HR-HPV-infected women will eventually be detected with dysplasia, may account for the lower overall CIN3 risk among infected women in the Portland study. The increased CIN3 risk among women in Manchester with a history of cervical abnormality (OR 3.0: ) presumably reflects their high probability of having an HR-HPV infection, and perhaps also a lower probability in Britain than in the US that their dysplasias were biopsied or destroyed.
Cytological regression of CIN3
The CIN3 prevalence in smears taken more than 5 years after a normal smear appears to have reached a plateau at about 1% ( and ). The initial linear increase in CIN3 prevalence with increasing screening interval shows that CIN3 often remains detectable for at least 5 years, but this subsequent plateau suggests that most cases eventually regress cytologically. The rising prevalence thus represents a balance between increasing CIN3 incidence and cytological regression. The inference that most CIN3s are only transiently detectable is supported by the low prevalence of CIN3 in women aged 45 or older with no record of a previous smear, which was only 0.5% (; 11 out of 2087). A proportion of these women would have had unrecorded smears, but the CIN3 incidence rates in suggest that 8% of women in this cohort develop CIN3 by age 45. Most earlier CIN3s must therefore have regressed cytologically by the time these unscreened women entered our study. Even in the high-risk rural population of Costa Rica, the prevalence of persistent HSIL at initial recruitment was less than 1% above age 45 (Herrero et al, 2000
). Reversion to cytological normality in women with underlying carcinoma in situ
was observed directly in the prospective study of Dutch women referred for colposcopy with abnormal smears, in which six of the eight women who regressed to normal cytology and colposcopy but remained HR-HPV positive had occult CIN3 (Nobbenhuis et al, 1999
The underlying risk in women who choose not to be screened may differ between different populations. The CIN3 rate at entry in women with no previous smear history was similar to that in the screened population both in our cohort () and in a Norwegian study (Forsmo et al, 1996
), and the rate at their second smear was slightly less than in previously screened women (). A much lower CIN3 rate was observed at the second smear than at entry in a national study of inadequately screened low-income American women (Sawaya et al, 2003
), but this may reflect confounding between screening interval and number of negative smears. Above age 30 both CIN3 and cancer were slightly less common in unscreened women in the American study (Sawaya et al, 2003
) than in our cohort (), and CIN3 prevalence after a negative smear was similar to the rate we observed for screening intervals of under 2 years (0.35%: ).
National data linkage
The underlying age-specific incidence rate of new cases of CIN3 in Manchester shown in could be estimated in the same way in any population with routine screening records linked to histology records. This could now be done for the whole of Britain, as the computerised call–recall system introduced in 1988 includes details of almost all cervical smears for the whole population aged 20–64. Individual women's smear records are already linked to histological diagnoses of CIN3 in the computer systems of the cytology laboratories that provide these data. Linkage to cancer registrations and deaths would provide CIN3 incidence as well as cancer incidence and mortality rates by screening interval and age in all regularly screened women, complemented by cancer rates in the inadequately screened remainder of the population. Sasieni and Cuzick (2001)
recommended that such audit should be established as a routine procedure. The files could also be linked to small area statistics such as deprivation index that can be derived from address postcodes. The files would be anonymised after linkage, so informed consent would not be needed for such analyses.
Progression to carcinoma in situ and cancer
Three observations within our cohort reflect the distinction between the majority of HR-HPV infections that resolve within a year or two and the minority that persist and frequently progress to CIN3 (Nobbenhuis et al, 1999
). First, the prevalence of CIN3 increases linearly with time since the last normal smear, at least up to 5 years, while that of lesser (low-grade) abnormality is almost independent of screening interval (). Second, there is a very strong relationship between HR-HPV infection at entry in cytologically normal women and the risk of CIN3+ at their next smear (OR 17.2, P
<0.001), but a much weaker and nonsignificant relationship for abnormalities other than CIN3+ (OR 2.3 for CIN2 and 1.4 for lesser abnormality). Third, the much lower incidence of CIN3 in women aged under 25 despite their higher HPV prevalence compared with those aged 25–34 () indicates a substantial delay between initial HPV infection and the development of carcinoma in situ
. In a case–control study nested within the cohort (Deacon et al, 2000
), the number of sexual partners was the main risk factor for detectable HPV infection, but progression to CIN3 diagnosis among HPV-infected women depended more on early age at first intercourse, providing further evidence that carcinoma in situ
often originates in a much earlier HPV infection.
