Invasive cancer of the cervix occurring after treatment for cervical intraepithelial neoplasia (CIN) is becoming more important, as screening reduces the overall incidence of invasive disease. In 1997, a multicentre UK study of 44,699 women years of follow-up after conservative treatment of CIN showed that the cumulative rate of invasion after 8 years was 5.8 per 1,000 women and that the rate remained the same throughout the 8-year period such that these women remained at ˜5 times greater risk than the general population.1 The second edition of the NHS Cervical Screening Programme Guidelines,2 published in the same year, described the controversy surrounding the optimum duration of annual cytology testing after the treatment of CIN. It also referred to the importance of confirming that the rate of invasive disease did not fall and of investigating the reasons for this continuing high risk.
The present study was commissioned by NHS Cervical Screening Programme to undertake an analysis of long-term follow-up studies of women treated for CIN. The first objective was to determine from international data how long the rate of invasive cervical or vaginal cancer following treatment for CIN remains elevated. The second objective was to establish whether the rate of invasive disease was a reflection of the rate of posttreatment CIN. The aim was to investigate the reasons for the continuing high risk of invasive disease in these women so as to inform decisions about the follow-up protocol for these women.
Material and methods
The search strategy
A Medline search was constructed to identify articles published after 1977 that referred to the long-term success rate of the treatment of CIN with 5 or more years of follow-up. The search formula was [CIN or CERVICAL INTRAEPITHELIAL NEOPLASIA or “CERVICAL-INTRAEPITHELIAL-NEOPLASIA”/ or “CERVIX-DYSPLASIA”/ all subheadings or “CARCINOMA-IN-SITU”/ all subheadings or “CERVIX-NEOPLASMS”/ all subheadings] and [“TREATMENT-OUTCOME” or “TREATMENT-FAILURE” or FAILURE or OUTCOME or “DISEASE-PROGRESSION” or explode “COHORT-STUDIES”/ all] and (PY >= “1978”). This search strategy was intended to capture all potential articles at the expense of specificity. Potential long-term studies of the results of treatment of CIN were identified from the title and abstract. Copies of these articles were obtained and further articles were identified from the reference lists. These were examined to identify studies that met the inclusion criteria (Table I). No distinction was made between persistent disease, recurrent lesions or newly incident disease. In our study, all are referred to as posttreatment lesions.
Table I. Inclusion Criteria
1. The study must describe the results of treatment of women with CIN.
2. Recurrence of disease must be defined.
3. Follow up data for 5 years or more must be presented.
4. The numbers of women still under follow up during specified intervals must be shown.
5. The numbers of women developing a recurrence of CIN or cervical cancer during the specified intervals or the rates of recurrence during these intervals must be shown.
Using a scoring system devised for our study, the articles that met the inclusion criteria were assessed for the quality of the data and to determine whether the data were likely to be in a form capable of being analysed (Table II). Only studies that scored more than 10 out of 25 points were considered further. From these articles, the number of women years of observation and the numbers of women who developed posttreatment CIN or invasive cancer in each interval were to be determined. However, it proved impossible to do this in several cases and more articles were excluded at this stage because of uncertainty about various aspects of the data. These exclusions are described individually in the results. The authors of several articles generously made original data available3, 4 or sent the original protocol and data collection forms5 to assist in this analysis. We were unable to make contact with some others.
Table II. Scoring System for Assessment of Quality and of Suitability for Analysis
In some studies, the numbers or rates in each interval had to be calculated from the data presented.
The number of women years of follow-up were calculated by assuming that women were “lost-to-follow-up” uniformly through the interval using the formula: (number at start + number at end)/2 × length of interval in years. This changed slightly the results of calculations in the metaanalysis by Soutter et al. but does not change the conclusion of that analysis.1
Kendall's rank correlation test was used to assess the association between nonnormally distributed continuous variables. The data for posttreatment CIN were modelled using Poisson regression with a separate underlying rate for each study and a common curve to model the effect of time since treatment on the relative risk of recurrence. In order to deal with data from studies that reported rates in wider intervals, we maximised the mixture likelihood in Stata using a likelihood program written especially for this problem. Confidence intervals were estimated using Monte Carlo simulations. Recurrence rates for different treatments were compared with the Kruskal-Wallis test and with the Mantel-Haenszel test. All of the above were calculated with the aid of StatsDirect statistical software version 1.9.5 and Stata version 8.
