Int. J. Cancer: 71, 4–8 (1997) r 1997 Wiley-Liss, Inc. Publication of the International Union Against Cancer Publication de l’Union Internationale Contre le Cancer TREATMENT OF PRE-INVASIVE CONDITIONS DURING OPPORTUNISTIC SCREENING AND ITS EFFECTIVENESS ON CERVICAL CANCER INCIDENCE IN ONE NORWEGIAN COUNTY Siri FORSMO1*, Harald BUHAUG1, Finn Egil SKJELDESTAD2 and Olav A. HAUGEN3 Institute for Hospital Research, Trondheim, Norway 2Department of Obstetrics and Gynecology, University Hospital of Trondheim, Trondheim, Norway 3Department of Pathology, University Hospital of Trondheim, Trondheim, Norway 1Norwegian Norway had until recently no organized screening programme for cervical cancer, but opportunistic screening was common. This study focuses on the effectiveness of treatment of pre-malignant cervical conditions (CIN III) on cervicalcancer incidence in the county of Sør-Trøndelag in Norway, prior to the introduction of organized mass screening. The study is based on cervical-cancer incidence rates during the years 1965–92 and treatment data for CIN III. The expected number of cervical-cancer cases prevented due to early intervention was expressed in a regression model with 2 unknown parameters: the probability, p, of cancer development in case of CIN III, and the time lag, t, between treatment and when clinical cancer would otherwise have been diagnosed. The estimated probability that a patient treated for CIN III would have developed cervical cancer if not treated was found to be approximately 20%, and the mean time delay was around 16 years. In the last period of study (1988–92), the incidence was reduced by nearly 40% of what would have been expected without early intervention. Based on equal treatment rates as in 1990, parameter estimates were used to predict future incidence reduction. Maximum effectiveness will be achieved around the year 2005, with a nearly 70% reduction. Opportunistic screening and treatment of CIN III seems to have had considerable influence on cervical-cancer incidence. The costs, however, are substantial over-treatment, since our results indicate that 4 of 5 women treated for CIN III would not progress into the invasive state. Int. J. Cancer, 71:4–8, 1997. r 1997 Wiley-Liss, Inc. In most industrialized countries, the incidence and mortality of cervical cancer have been declining over the last 25 to 30 years (Hakama, 1982; Parkin et al., 1993). It is commonly believed that regular Pap testing, which makes possible the diagnosis and treatment of pre-malignant cervical conditions, is the main reason for this reduction (Hakama et al., 1985; Miller, 1986). Although the effectiveness of Pap testing has never been evaluated in any randomized trial, it seems that the observed reduction has been most noticeable in populations subject to organized mass screening (Hakama et al., 1985; Miller, 1986). Reduced incidence of cervical cancer also has been observed in populations without a screening program, as is the case in Norway. It is assumed that the majority of the cervical cancers develop via a pre-malignant and reversible stage, cervical intraepithelial neoplasia (CIN), graded CIN I–III, according to severity (Richart, 1968). Without detection and treatment of precursor lesions, the maximum prevalence rate of carcinoma in situ (included in CIN III) is believed to manifest itself in women around the age of 35 years (Boyes et al., 1962), and the maximum incidence rate of invasive cancer appears about 10 years later. Development from a pre-malignant condition into invasive cancer is usually a slow process affecting only a minority of women with CIN. Hence, it will take many years to observe the effectiveness on cervical-cancer incidence rates of tracing and treating pre-invasive disease. An indirect approach to evaluating the efficiency and effectiveness of screening and early intervention is the estimation of progression time and transition rates from the pre-malignant to the malignant state in simulation models of the natural history of cervical cancer (Gustafsson and Adami, 1989, 1992; Yu, 1982; Parkin, 1985; Prorok, 1986; Van Oortmarssen and Habbema, 1995). Until 1992, there was no organized nation-wide screening program for cervical cancer in Norway. During the period 1965–85, however, a 10-fold increase in the annual number of Pap smears was observed (Norges Offentlige Utredninger, 1987). This was due mainly to frequent opportunistic Pap testing, especially of women of child-bearing age (Forsmo et al., 1994). In most of the other Nordic countries, organized screening was introduced about 30 years ago, and cervical-cancer incidence has been decreasing since the mid-1960s (Hakama, 1982). In Norway, the decline in cervicalcancer incidence occurred about 10 years later than in the neighbouring countries, and has been slower. Lack of a preventive strategy such as organized mass screening has been held as the most plausible explanation of this difference (Anonymous, 1985). The most important incidence reduction over the last 20 years has taken