Conflict of Interest: Marjolein van Ballegooijen, Dik Habbema, Matejka Rebolj: The Department of Public Health of the Erasmus MC received a grant from GSK, a manufacturer of an HPV vaccine, for research on the cost-effectiveness of HPV vaccination in 2007 and 2008. This research and manuscript were neither funded nor supported by GSK. Rob Boer has been participating since 1989 in the screening research group at the Department of Public Health of the Erasmus MC. He has been affiliated with RAND since 2000. Since 2007, he has been a Director of Evidence Based Strategies - Disease Modeling and Economic Evaluation at Pfizer, which develops and sells various medicines for cancer and other diseases. This research and manuscript were neither funded nor supported by Pfizer.
The number of women who benefit from cervical cancer screening (estimated at about 1%)1, 2 is small compared to the number that bear a considerable burden of non-negative screening outcomes (2–10% per screening round).3–8 Women with non-negative cytologic findings are advised to stay in follow-up, and sometimes need to undergo colposcopy and treatment for preinvasive lesions, which most often would not have developed into cancer.9, 10 It is therefore crucial to set an appropriate threshold for the minimal level of relevant smear abnormality.
In The Netherlands, the current definition of this threshold, i.e. of borderline dyskaryosis (ASCUS), was agreed upon in 1996 as part of the major changes in the cervical cancer screening programme.4, 11 Since 1996, inflammatory changes without dysplasia are seen as benign for cervical neoplasia and are included in the category of a negative rather than a borderline dyskaryotic Pap smear.12 This guideline definition is comparable to that in many other countries, e.g. England and the USA.13–15
This change had a tremendous impact on the Dutch screening practice. During the 1980's and the early 1990's, about 10% of smears were judged borderline dyskaryotic (including inflammation).16, 17 It was recommended to follow up this group of women with yearly smears until they had two consecutive negative smears and could thereafter rejoin the regular screening programme, or until their cytologic diagnosis worsened and they were referred to colposcopy.18 In practice, it took several years before such a decision could be made.4 After the guidelines began to recommend classifying inflammation as a negative screening outcome, the proportion of borderline dyskaryotic smears decreased to less than 2% (1.5% for programme smears), whereas, the remaining 80% of pre-1996 borderline smears are now classified as negative.19
Broadening the definition of a negative Pap smear with inflammation could cause a loss in smear sensitivity, though this has never been adequately studied. The Dutch observational experiment combined with the existence of nationwide comprehensive registration of total screening activity linked to diagnostic histological outcomes (including cancer) at the individual level since 1990 give a unique opportunity to study whether this has been the case. We analysed the changes in the incidence of invasive cancer after a negative smear (i.e., interval cancers) before and after the definition change.
Material and methods
Information on all cervix uteri cytological and histological tests in The Netherlands registered until March 31, 2007 was retrieved from the nation-wide network and registry of histo- and cytopathology (“PALGA”).19 The registration began in the late 1970's, and achieved practically complete coverage of pathology registries in 1990.20 PALGA identifies a woman through her birth date and the first four letters of the maiden name. This identification string enables the linkage of different tests belonging to the same woman, and therefore also to follow individual testing histories (dates and diagnoses). The problem of false identity matches21 was avoided by excluding women with 0.5% most common maiden names which corresponds to ∼30% of all women.22
Registered screening histories were organized into screening episodes. An episode is defined as starting with a primary test (a smear or a biopsy) followed by secondary tests in case this test was abnormal (at least borderline dyskaryosis) or of inadequate quality. Follow-up or secondary tests were defined as the tests made within 4 years following the primary test, unless the follow-up of this primary smear had already been completed according to the guidelines12, 23 (e.g., with two consecutive negative smears after a borderline dyskaryotic smear, or three consecutive negative smears after histologically confirmed cervical intraepithelial neoplasia (CIN)). All other tests were seen as primary tests.
We identified women with cervical cancer by selecting all PALGA records that included pathology codes for cervical cancer between 1994 and 2006. For these women, we reviewed the free text of all histology reports in PALGA. Complete follow-up (woman-years at risk and cases) was therefore left-censored at the beginning of 1994 and right-censored at the end of 2006. Cases and woman-years at risk were counted for at most 9 years from a negative primary smear until the primary smear of the next episode (or the date of the first histologically proven diagnosis of cervical cancer if it was diagnosed in the next episode (cases)), or else until December 31, 2006. Because during the analysed period the reason for smear-taking was not always registered in PALGA, all cancers originating in the cervix were counted as cases regardless of the reason for the primary investigation. Thus, all invasive cervical cancers diagnosed subsequent to a negative primary smear, including screen-detected, were counted as cases.
