This study was restricted to countries in North America and Europe where cervical screening has been in place for several years. Data were available from the United Kingdom, France, Germany, the Netherlands, United States and Canada. The studies included have published their initial results comparing HPV testing to cytology in routine screening.5, 6, 7, 8, 9, 10, 11, 12 However, these results have been reported in different ways, and here we provide a unified analysis including a comparison against cytology, both overall and for different age groups. Because each study used different entry criteria, cytology cutoffs and adjustments for verification bias (see later), our results did not agree exactly with those in the published reports of the individual studies. The main characteristics of each study are shown in Table I. All studies have a similar split-sample study design. They focussed on screening populations mostly aged 30–60, and the Hybrid Capture II (HC-II) test was used in all studies except for those based in the Netherlands and Jena, where consensus PCR with the GP5+/6+ primers was used. Data from the more recent studies in which women are randomised to cytology or HPV testings (with or without cytology) are only just becoming available and are not included. They are less efficient for assessing sensitivity and specificity, but if large enough will be essential for establishing overall effectiveness of programmes. We also have not included studies in the developing counties, where cytology was often not part of routine care and the overall characteristics of both cytology and HPV testing appear to be somewhat poorer, possibly because of other concomitant STIs. These studies have been reviewed by IARC (2005).3 Specific characteristics included for each study are briefly noted later.
Table I. European and North American HPV Screening Studies1
|HART (Cuzick et al., 2003)||10,358||30–60||No smear test within the last 3 year, never treated for CIN||HC-II||Immediate colpo if cyto ≥ mild dyskaryosis. Borderline cyto or HPV+ randomised to immediate colpo or HPV and cyto in 6–12 months||Blinded central review|
|German – Tubingen||4,2302||30–87||No abnormal smear in preceding year||HC-II||Immediate colpo if cyto ≥ Pap IIw or HPV+||Blinded central review|
|German – Hannover (Petry et al., 2003)||4,7372|
|Jena (Schneider et al., 2000)||4,761||18–70||No abnormal smear in preceding year||GP 5/6||Referred Colpo and biopsy if abnormality found in colpo, cyto or HPV||Blinded single pathologist any abnormality reviewed by two pathologists|
|Remainder had follow up cyto and colpo at 4–8 months|
|French – Public||8,8582||15–80||No abnormal smear, untreated cervical lesion in last 2 years||HC-II||Immediate colpo for cyto+ results. If neg cyto and HPV+, repeat testing at 6 and 12 months||Two blinded pathologists, discrepancies resolved by third pathologist|
|French – Private (Clavel et al., 2001)||5,2652|
|Seattle (Kulasingam et al., 2002)||4,075||18–50||Routine screening at planned parenthood clinics||MY09-11 (4,075) and HC-II (1,150)||Colpo and biopsy if cyto ≥ ASCUS or HPV+||Single blinded pathologist|
|Only 1,150 with HC-II included||Never treated for CIN||1 in 10 women with cyto neg and HPV neg randomly chosen for colpo|
|Canada-Newfoundland (Ratnam et al., 2000)||2,098||18–69||Routine screening||HC-I (1,440) and HC-II (658)||Cyto ≥ ASCUS or HPV + referred for colpo and biopsy.||Local hospital labs|
|Only 658 with HC-II included||1 in 10 women with cyto neg and HPV neg were randomly chosen for colpo|
|Hammersmith3 (Cuzick et al., 1999)||2,988||35–65||No cytological abnormality in previous 3 years, or treatment for CIN||PCR plus HC-I (1,285) and HC-II (1,703)||Colp if cyto ≥ borderline or HPV(PCR) +||Two blinded pathologists|
|Only 1,703 with HC-II included|
|Netherlands3 (Bulkmans et al., 2004)||21,996||30–60||No abnormal cytology or CIN in last 2 years||GP 5/6||HPV + followed up at 6, 18 and 60 months||Four national labs|
|Italian||23,500 (not included)||25–60||Routine screening||HC-II||Age 35+ Immediate colpo if ≥ ASCUS and/or HPV +||Local pathologist Central review of all CIN2+ and a sample of CIN 1|
|Age <35, immediate colpo if cyto ≥ ACSUS; if cyto neg and HPV+, retest at 12 months and colpo if ≥ASCUS or HPV+|
In the HART study,11 women who were cytology negative and HPV positive, or whose cytology showed borderline changes (regardless of HPV status), were randomised to either immediate colposcopy or 6 monthly surveillance with HPV and cytology (with an exit colposcopy at 1 year).
The most recent study in Germany10 was conducted in two areas, one more rural (Tuebingen) and the other more urban (Hannover). The baseline rates of HPV positivity in these areas were different and these have been split into separate groups for this analysis. Pap IIw or higher was used as the cut-off for cytology positivity. Women who had a positive result for either cytology or HPV were referred for colposcopy.