The search and exclusions
The Medline search was completed in March 2004 and included publications from the period January 1977 to December 2003. This identified 1467 articles and a hand search of the bibliographies performed by 2 of the authors (TP and WPS) identified a further 381 potentially suitable articles, bringing the total number of articles to 1,848. An examination of the abstracts resulted in most being rejected as obviously unsuitable, usually because the duration of follow-up was very brief or because the topic of the article was unrelated to the objectives of this analysis. The remaining 149 articles were examined in greater detail. Only 30 appeared to meet the inclusion criteria and had a quality score greater than 10 (Table III). Five of these were excluded during a more detailed examination of the data undertaken by 2 of the authors (PS and WPS). In some of these, the authors kindly provided further information.3, 5
Table III. Details of 33 Cohorts from 30 Articles that Appeared to Meet the Inclusion Criteria and had a Quality Score Greater than 10
Total women years
Follow up (yr)
Posttreatment CIN confirmed by
Mitchell et al. was a case control study. All the rest were retrospective observational studies. Mettlin, Pettersson, Richart and Soutter were multicentre studies.
Cohorts excluded from analysis.
The first figure is the woman years of follow up for detection of CIN. The second figure relates to follow up for posttreatment cancer because this was performed using cancer registry data. Consequently complete follow up is assumed for all women upto 12 years.
The article by Bekassy3 was excluded because the “women years at risk” in any period in this study seemed to have been determined by the number of smear results reported in that period rather than the number of women under observation. A second study6 was excluded because it proved impossible to determine recurrence rates in different intervals in spite of initial appearances to the contrary. A third study5 was excluded because it was not possible to separate posttreatment CIN from recurrence as invasive cancer. Nor was it clear whether recurrences in the vagina after hysterectomy were counted separately or with second primaries of vagina and vulva. A 4th study7 was excluded because recurrences included cytological abnormalities of any degree including those with only HPV abnormalities. The fifth study8 was excluded because the data were already included in the earlier metaanalysis.1 The final analysis was based on 26 cohorts in 25 studies.
The eligible studies
All of the remaining studies were retrospective observational studies and most were based upon more than 1 institution. One very large study was based on data from the Swedish National Cancer Registry.9 Most of the articles described the results of one particular treatment method, but several described the outcome for women who had been treated with one of a number of methods (Table III). The latter was particularly the case in the large studies of invasive cancer rates. Ten of the women in the New Zealand study who developed cancer and 15 who did not were excluded from this analysis because they were not treated with any recognised, potentially effective therapy.10 The grade of the pretreatment histological diagnosis was not stated explicitly in 2 studies.11, 12 In the remainder, CIN III had been found in 32–100%. None of the studies included women treated for microinvasive disease. In nearly all the studies, the policy was to perform annual smears after an initial period of 12–24 months of more intensive surveillance with cytology. Colposcopy was usually included for part of that time at least. A histological diagnosis of posttreatment CIN or invasive cancer was required in 19 of the studies. In 3 of these, the posttreatment CIN had to be at least CIN II (Table III). In 7 studies, a dyskaryotic smear was used as an endpoint for at least some patients described as having posttreatment CIN.
Studies looking at the rate of cancer following treatment for CIN report variable rates (Table IV). There was no relationship between the year of publication and the rates of posttreatment invasive cancer (Kendall's tau b = −0.18, z (continuity corrected) = −1.14, two-sided p = 0.26). The groups containing a high proportion of women with CIN III had higher rates of invasive recurrence (Kendall's rank correlation tau b = 0.47, z corrected for continuity = 2.72, two-sided p = 0.007). Although 7 studies reported no cases of cancer, with fewer than 9,000 women years of follow-up in each study, the upper limit of their individual 95% confidence intervals is over 40 per 100,000 women years. The smaller studies with less than 2,500 women years' observation tended to have much larger variation between the studies. There was no other relationship between the size of the study and the rate of posttreatment cancer.
Table IV. Recurrence of Invasive Cancer Per 100,000 Woman Years
% CIN III
Rate per 100,000 woman yr
Rate after 5 yr
Rate after 10 yr
Years of follow up
% CIN III, % of women in the study treated for CIN III. NS, not stated. Rates after 5 yr or 10 yr are the cumulative annual rates after 5 or 10 years. ND, no data.
Because invasive cancer in women lost to follow up was identified from a register of cancer cases, the women years of follow up with respect to invasive cancer have been calculated on the assumption that all women were followed for a minimum of 12 years.