We compared negative primary smears taken in calendar periods 1990–1995 and 1998–2006. Smears made in 1996–1997 were excluded from the analysis because this was a transition period in which the proportion of borderline smears was still high but decreasing (1996: 5.5%, 1997: 3.3%).19 First, the cumulative incidence rate (CIR) of interval cancer was calculated for each period for women aged 30–64 at the time of the primary smear. We focused on the 6-year CIR because this period covers the next screening round scheduled to take place 5 years after a negative smear. The 95% confidence intervals were estimated by non-parametric Kaplan-Meier product-limit estimator for log(hazard).24 Second, the relative hazards were estimated by univariate and multivariate Cox regression with left and right censoring. In the univariate model, period (1998–2006 vs. 1990–1995) was entered as the explanatory variable; in the multivariate model also age at primary smear (grouped as 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64 years), the number of previous negative primary smears (First negative smear vs. second or later consecutive negative smear), and the history of abnormalities (no cytologic or histologic abnormality vs. at least a borderline negative smear in the past) were entered. Time dependency of the relative hazards was statistically tested by splitting the total follow-up time in 2 periods with a roughly equal number of cases. This leads to 2 new datasets where the full model is fitted on again. If the sum of the deviance of both sub-models is significantly lower than the deviance of the original model, the parameter estimates differ significantly between the 2 periods.
The detection rates of CIN grades 1, 2 and 3, and invasive cancer, i.e. the proportion of primary smears with a histologically confirmed cervical lesion, were calculated as the number of those lesions (numerator) per 1,000 smears in previously unscreened women aged 30–64 (denominator). To allow for follow-up, the most severe histologically confirmed diagnosis within 27 months of an abnormal smear was used as the final diagnosis. The age-adjusted odds ratios for periods January 1998–September 2004 versus January 1994–December 1995 were estimated with logistic regression.
We identified 1,546,252 negative smears made in 1990–1995 that contributed woman-years in the analysis, and 3,552,716 negative smears in 1998–2006 (Table I). These were made in 1,136,631 and 2,005,627 women, respectively (2,314,250 women in total). The age at which these smears were taken was somewhat higher in the 1998–2006 period. The negative smears from the 1990–1995 period were more likely first rather than second or later consecutive negative smears, and had therefore fewer previous abnormalities than those from the 1998–2006 period. Three hundred and seventy-seven women were diagnosed with an invasive cancer in 5,232,959 woman-years at risk following a negative smear made in the 1990–1995 period, and 619 women in 11,210,675 woman-years at risk following a negative smear made in the 1998–2006 period.
Table I. The Characteristics of Negative Primary Smears, by Period: 1990–1995 (1,136,631 Women) and 1998–2006 (2,005,627 Women)
Number of negative primary smears
Age at negative primary smear
Length of follow-up
Less than 2 years
6 years or more
1st negative smear
2nd or later negative smear
No previous abnormality
Previous at least cytological borderline dyskaryosis
The cumulative incidence rate (CIR) for the 1998–2006 negative smears was never significantly higher than that for the 1990–1995 smears (Fig. 1). At 6 years, the CIR were 46 (95% CI: 41–52) and 48 (43–54) per 100,000 negative smears in the 1990–1995 and the 1998–2006 periods, respectively (p = 0.59). The univariate hazard ratio was 0.88 (0.77–1.00) for 1998–2006 versus 1990–1995. The multivariate hazard ratio adjusted for age at the primary smear, the number of previous consecutive negative smears, and the history of abnormalities was 0.90 (0.78–1.03) and,therefore, not substantially different from the univariate hazard. The test for time dependency of the relative hazards was statistically non-significant (p = 0.45).
The detection rate of CIN 1 in previously unscreened women showed a 38% statistically significant decrease after 1998, whereas the detection rate of CIN 2+ increased by 22% (Table II). The positive predictive value (PPV) of an abnormal primary smear (borderline dyskaryosis or worse) for CIN 2+ more than tripled in the recent period.