For an earlier German study in Jena,7 all women received a screening colposcopy (without biopsy), HPV test and cytology. Pap III or higher was used as the cutoff for cytology positivity. If any of these results were abnormal, women were referred for a further colposcopy and biopsy. In addition, follow up cytology and colposcopy at 4–8 months was offered to all women who were initially negative for all 3 tests.
The French study6 offered HPV testing from age 20 and was conducted in Reims. It also had two distinct populations—one group of women who had their screening performed by private gynaecologists and the other who were screened in public hospitals. These two groups had different baseline HPV rates and social class distribution, and were also analysed separately. All women with a cytological abnormality were referred for immediate colposcopy, whereas those who were cytology negative and HPV positive were retested at 6 and 12 months by cytology and HPV. They were referred if either of these tests were positive at either visit.
The Hammersmith study5 enrolled only women above the age of 35. Management was based on the SHARP/PCR system, which used MY09/11 consensus primers. This detection system was found to be suboptimal. A second sample was then retrospectively analysed by the Hybrid Capture method using version 1 for the first 1,440 women and version 2 (HC-II) for the last 1,703 women. HC-II was found to be a very sensitive test, but because management was based on the SHARP results, it was not possible to properly adjust for verification bias as in the other studies, and consequently this study has been used only to evaluate age-specific HPV positivity in this analysis.
A large study based in the Netherlands12 has published initial results. Women with a high-grade cytological abnormality were referred for immediate colpscopy, but follow-up for women who were HPV positive and had low grade or normal cytology has not yet been completed. Therefore, at this stage, it is not possible to evaluate the performance of HPV testing, and so here these data have only been used for evaluating age-specific HPV positivity rates.
In the Seattle study,9 all women entering the study after January, 2000 received a HC-II test, and only these women were included here. Colposcopy and biopsy was performed on those who had had either a positive cytology or HPV test result, as well as on a random sample of women with completely negative results. This study was based at family planning clinics and contained a preponderance of younger women.
The Canadian study,8 like the Hammersmith study, converted to from using HC-I to HC-II during the trial and management was based on HC and cytology results. Only the HC-II part of the study was used in this analysis. All women with an abnormal smear or a HPV positive test result were referred for colposcopy.
A large Italian study has also been completed, but the results are not yet published, and their data are not included in this analysis.
The HC-II results were used in all studies except for Jena and the Netherlands, wherein consensus PCR with GP5+/6+ primers were used. Only the high-risk probe set was used for HC-II, which detects HPV types (16, 18, 31, 33, 35, 39, 45, 51 52, 56, 58, 59 and 68). Previous studies have shown that consensus PCR and HC-II give highly concordant results,13 and we have not separated them for this analysis.
Individual studies have evaluated cytology at different thresholds, but because cytological classification systems differ between countries, comparisons at different thresholds could not be done in this overview. Here, we have dichotomized cytology outcome at the lowest reported degree of abnormality (ASCUS, borderline changes, Pap IIw or equivalent).
Not all patients with abnormal screening results received colposcopy, mostly because of non-compliance. Therefore, not all cases of CIN2+ were identified, and an adjustment for verification bias was needed. Because the proportion of women with both negative cytology and HPV, who received colposcopy is very low, any attempt to adjust for missed disease in this group will lead to unstable results.14 However, any disease in this group of women does not influence the relative sensitivity or specificity of HPV vs. cytology, because the proportional effect is the same in both groups and their ratios are unchanged. Also, it can be seen by reviewing the reports from the individual studies that there are very few high-grade CIN cases found within this group. Thus, we chose to not adjust for the component of verification bias arising from missed cases in this group. As a result, sensitivities are slightly overestimated and specificities very slightly underestimated, both for cytology and for HPV.
Age specific rates were fit by logistic regression, with a spline containing knots at 25, 30, 40, 50 and 60 years. The individual curves for Seattle and Canada were truncated because of a small number of older women and the knots were at 5-year intervals from 25 to 45. For the ‘common shape’ analysis, a single spline was used for all studies combined, but additional indicator variables were included to model ‘shifts’ for each study. The purpose of this model was to more clearly separate the common characteristics of the shape of the age-prevalence curve from area-specific differences in prevalence. A stratified model was used to compute sensitivities and specificities for individual studies and overall, where the weights were the inverse of the stratum specific compliance levels. This approach allows one to adjust for verification bias as discussed earlier, in computing both means and confidence intervals. Thus, the sensitivities and specificities reported here do not agree with naïve estimators based on unadjusted counts, which were sometimes used in the original reports. The 95% confidence intervals, indicated in brackets, were computed by asymptotic methods, as used in sample survey methods. Trend tests were derived from generalised linear models. All calculations were performed by using STATA (Version 8.2).