A Longitudinal Swedish Study on Screening for Squamous Cell Carcinoma and Adenocarcinoma: Evidence of Effectiveness and Overtreatment

The use of cytologic screening to detect and remove precursor lesions has had a huge impact on both the incidence and mortality of cervical cancer. However, despite the general decline in squamous cell carcinoma (SCC) incidence within some countries employing organized or opportunistic cytology screening (1-4), cervical cancer incidence remains at unacceptable levels globally (5). It is also worrisome that a high risk remains even in countries where organized screening occurs at regular intervals, suggesting that an upper limit for effectiveness has been reached using cytology screening. Adenocarcinoma of the cervix (adenocarcinoma) is potentially even more problematic, with the majority of countries (including Sweden) exhibiting increasing incidence rates for adenocarcinoma over the last couple of decades (6-8).

Although the two histology types of malignant cervical cancer exhibit slightly different etiologies, they do possess some similarities. Both cancer types are associated with similar oncogenic human papillomavirus (HPV) types, and both of them seem to be associated with precursor lesions, namely, squamous cell cervical cancer in situ (CIS) and adenocarcinoma in situ (AIS; ref. 9). Interestingly, there is a much larger variation between Swedish counties in the recorded incidence of these precursor lesions than in the incidence of invasive cancer. As previously described (1), this variation suggests that the criterion used to classify CIS varies markedly between clinical pathologists, although it is also possible that this may be the result of there being more CIS cases overall. Presumably, a certain level of variability (between clinical pathologists) would also extend to the characterization of AIS.

A previous Swedish study made use of this phenomenon to estimate whether the effectiveness of Pap smear screening with respect to SCC was predicted by the recorded incidence of CIS (1). We were interested in using a similar methodology to investigate the effectiveness of Pap smear screening also on adenocarcinoma incidence. This approach makes use of the aforementioned fact that historically, there have been substantial differences in incidence rates of both CIS and AIS between counties. In Swedish counties where the recorded AIS incidence was higher, implying that a larger number of precursor lesions were detected and removed, we would expect to see a correspondingly lower incidence of adenocarcinoma compared with counties exhibiting a lower AIS incidence rate for the same period after an appropriate time. This time would represent the lag time for progression from detectable AIS to adenocarcinoma. We did similar analyses with CIS and SCC to see whether the lack of association observed in the previous study (1) on CIS and SCC continued over the last decade.

The Swedish National Cancer Registry was established in 1958. Reporting of all cancers and in situ lesions has been ongoing and, in more recent years, is considered to include virtually 100% of incident cancer cases in Sweden (10). This is achieved through the contribution of the pathologists and cytologists of 21 counties who separately report their results to the registry. The vast majority of their reports include information on individuals from within their county because it is uncommon for individuals to contact hospitals outside of their place of residence.

The end result is a registry containing cancer information for all Swedish residents, which stretches back to 1958 and can be separated into county-specific data sets. Cervical cancer reports contained within the cancer registry are mostly diagnosed based on biopsy (rather than cytologic) interpretation.

Papanicolaou smear (Pap smear) screening began in Sweden during the early 1960s as opportunistic screening. Organized screening was introduced from 1967 onward, with some counties refraining from participation until the early 1970s. From early 1970s, all women were invited to attend Pap smear testing once every 4 years. Since the mid-1970s, this has increased to once every 3 years.

In Sweden, the typical collection method for cervical screening before the 1980s was the spatula. Post-1980s, the cytobrush was incorporated in combination with the use of spatula collection. Liquid-based cytology has not been used routinely in Sweden.

Before the 1980s, high-grade cervical dysplasia was treated using cold knife conization. During the 1980s, however, more conservative modes of treatment such as cauterization, cryotherapy, and laser vaporization or conization were more readily employed. In 1990, the large loop excision of the transformation zone (LLETZ) was introduced and remains the procedure of choice for management of high-grade cervical lesions.

To increase the reliability of our data with respect to the screening/detection of cervical cancer and its precursors in counties, we restricted our study period from 1968 onward. The latest date for which we possessed data was 2002. Between 1968 and 2002, we observed 125,543 cases of CIS and 17,399 cases of SCC. In contrast, the rare AIS and adenocarcinoma lesions were reported in 1,260 and 3,602 women, respectively, during the same period. The source population of women from whom these cases arose was numbered at 3,991,867 women in 1968 and grew to 4,507,708 women in 2002. There were a total number of 145,252,306 person-years observed for Swedish women during follow-up, estimated using mid-populations for each calendar year.