place in the 40-to-55-year age group, and maximum incidence rates are now observed in women aged 60 to 65 years (Bjørge et al., 1993). This study focuses on the effect of early intervention on cervical cancer in one Norwegian county (Sør-Trøndelag, mid-Norway) during the period 1965–92. In the mid-1960s, cervical-cancer incidence in Sør-Trøndelag was among the highest in Norway (relative rate 1.4), but has been decreasing since then. The incidence of cervical cancer in the county today is approximately equal to the national cervical-cancer incidence (Cancer Registry of Norway, 1995). In a regression model, we aimed to estimate the effectiveness of treatment of cervical premalignancies prior to the introduction of organized mass screening, measured as cervical cancer incidence reduction. MATERIAL AND METHODS The study is based on incidence data on pre-invasive and invasive cervical cancer and on treatment records from the county of Sør-Trøndelag, which had a total female population of 127,000 in 1990. Cervical-cancer incidence data for the years 1965 to 1992 were provided by the Norwegian Cancer Registry. For the analyses, the cases were grouped by age in 5-year increments. Effect on cervical-cancer incidence of treatment for pre-invasive cervical conditions was measured in 3 periods of 5 years from 1978 to 1992. The years 1965 to 1977 were used as a reference period to measure incidence rates before intervention had had any major influence. The length of the reference period is necessary in order to obtain precise estimates, and also a period close in time to the study periods in order to minimize the effect of possible incidence trends. Surgical intervention records, e.g., cervical conization and hysterectomies during the years 1986 to 1992 were collected from *Correspondence to: SINTEF UNIMED Norwegian Institute for Hospital Research, N-7034 Trondheim, Norway. Fax: 1 47 73596361. E-mail: firstname.lastname@example.org. Received 25 July 1996; revised 14 November 1996 EFFECTIVENESS OF TREATING CERVICAL PRE-CANCER 5 TABLE I – AGE-SPECIFIC INCIDENCE OF CERVICAL CANCER IN THE COUNTY OF SØR-TRØNDELAG, 1965 TO 1992, AND MEAN AGE AT DIAGNOSIS1 Age (years) 20–29 30–39 40–49 50–59 60–69 70–79 801 Total Age-adjusted Mean age 1The 1965–1977 N 6 51 98 104 69 46 16 390 54.8 1/100,000 2.3 37.7 59.1 56.5 34.8 44.0 32.9 25.3 21.2 1978–1982 N 1/100,000 10 15 19 24 26 21 8 123 10.9 18.1 32.9 35.5 34.3 43.2 35.9 19.9 15.1 56.2 1983–1987 N 7 17 26 12 22 24 14 122 57.3 1/100,000 7.5 19.2 38.5 20.4 32.5 46.1 51.8 19.5 14.1 1988–1992 N 5 29 11 18 28 14 12 116 1/100,000 5.1 32.2 13.1 32.3 43.3 25.6 39.5 18.2 13.4 55.8 years 1965 to 1977 represent the baseline period. computerized registries and medical records at the University Hospital of Trondheim. As part of a larger project, a record linkage between the computerized registry at the Department of Pathology and the surgery records was established in order to create a complete registry of all Pap smears, histological investigations, conizations and hysterectomies for the period 1986 to 1992. As the only pathology laboratory in the region, the coverage is virtually complete for women resident in Sør-Trøndelag. A system of automatic calls for follow-up smears at specific intervals in the event of a positive smear result was introduced in 1986. From the registry records we know that virtually all histologically confirmed cases of CIN III were treated, and we assumed that this was also the case for patients prior to 1986. Annual rates of CIN III (severe dysplasia and carcinoma in situ) diagnosed at the laboratory between 1975 and 1985 were used to estimate annual treatment rates. In the years 1975 to 1992, the annual number of carcinomas in situ varied in proportion with the number of CIN-III cases, and accounted for about 50% of the cases treated for CIN III. Assuming that this was also true prior to 1975, treatment rates in the period 1961 to 1974 were estimated on the basis of in situ cases recorded by the Cancer Registry since 1961. Since 1983, the preferred treatment of pre-malignant cervical conditions has been laser conization. About 90% of the cases receiving conization were treated after the diagnosis of CIN III, but a few cases of CIN I, II and invasive cancer are recorded. This study includes treatment rates of CIN III only. Statistical methods CIN III involves a risk of cancer progression. For the analyses, we assume a probability that treatment of one case of CIN III will prevent one case of invasive disease. The time lag between treatment of a progressive CIN-III case and the time at which invasive cancer would otherwise have been diagnosed is assumed to be a random variable, i, with a probable density function f(i). Thus, we express the expected number of prevented cancer cases in year t in terms of the number of CIN-III cases treated by (conization or hysterectomy) in previous years: Prev(a,t) 5 p S i Con(a-i,t-i) f(t) Surv(a-i,a) 5 p X(a,t) (1) where Prev(a,t) is the prevented number of cervix-cancer cases of age a in year