In this series, 35/195 hysterectomies were done for a variety of other reasons more than two years after the cone biopsy.
There was no statistically significant difference in the incidence of invasive recurrence between those series in which women were treated with hysterectomy and those in which one of the local, conservative methods of treatment was used (39 vs. 56 per 100,000 woman years, RR 0.69, 95% CL 0.27, 1.44, p = 0.33) nor were there any differences between the various conservative methods. The invasive disease was found at the vault of the vagina in 24 of the 26 women who developed invasive cancer after hysterectomy and whose clinical details were described.6, 10, 12, 13, 14, 15 Thus, nearly all would have been detectable by vault cytology. Indeed, an abnormal smear did identify 14 of the 15 who were having smears taken and whose presentation was described. In several cases, many years elapsed between the abnormal smear and the diagnosis of cancer.10, 15
Time-dependent changes in the rate of invasive recurrence
High rates of cancer were observed in the New Zealand study, in which posttreatment CIN was not necessarily treated.10 In that study, the rate of cancer was about 864 per 100,000 in the first year after treatment and 250–350 thereafter. In contrast, the Swedish National Cancer Registry study reported a rate of just over 50 per 100,000 women years (95% CI 46–61).9 The Swedish study did not report cancer rates in the first year after treatment, but included more women years of follow-up thereafter than all the rest of the studies combined. The incidence rate in the Swedish study was between 51 and 57 per 100,000 in each 5-year follow-up interval up to 20 years. In the remaining studies, the average cancer incidence rates were static at 63–69 per 100,000 woman years throughout the first 10 years. Excluding the New Zealand study, the combined rate for all the remaining studies were 69 per 100,000 woman years in the first year and 56 per 100,000 woman years until at least the 20th anniversary of treatment. These studies were conducted in 10 different countries between 1952 and 2001 during part of which time the background rate of invasive cervical cancer fell by upto 50%.9, 16, 17, 18, 19, 20, 21, 22, 23 In all, 91% of the women in the studies that reported invasive cancer had been treated in Sweden or UK. In the Pettersson study, the expected rate of cervical cancer, adjusted for age and birth cohort, was about 22 per 100,000.9 With the exception of that Swedish study, most women had been treated and followed up in the late 1980s and 1990s. In that period, the incidence in UK fell from about 25 to about 15 per 100,000.21 On that basis, the overall background rate of invasive cancer was estimated to be about 20 per 100,000.
The conclusion with respect to invasive cancer is that there is some danger of a cancer not detected at the time of treatment being diagnosed shortly after treatment. Thereafter, the risk of cancer in women being asked to attend for annual smears after treatment for CIN remains elevated for at least 10 years and probably for up to 20 years at a rate ˜2.8 times the background.
There was considerable variation in the incidence of posttreatment CIN between the studies, ranging from 76 to 6,036 per 100,000 woman years (median 1,413, lower quartile 627 and upper quartile 2,617) (Table V). The variation in this rate was greater in the first year of treatment than in subsequent years. This was partly because some studies distinguished between “residual disease” and “recurrent disease” and excluded the former. Smaller studies had higher rates of recurrence of CIN (Kendall's rank correlation coefficient tau = −0.38, z corrected for continuity = −2.48, two-sided p = 0.013). The 5 studies that used only a cytological endpoint had a higher incidence of posttreatment CIN than those that used a histological endpoint (2,704 vs. 967 per 100,000 woman years, rate ratio 2.80, 95% CL 2.41, 3.23). The incidence of posttreatment intraepithelial disease was lower in women treated by hysterectomy than in those treated by local therapy (397 vs. 1,605 per 100,000 woman years, rate ratio = 0.25, 95% CL 0.19, 0.32).
Table V. Recurrence of CIN per 100,000 Woman Years
% CIN III
Rate per 100,000 woman yr
Rate after 5 yr
Rate after 10 yr
Rates per 100,000 women in each year of follow up
% CIN III, % of women in the study treated for CIN III. NS, not stated. Rates after 5 yr or 10 yr are the cumulative annual rates per 100,000 after 5 or 10 years. ND, no data. The article by Robertson et al. appears twice in this table because those treated by hysterectomy are listed separately from those treated by knife cone biopsy.
In these series overall rates for a number years were reported. The rates shown are estimates based on theses values.
In this series, 35/195 hysterectomies were done for a variety of other reasons more than two years after the cone biopsy.