Ethical approval was obtained from the Karolinska Institutet Ethics Committee.

We calculated age-standardized incidence rates (ASR) for each county and time period using the Swedish census population in 1970 as a reference. Incidence rates were reported as the number of case women per 100,000 women in the population.

For each county, Spearman correlations between CIS and SCC rates, in addition to AIS and adenocarcinoma rates, were estimated. To take into account the expected lag period between CIS development and SCC detection, we compared current age-standardized incidence rates for SCC to those for CIS in the same county and cohort 5, 10, or 15 years before. The same analyses were done for AIS and adenocarcinoma.

To more effectively compare CIS/SCC correlations between counties, we used a generalized estimating equation (GEE) model, where we allowed the variances to differ for different counties and time periods. We then added the CIS incidence (current, −5 years, or −10 years) as a continuous covariate, estimating the association between SCC and lagged CIS. The 15-year lag period was omitted from these analyses because of power constraints.

The distribution of women for the whole of Sweden in various age groups during different time periods are shown for CIS and SCC of the cervix in Table 1 and for AIS and adenocarcinoma in Table 2. The majority of women with CIS were aged between 20 and 40 years at the time of diagnosis. The number of women with CIS under the age of 50 decreased between the first time period (1968-1979) and the following two time periods. This trend seemed to be reversed in the older age groups, however.

The age distributions of SCC differed to that of CIS. Between the years 1968 and 1979, the majority of women (65%) diagnosed with SCC were aged between 40 and 70 years. Twenty years later, however, there seemed to be a shift to younger ages (30-50 years of age). In all age groups of women, apart from those who were 80+ years, there were a decreasing absolute number of women developing SCC during the following two time periods (1980-2002).

The variation in incidence of AIS and adenocarcinoma cases during the three time periods (Table 2) was quite different to that of CIS and SCC. Although the numbers of observed AIS cases were far fewer than CIS or SCC cases, there was a large increase in absolute numbers of AIS cases between the first and last time periods. Although the increase in number of adenocarcinoma cases was not as pronounced as it was for AIS, nevertheless, there existed a clear increase in adenocarcinoma during the same time periods. AIS was predominantly found in women between the ages of 20 and 50 years in the most recent 10-year period; however, adenocarcinoma occurred mostly in women aged 30 to 60 years.

The standardized incidence rate patterns for all counties combined (the whole of Sweden) are shown in Fig. 1A and B (SCC, CIS and adenocarcinoma, AIS, respectively). An overall decrease in both SCC and CIS can be seen in Fig. 1A between the years 1968 and 2002. The opposite was observed for adenocarcinoma and AIS, however (Fig. 1B), particularly with respect to AIS, which showed an almost exponential increase in incidence from 1973 onward.

We next calculated standardized incidence rates for each of the four cervical cancer types for each Swedish county (Tables 4–7). For a selection of Swedish counties (including the larger counties), these ASRs were plotted to inspect the variation in incidence between counties for the various histologic cervical cancer types (Fig. 2A-D). Clearly, there are tangible differences in reporting of the various histologic cancer types. These differences are least for SCC, possibly due to the relative ease in classification (compared with CIS) and less random variation due to the larger numbers of cases of SCC compared with AIS and adenocarcinoma.

Before Pap smear screening had any impact, incidence rates of SCC varied about 2-fold between counties. Because cancer registration was considered about 100% complete, this variation likely reflected geographic differences in past exposure to causal factors. During subsequent decades of extensive screening, incidence rates were reduced by at least 50%, and geographic variation becomes limited.

For CIS, a different pattern was revealed. The recorded incidence varied up to 7-fold between different counties before 1990, and reported incidence rates overall in Sweden were reduced by <30% in later years. In addition, we found drastic changes, decreases as well as increases, in incident rates within counties from one time period to the next. True incidence trends are unlikely to follow such patterns, and the pattern of organized screening has changed little over time. Therefore, these geographic differences and temporal trends likely reflect that the histopathologic criteria used to classify CIS were poorly standardized among clinical pathologists.

For adenocarcinoma and AIS, interpretation is hampered by random variation due to small numbers when incidence rates were calculated for each county. The recorded incidence of adenocarcinoma varied 3-fold between counties throughout most periods; however, the overall reported incidence of adenocarcinoma visibly increased over the calendar period until the last period where the incidence dropped slightly. In a similar fashion to CIS, some counties reported dramatic differences in incidence of adenocarcinoma between different periods.