An increase in prevalence of CIN2 and CIN3 but not of lesser abnormality with increasing interval since the last normal smear was reported in Icelandic women, but data for CIN2 were not presented separately (Sigurdsson and Adalsteinsson, 2001
). There are several epidemiological differences between CIN2 and CIN3. The CIN2
CIN3+ ratio was 1.5 (19
13) in smears taken within 18 months of a normal smear and 0.5 (211
415) for longer intervals (P
<0.005). CIN2 is also commoner in young women. The CIN2:CIN3+ ratio fell from 1.1 (58
53) below age 25 to 0.5 (142
308) at older ages (; P
<0.001). Persistently detectable HPV preceding cytological abnormality is common in the development of CIN3 and cancer but not of CIN2. In a prospective cytological and colposcopic follow-up of women in the Netherlands with abnormal smears (Nobbenhuis et al, 1999
), the proportion of incident cases that were preceded by persistent HR-HPV infection was 95% (98 out of 103) for CIN3, but only 27% (eight out of 30) for CIN2 and 10% (six out of 63) for CIN1. HR-HPV was detected in the preceding normal smear in 50% (: 42 out of 84) of CIN3s and in 13% (: two out of 16) of CIN2s in our cohort (P
<0.01). Despite disagreement between pathologists over the classification of individual cases, these objective differences between CIN2 and CIN3 support the view that most CIN2s are biologically and clinically indistinguishable from the lower grade HPV lesions that usually regress (Syrjanen et al, 1992
). Among women with HPV infection, these authors observed progression or recurrence in 81% of CIN3s, 24% of CIN2s and 17% of CIN1s (Syrjanen et al, 1992
). The weak increase in CIN2 prevalence with increasing screening interval () confirms that some cases are persistent, perhaps because they harbour undetected carcinoma in situ
or because they are precursors in its progression. For clinical purposes, the inclusion of CIN2 with CIN3 in the HSIL category of the Bethesda classification may therefore be justified, although a repeat smear would avoid colposcopy and biopsy in the transient majority of CIN2s (Syrjanen et al, 1992
). Epidemiological studies in which CIN2 and CIN3 are combined are unnecessarily uninformative, however, particularly in younger women or if the screening interval is 18 months or less. Clonality, HPV integration, loss of heterozygosity and patterns of gene expression are likely to reflect different stages of carcinogenic progression (Evans and Cooper, 2004
), and results for CIN2 and CIN3 should also be presented separately in cellular and molecular studies. It is perhaps unfortunate that the earlier categories of severe dysplasia and carcinoma in situ
were superseded (Syrjanen et al, 1992
), despite difficulties of reproducibility (Sherman, 2003
The age-specific CIN3 incidence pattern shown in suggests a simple model for the relationship between CIN3 and invasive cancer. Before the introduction of the national screening programme in 1988, cervical cancer death rates in Britain rose sharply up to about age 45, then remained almost constant above age 50 within each birth cohort (Peto et al, 2004
). The cumulative risk of developing CIN3 () shows a similar pattern. In view of this similarity and the effectiveness of regular screening in preventing cervical cancer, it seems likely that the incidence of cervical cancer remains roughly constant at about 1% per year for the rest of their lives in women who have developed undiagnosed CIN3, and about 40% will eventually develop invasive cancer (Peto et al, 2004
Implications for screening policy
If most CIN3s eventually regress cytologically, an increasing proportion will be missed as the screening interval increases and many may not reappear until they become malignant. CIN3 regression is also relevant in choosing the age at which screening should begin. The age at which routine screening begins has recently been raised from 20 to 25 years in the UK, and it has been suggested that this should be further increased to 30 years because invasive cancer is so rare in young women (Miller, 2002
). Registered CIN3 rates for England for 2000 imply that 2.9% of women are now diagnosed with histologically confirmed CIN3 by age 30, and the underlying risk must be even higher, as coverage is still incomplete. If screening began at age 30, therefore, at least 3% of women would already have had undiagnosed CIN3. suggests that 3.9% of women in our cohort develop CIN3 by age 30. The majority of such cases may no longer be cytologically detectable at age 30, as the prevalence of CIN3 in women with no previous smear () was only 0.8% (16 cases) at age 25–29 and 1.1% (9 cases) at age 30–34.