It was not possible to analyse fully the effect of the grade of CIN treated on the recurrence rates, but there was no relationship between the proportion of women treated for CIN III in a study and the incidence of posttreatment CIN (Kendall's rank correlation tau b = −0.02, z corrected for continuity = −0.12, two-sided p = 0.91). There was no relationship between the year of publication and the rates of posttreatment CIN (CIN: Kendall's tau b = −0.04, z (continuity corrected) = −0.21, two-sided p = 0.83). There was no evidence that higher rates of detection of posttreatment CIN were linked to reduced rates of invasive recurrence (Kendall's rank correlation coefficient tau b = 0.20, z (continuity corrected) = 1.20, two-sided p = 0.23).
Time-dependent changes in the rate of posttreatment CIN
The data for posttreatment CIN were modelled using Poisson regression with a separate underlying rate for each study and a common curve to model the effect of time since treatment on the relative risk of recurrence. This model fitted the data poorly. The primary reason for the lack of fit was the variation between studies in the relative rates of recurrence in the first year after treatment compared to subsequent years. When the first year of follow-up was excluded, the model fitted reasonably well (but not perfectly). Figure 1 illustrates the rates for each of the 4 largest studies (6,000 women years or more excluding year 1) together with that estimated from the 9 very small studies (less than 2,000 women years of follow-up after year 1); the 8 small studies (2,000–3,999 women years of follow-up after year 1); the 2 medium studies (4,000–5,999) and the overall estimate. Although there is some variation between the individual plots, all plots are reasonably consistent with the overall result, showing that the risk of recurrence in year 3 is approximately half of that in year 2. The risk then continues to decline slowly, halving again over the next 5 years or so. Table VI provides the year-by-year relative risk estimates for all studies combined. None of the studies provide sufficient data on excision margins to assess the impact of this factor on the long-term rate of posttreatment CIN or invasive cancer.
Table VI. Relative Risk of Recurrence Compared to the Risk of Recurrence During Year 2, Estimated from all Studies on Recurrence Rate of CIN
95% confidence interval
The aim of an effective programme should be to reduce the risk of invasive cancer in treated women to a level close to or below the background rate in the population. However, it is possible that those who have developed CIN once may be at a higher risk of developing a second primary lesion and that the cancer rates may remain higher in these women. Unfortunately, it is not possible to distinguish between a reoccurrence of the original lesion and new disease.
The main findings
These data show clearly that the rate of invasive cancer following treatment for CIN remains undiminished throughout the period of follow-up at about 56 per 100,000, a rate substantially greater than that found in the general population. The population rates of cervical cancer have fallen steeply during the period in which these studies have been conducted and one concern was that this might have influenced the rates of invasive recurrence. However, there is no relationship between the recurrence rates and the epoch in which the patients were observed. As further confirmation of these data, the rate of subsequent invasive cervical or vaginal cancer was 48 per 100,000 in women registered as having CIN III, with the Thames Cancer Registry between 1960 and 1999.24
In contrast to invasive disease, the rate of posttreatment CIN falls throughout the first 10 years after treatment to a rate of about 280 per 100,000 women in the 8th year and to about 190 per 100,000 women in the 10th year. Thereafter, there are too few women under observation to provide reliable estimates of the rate. To put these figures into perspective, the rate of high grade disease in a population of screened women of similar age in England is about 400 per 100,000.25
Evidence from other sources
Confirmatory evidence of the falling rate of CIN with time after treatment is provided by a recent, very large study of 6,849 Australian women, with 42,463 woman years of observation for a mean of 6.2 years follow-up, which could not be included in this analysis because the number of women under follow-up at different times after treatment could not be determined.26 In that study, the rate of posttreatment CIN was highest in the first year, but then fell at a steady rate up to about 9 years. At that point, the rate seemed to rise abruptly, probably because of the relatively small numbers of women followed for more than 9 years and to differential ascertainment with more frequent smears being taken from those at highest risk of recurrence. Excluding the data after year 9, there is some evidence that the rate falls more slowly towards the end of this period. The same pattern of a falling risk of recurrence of CIN can be seen in the very large study by Mettlin et al. in which the risk relative to year 2 changes little than in years 3–5, but has fallen to 0.34 by years 8–10.5 Although these recurrences include invasive disease, the great majority would have been CIN.