Table II. Detection Rates of CIN and Invasive Cancer Per 1,000 Primary Smears in Previously Unscreened Women Aged 30–64, by Calendar Period, and Age–Adjusted Odds Ratios (95% CI)
In The Netherlands, treating inflammatory smears as a negative screening outcome decreased the proportion of borderline smears by 80%.17 In spite of this, no increase in the risk for cervical cancer among women with a negative smear could be observed (Fig. 1). This conclusion was based on a large observed series of cases that produced relatively narrow confidence intervals.
An increased cancer risk associated with inflammation could have been diluted in this comparison because it represents a small proportion (less than 10%) of all negative smears. To gain a more detailed insight into the risk associated specifically with inflammation, we additionally compared the detection rates of CIN and cancer before and after the change in the definition. We limited the analysis of the detection rates to previously unscreened women in order to obtain an unbiased comparison regarding their screening history. An unchanged detection rate would imply that there is no increased risk associated with inflammation, whereas a decrease would suggest an increased risk. The observed overall detection rates for CIN 2+, however, show that the programme's ability to detect cervical lesions was not diminished after 1996 (Table II). Limiting these detection rates to the CIN lesions found after a borderline dyskaryotic smear also showed an increase (data not shown). The observed increase in the detection rates must be due to other factors. Indeed, since 1996 borderline smears had a more complete follow-up, and biopsies were taken more often than before 1996.4 On its own, the latter suggests that the narrower definition of a borderline dyskaryotic smear has been perceived as a more serious screening outcome, and tended to improve the quality of follow-up.
Since screen-detected cancers were included in the analysis, we evaluated whether our results were affected by a change (i.e., a decrease) in the screening frequency between the periods.11 At 6 years, the 1998–2006 cumulative “incidence” of a subsequent smear after a negative smear was 9% lower than after a smear made in 1990–1995.19 The effect of this small difference was further reduced because only part of the cancers (estimated <50%)19 are screen-detected.
All smears and histologically diagnosed CIN and cervical cancers are in principle registered in PALGA. In The Netherlands, this is the only comprehensive registry that links the cancer cases with their screening history. In PALGA, the total number of women aged 30–74 with an incident cervical cancer in the period 1994–2003 was 10% higher than published by the Cancer Registry.25 There was no trend in the differences between the two sources by calendar year. Thus, although the interval cancer rates may have been somewhat overestimated, the comparison of the periods was not biased.
Even though the recommended definition of borderline dyskaryosis is similar in several countries, the observed smear abnormality rates vary substantially. The diagnosis of borderline dyskaryosis is made in 1.5% of the programme smears between ages 30 and 64 in The Netherlands, 3% in England and almost 5% in the USA.19, 26, 27 This variation could be due to genuine differences in the background risk, or to the differences in smear interpretation. The differences in the observed incidence rates are small,1 while the country-specific screening intensities are similar.4, 6, 28 This suggests that the background risk is comparable. On the other hand, systematic differences in the interpretation of borderline dyskaryosis have been shown to exist, with the tendency in England and the USA to classify subtler lesions as borderline abnormal.29 This suggests that there may still be room for decreasing the smear abnormality rates in these other countries.
This observational experiment with broadening the definition of a negative smear took place in a setting with high-quality cytology practice. In The Netherlands, strict quality assurance measures aimed at improving and standardising smear interpretation are set out in professional guidelines.12, 30 Nation-wide pro-forma reporting for cytology was introduced in 1996, and instruction CD-ROMs with guidelines and visual analogues were distributed to every cytotechnician.12, 31 Laboratory-specific feedback is provided to all pathology laboratories in the framework of continuous monitoring.
The balance between the positivity rate (i.e., the proportion of smears with a positive outcome) and increased cancer prevention is also relevant in HPV screening. The positivity rates of the most widely used HPV tests are 2 to 4 times higher than those of conventional cytology.32, 33 Partly, the higher abnormality rate will represent a loss in specificity, but unlike with inflammation, a gain in sensitivity can be expected.32, 33 These will have to be weighted against each other after pooling the data on interval cancers from the currently on-going HPV trials,34 which will allow an even more accurate estimate of the true increase in the sensitivity.
In conclusion, broadening the definition of a negative smear with inflammatory changes in The Netherlands led to an 80% decrease in borderline abnormal smears without increasing the risk for cervical cancer after a negative smear.