CIN3 in an older woman who has not been screened regularly may often be a recurrence of an earlier undiagnosed lesion. The CIN3 rate in England was increasing at all ages before 1988, but the rate among women aged over 45 has fallen progressively since the introduction of national screening. Women whose CIN3 has already regressed before their first smear and does not recur may suffer an increased risk of invasive cancer for the rest of their lives, and this long-term risk could be an important disadvantage of delaying the age at which routine screening begins (Peto et al, 2004
Our results support the conclusion that the screening interval should be less than 5 years, except perhaps for older women who are HPV negative, whose risk of CIN3 is very low (). Our analysis suggests that more frequent screening will have little effect on the overall number of CIN3 cases diagnosed in a regularly screened population. More frequent screening will prevent more CIN3s by ablation of lesser lesions; but it will also detect more before they have regressed cytologically, and will therefore reduce cancer incidence. The anomalous long-term protective effect of abnormal cytology among HPV-infected women in Portland suggests that the annual cytology, routine colposcopy of ASCUS and more frequent ablative treatment still practised by most American gynaecologists (Noller et al, 2003
) will over time lead to a reduction in CIN3 incidence. The additional reduction in invasive cancer achieved by annual rather than 3-yearly screening is likely to be small (Sawaya et al, 2003
), however, and many lesions that are HPV-negative and hence harmless will also be treated. Sensitive HPV testing is therefore a useful adjunct to cytology (Cuzick et al, 1995
), as mildly abnormal smears that do not contain HR-HPV can be identified and ignored (Cox et al, 2003
). The potential superiority of HR-HPV testing over cytology for sensitive primary screening is now widely recognised. Very sensitive HPV testing will reduce still further the CIN3 or cancer risk in HPV-negative women and increase the proportion of CIN3s that are preceded by persistent HPV detection, but it is also likely to increase the number of referrals for colposcopy. Other practical questions that must be resolved before HPV testing can replace cytology are the appropriate screening interval, whether HR-HPV infection should be observed or treated immediately by cervical ablation, and whether cytology is worthwhile in addition to HPV testing.
Manchester: Abbey Hey Clinic, Alexandra Park Health Centre, Ancoats Community Clinic, Ancoats Hospital, Assheton Road Clinic, Atherton Clinic, Baguley Clinic, Beswick Health Centre, Bodey Medical Centre, Boothtown Clinic, Brunswick Clinic, Central Drive Clinic, Charlestown Health Centre, Child Health Centre, Clayton Health Centre, Corporation Road Clinic, Crumpsall Clinic, Derbyshire House Clinic, Gorton Combined Clinic, Harpurhey Health Centre, Hulme Clinic, Irlam Health Centre, Ladybarn Group Practice, Levenshulme Health Centre, Little Hulton Clinic, Longsight Health Centre, Monton Road Clinic, Moss Side Health Centre, Newton Heath Health Centre, Nicholas Road Health Centre, Northenden Group Practice, Northern Hospital, Partington Lane Clinic, Plant Hill Clinic, Poplar Street Clinic, Robert Darbyshire Practice, Smedley Street Clinic, St Mary's Hospital, The Palatine Centre, Varley Street Clinic, Walkden Clinic, Walmer Street Clinic, Wellwoman Clinic, Wilmslow Road Clinic, Woodhouse Park Clinic, Wythenshawe Health Care Centre. Rochdale: Whitworth Medical Centre. Salford: Bevendon Square Clinic, Higher Broughton Health Centre, Hope Hospital, Lanceburn Health Centre, Liverpool Street Clinic, Lower Broughton Clinic, Ordsall Health Centre, Langworthy Clinic. Stockport: Adswood Clinic, Bramhall Clinic, Brinnington Clinic, Burley House Clinic, Cheadle Heath Clinic, Hazel Grove Clinic, Heaton Norris Clinic, Maple Clinic, Marple Cottage Surgery, North Reddish Clinic, Offerton Clinic, Offerton Green Clinic, Robins Lane Medical Centre, Romiley Clinic, Romiley Health Centre, Shaw Heath Clinic, Stepping Hill Clinic, Wellwoman Clinic, Woodley Clinic, Bredbury Clinic, Brookfield Clinic, Heaton Norris Clinic. Warrington: Derby Road Clinic. Wigan: Beech Hill Health Centre, College Road Clinic, Grasmere Street Health Centre, Haig Road Clinic, Liverpool Road Health Centre, Longshoot Lane Health Centre, Orrell Road Clinic, Pemberton Clinic, Platt Bridge Clinic, Queens Road Clinic, Standish Clinic