The Australian study also showed that the rate of invasive recurrence remained the same during the first 10 years of follow-up.26 Relatively small numbers towards the end of the period of observation may make the rates in the last year or 2 less precise. Others have recently described the risk of late invasive recurrence, with 5 of 6 invasive recurrences being diagnosed more than 5 years after knife conisation for CIN III.27 Two of these were detected 12 and 23 years after the treatment.
Weaknesses of the study
The weakness of this analysis is the relatively scanty information provided in most of the articles and the small number of women studied in many. This applies particularly to the data on women treated by hysterectomy many of whom may have been treated without prior colposcopy. The lack of clear information or the discovery of inappropriate methods of analysis led to the exclusion of several studies. In spite of these difficulties, the data show clearly that the rate of detection of posttreatment CIN falls over the first 7–10 years after treatment, while the rate of posttreatment invasive disease remains elevated.
Interpretation of findings
It might have been expected that the rate of detection of posttreatment CIN would fall gradually over time. Some lesions not fully removed by treatment might take some time to regrow before reaching a detectable size, especially if the residual lesion is very small. In addition, the sensitivity of cervical cytology is only of the order of 50%,28 and so several tests might be required to detect these residual lesions. HPV testing is probably a more sensitive test in this situation, but there was variability in the results described in a recent metaanalysis.29 Specificity was also a problem. What will be needed are long-term studies with multiple testing of large cohorts of women. Some believe that posttreatment CIN detected more than 1–2 years after treatment is the result of a new lesion developing rather than regrowth of the original abnormality. There is no evidence to confirm or refute this idea, but the result is the same from the woman's point of view.
It is very disappointing that the rate of invasive disease posttreatment remains elevated, in spite of falling rates of detection of posttreatment CIN. This suggests that many cases of post treatment CIN are not being detected. It is possible that the rate of invasive cancer is inflated by differential ascertainment, in that women who no longer participate in follow-up may return to the study only when they develop invasive cancer. However, that would not apply to the Swedish National Cancer Registry study in which all the women in the study were assumed to be under follow-up and to contribute to the denominator. Although it is possible that these are new lesions that have developed de novo without going through a premalignant stage, there seems to be no evidence to support that idea.
Another possible explanation for this observation is that the process of detecting posttreatment CIN becomes less effective with the passage of time because these later cases arise from areas of buried CIN that are not readily detected.30 If this were so, it might apply particularly after hysterectomy but nearly all of these lesions appeared at the vault and most seem to have been detected at an early, asymptomatic stage when effective treatment was still possible.
It seems very likely that the most important cause of the persistently raised cancer rate is an increasing rate of default from follow-up over time. All of the cohort studies in this analysis refer to substantial loss to follow-up and several mention women presenting with invasive cancer after defaulting from follow-up many years before. Others have found that half of those who develop invasive cervical cancer after previous treatment for CIN had been lost to follow-up.31 In addition, we have recently completed a study of compliance with follow-up after treatment for CIN (manuscript in preparation). This shows that compliance deteriorates steadily with time since treatment.
The development of invasive cervical cancer after treatment for CIN is a significant problem. We estimate that about 390 (16%) of the ˜2,400 women in England who develop invasive cervical cancer each year will have been treated for CIN previously. Even in 1982–1991 in Aberdeen when 83% of cases occurred in unscreened or poorly screened women, 6% of the women with invasive cancer had been treated previously for CIN.31
The data presented here show that women treated for CIN remain at a substantially increased risk of subsequently developing invasive cervical cancer even when the rate of detection of posttreatment CIN appears to have fallen. It seems most likely that the sustained rate of invasive disease is due to reduced compliance with follow-up. Although there is no direct evidence to confirm the value of annual smears after treatment, it would be expected that more frequent screening of treated women would increase the efficacy of detection of preinvasive disease, as it does in the general population where decreasing the interval between smears from 3 years to 1 year halves the risk of invasive squamous cell cancer.32
Those who advocate reducing the frequency of smears after 5 years do so because they believe this would be welcomed by women and would reduce the costs of the programme. But this would only be acceptable if women could be assured that their safety was not being compromised. These data provide no grounds for such an assurance but suggest instead that encouraging women to persevere with regular annual smears for 10 years after their treatment would offer them the best chance of detecting any recurrence at a treatable stage. The most recent NHS Cervical Screening Programme Guidelines have been amended to reflect this information.33 Because women treated by hysterectomy appear to have a similar risk of invasive recurrence to those who have received conservative therapy, it seems prudent that this small number of women should be followed up in the same way. There is a need for large studies to evaluate the organisation of the follow-up of women treated for CIN.