The reported incidence of AIS during early periods of screening was negligible for all counties. During the 1980s, however, an increase in reported incidence of AIS occurred. This incidence of AIS in Sweden continues to increase steadily, showing no signs of abate. AIS incidence has, and continues to be, highly variable between counties (higher than the aforementioned histologic types). This variation most likely reflects difficulties in the classification of AIS, although it may be partly related to the rarity of the lesion.

Spearman correlation analysis of the standardized incidence rates for SCC and CIS (5 or 10 years before SCC) showed weakly (but statistically nonsignificant) positive correlations between the incidence of SCC and that of CIS, 5 years (r = 0.19; P = 0.051) or 10 years (r = 0.16; P = 0.095) prior (Table 3). Similar analyses for adenocarcinoma and AIS (5 or 10 years before adenocarcinoma) showed slightly stronger positive correlations of r = 0.31 (P = 0.001) for the 5-year and r = 0.28 (P = 0.003) for the 10-year lag periods (Table 3). In addition, we did similar analyses using a lag period of 15 years. There were no apparent correlations between incidence of SCC and CIS (−15 years) or between adenocarcinoma and AIS (−15 years).

These findings were verified using GEE models. The relative change in SCC for an increase of 100 CIS cases per 100,000 women-years (Tables 4, 5, 6, and 7) was positive rather than inverse and estimated to be 1.05 [95% confidence interval (95% CI), 0.99-1.10] for the 5-year and 1.02 (95% CI, 0.96-1.08) for the 10-year lag periods. For the adenocarcinoma and AIS relationship, there were fewer cases and, as a consequence, lower statistical power. The relative change in adenocarcinoma for an increase of 10 AIS cases per 100,000 women was estimated to be 1.17 (95% CI, 0.90-1.52) and 1.08 (95% CI, 0.81-1.43) for −5 and −10 years lag, respectively. The lack of an inverse correlation suggests that increased reported incidence of CIS in certain counties does not forecast a reduction in SCC for those counties. Thus, whereas the GEE statistical model reduces some of the variability present in the Spearman correlations, there was still no support for an association between SCC and −5- or −10-year CIS rate or between adenocarcinoma and −5- or −10-year AIS rate.

Our study investigates the effectiveness of Pap smear screening in Sweden between the years 1968 (when organized Pap smear screening began) to 2002. For the most common cervical cancer type (SCC), one might expect that differences in the incidence of its precursor lesion (CIS) in different counties within Sweden would influence the SCC incidence. We show that although there are indeed large differences in CIS incidence for different counties, this seems not to influence the pattern of SCC incidence in either a positive or negative fashion. Because adenocarcinoma and AIS have been increasing alarmingly over the last 30 years, we also investigated these histologic cancer types in a similar manner. We found no evidence suggesting that the detection and removal of AIS lesions predicts a subsequent decrease in incidence of adenocarcinoma.

In our study, we assume that an inverse association between CIS (or AIS) incidence and subsequent SCC (or adenocarcinoma) incidence (5-15 years later) suggests a lack of effectiveness in Pap smear screening. To a certain extent, this relies on the assumption that migration between counties is negligible. If there was a large amount of migration between counties, there may be an impact on the correlations between precursor lesions and cervical cancer. However, it is unlikely that between-county migration patterns would be associated with the incidence of cervical cancer. We therefore do not expect there to be any differential bias problems associated with this type of migration. Although a certain degree of nondifferential bias may exist as a result of between-county migration, the number of individuals who migrate is unlikely to substantially affect our correlation estimates.

Although there is no doubt that the introduction of organized Pap smear screening has had a major impact on the incidence of cervical cancer (4, 11), a number of studies have shown data supporting a possible overkill with respect to Pap smear screening (1, 11, 12). There seems to be little substantial benefit either from broadening the age range for screening women or from reducing the time interval for screening from 5-yearly to 3-yearly or annually (12-14).

The significance of this overtreatment relates to the contraindications associated with the removal of CIS lesions. In addition to the patients' emotional trauma associated with discovery of CIS, there is also the immediate physical discomfort and potential postoperative complications from the treatment. Future effects of various treatments for CIS removal have been explored by other researchers. The more recently preferred treatments are the loop electrosurgical excision procedure (LEEP) and, to a lesser extent, laser conization. Both of these treatments have been associated with an increased risk of premature rupture of membranes before gestation of 37 weeks, resulting in preterm delivery (15). Moreover, infants to mothers who have undergone LEEP may have a lower birth weight as a consequence (16), particularly if the electrosurgical loop is large (17).

Our investigation, which used more than 30 years of data on cervical cancer screening in Sweden, pinpoints the relatively nonspecific nature of Pap smear screening. This is not entirely surprising, given that the estimated number of CIS lesions expected to regress spontaneously is ∼80% or higher (14, 18, 19). Nevertheless, it is important to verify with greater accuracy and further draw attention to this problem. Perhaps future identification of more specific molecular assays than traditional cytology detection methods could be used to determine which women with CIS require further treatment.

In terms of the ability of Pap smear screening to prevent adenocarcinoma, there has been considerably less benefit observed than for SCC. Most countries have seen an increasing incidence of adenocarcinoma in women born following the years 1930 to 1945 (8). This may be a consequence of an increasing ability to identify this subtype over time and/or may be related to increases in certain background risk factors (such as HPV prevalence) or cofactors (such as tobacco smoking; ref. 8). In our study, the slight leveling-off of adenocarcinoma incidence since the early 1980s lends credence to the likelihood that some of the increase in incidence can be attributable to improved (and more standardized) characterization of adenocarcinoma in Sweden. Presumably, after a given amount of time, one would expect the improvements in characterization of adenocarcinoma to reach an optimum, which would result in a plateau of adenocarcinoma incidence.

Interestingly, the observed increase in incidence of AIS parallels the introduction of cytobrush techniques (in combination with the use of spatula) during the early 1980s. This may partly explain the sharp increase in AIS incidence since the 1980s. Alternatively, the ever-increasing incidence of AIS could also suggest that there is an increase in the background risk factors for AIS and adenocarcinoma. If the incidence of AIS was merely a function of detection efficiency, and if AIS truly is a precursor to adenocarcinoma, one would expect to see a more pronounced decline in adenocarcinoma associated with this exponentially increasing AIS incidence. Also, detection and characterization of AIS should probably have reached a plateau in a similar manner to CIS and SCC after so many years of screening due to more standardized screening. This was not apparent in our study where the AIS incidence continues to increase at a steady rate with no dramatic changes in adenocarcinoma.

Related to this is the lack of association between AIS and adenocarcinoma in our study. We did not see any correlation between the incidence of AIS in cohorts, 5, 10, or 15 years before adenocarcinoma. Although the lack of association between SCC and CIS is probably related to Pap smear screening reaching a plateau of effectiveness, for adenocarcinoma and AIS, it seems more likely that this is due to an increase in background risk factors in this population.

Historically, there has been only a limited ability to reduce adenocarcinoma incidence through treatment of AIS lesions (8, 20-22). This may be related to difficulties in detecting and successfully removing the precursor cells (23). Hence, it is possible that the incidence of AIS and adenocarcinoma over time more clearly reflects the true effects of risk factors for adenocarcinoma in the population than CIS or SCC, which may not be seen due to the effects of Pap smear screening. Also, this inability to effectively remove the precursor AIS cells adequately would in part explain why AIS incidence increases so much with no measurable decrease in adenocarcinoma incidence when we plot their respective standardized incidence rates over the calendar period.

One alternative explanation for the ever-increasing AIS incidence without a corresponding decrease of adenocarcinoma is the possibility that AIS (or a large proportion of AIS) is not a precursor to adenocarcinoma. Although AIS seems to be accepted as a precursor lesion to adenocarcinoma, much of the evidence is circumstantial and/or underpowered (24-29). A second plausible alternative is that the proportion of AIS that progresses to adenocarcinoma may be even lower than the proportion of CIS that progresses to SCC. This could explain the relatively slight effects in reducing incidence of adenocarcinoma (associated with AIS treatment) compared with the clear decline in SCC as a result of CIS treatment.

Our data indicate that women develop SCC, regardless of the variations in cervical cancer in situ incidence, which occurs between the regions they live. The most likely explanation for this is that there is an overtreatment of CIS lesions. Women developing cervical adenocarcinoma are also unaffected by levels of AIS incidence in their counties, although this phenomenon is more likely to be attributable to a lack of effectiveness of Pap smear screening in detecting AIS. Moreover, there has been a disturbing increase in AIS over the last 30 years, which would seem to be the result of an increase in background risk factors and has yet to abate. This, coupled with the lack of effectiveness in Pap smear screening for adenocarcinoma, could ensure further increases in future adenocarcinoma incidence.