t (1/100,000); p is the probability that treatment of pre-invasive disease will prevent a future case of invasive cervical cancer; Con(a-i,t-i) is the number of cases of age a-i treated for preinvasive cancer in the year t-i (1/100,000); f(t) is the probability that a case of invasive cancer, prevented due to intervention, would have been diagnosed t years later if not prevented; Surv(a-i,a) is the probability that the patient will survive from age a-i to age a and will not be subject to a hysterectomy during this period; X(a,t) is the result of Si (Con(a-i,t-i) f(t) Surv(a-i,a). We used a Poisson distribution to represent the ‘‘delay function’’ f(t) after testing series of distributions belonging to the Erlang family based on an exponential distribution of the time delay. With the actual parameters, the Poisson distribution is a close approximation to the normal distribution. The survivor function Surv(a-i,a) is based on female age-specific mortality rates from official statistics and on hysterectomy rates for reasons other than pre-invasive and invasive cancer of the cervix. Mortality and hysterectomy rates are relatively low in the relevant age groups, and the inclusion of this function has only a minor effect on the final results. There was practically no net migration to or from the county during the study years. Model parameters were estimated by the least-squares method, using the following equation: Y(a,t) 5 c 1 p X(a,t) 1 ea,t (2) where Y(a,t) is the reduction in the incidence of cervical cancer at age a, measured as the difference in incidence between the reference period and the year t. The constant term c represents a possible general shift in the incidence level from the reference to the study period. A negative value of c indicates an increase in cancer incidence. The parameter ea,t is an error term. X(a,t) depends on the delay function f(t), and the equation is not linear in the unknown parameter t. Ordinary linear regression cannot be applied, and the parameters are estimated by minimizing the sum of squares, using an iterative method. Confidence intervals are estimated by Monte Carlo simulation (Press et al., 1988). Based on the model and the parameter estimates, simulation is used to generate a number of invasive cases for each age group and time period. The simulated numbers are used to calculate new incidence rates which are fed back into the model, and new parameter estimates are computed. This procedure was repeated 100 times, and the sample of estimates for each parameter was used for the calculation of 95% confidence intervals (CI). RESULTS Age-specific cervical-cancer incidence rates and age-adjusted incidence throughout the study period 1965 to 1992 are presented in Table I. There was a marked reduction in incidence, most pronounced in the first period after the reference period 1965 to 1977 (Table I). Registration rates for carcinoma in situ in women from Sør-Trøndelag in the Cancer Registry between 1961 and 1992 are presented in Table II. Until the beginning of the 1980’s, there was a steady increase in carcinoma-in situ registrations, but about 10 years later the registration rates seem to be decreasing. Mean age of patients with in situ cancer decreased from 44 years in the early 1960’s to 35 years in the last period, equal to the mean age of women treated for pre-invasive disease. Least-square estimates of model parameters with 95% CI for the 3 successive time periods are presented in Table III. Mean time FORSMO ET AL. 6 TABLE II – REGISTRATION RATES (1/100,000) OF CARCINOMA IN SITU IN THE NORWEGIAN CANCER REGISTRY IN THE COUNTY OF SØR-TRØNDELAG DURING THE YEARS 1961 TO 19921 1961–1967 Age N 20–29 30–39 40–49 50–59 60–69 70–79 801 Total Age-adjusted Mean age1 1Mean 0 13 17 4 0 2 0 36 1968–1972 1/100,000 0 15.4 16.3 4.1 0 4.3 0 4.7 4.4 44.2 N 1973–1977 1/100,000 21 39 37 9 4 1 0 111 N 24.0 67.1 53.3 12.4 6.6 2.5 0 18.8 20.3 63 83 42 20 6 1 2 217 41.6 1978–1982 1/100,000 68.2 121.0 69.3 27.5 9.4 2.2 11.2 35.7 36.8 N 144 160 55 30 6 0 2 397 37.2 1983–1987 1/100,000 156.0 192.5 95.4 44.5 9.0 0 8.9 64.7 64.4 35.1 N 1988–1992 1/100,000 148 154 45 27 16 7 1 398 157.9 173.7 66.5 45.9 23.7 13.5 3.7 64.1 60.9 N 117 130 50 11 6 4 0 318 35.9 1/100,000 118.4 144.8 61.1 21.1 12.7 7.7 0 46.5 50.5 35.0 age at registration. TABLE III – OUTCOME OF REGRESSION MODEL FOR 3 SUBSEQUENT 5-YEAR PERIODS Time periods 1978–1982 1983–1987 1988–1992 Delay before invasive cancer (T)1 Probability of future cancer if no treatment2 Constant term Years (95% CI) 9.8 16.1 15.3 (7.8–11.7) (13.5–18.7) (14.1–16.6) 20.7 1.5 22.7 (95% CI) p (95% CI) R2 (25.4–3.9) (21.5–4.5) (25.0–20.4) 0.21 0.27 0.18 (0.08–0.34) (0.14–0.40) (0.14–0.22) 0.38 0.27 0.60 1Mean delay between treatment and when invasive disease would have been diagnosed without intervention.–2Probability that a woman treated for CIN III would have developed invasive disease if not treated. TABLE IV – NUMBER OF CERVICAL-CANCER CASES PREVENTED DUE TO TREATMENT OF CIN III IN WOMEN AGED 25 TO 79 YEARS IN 3 SUCCESSIVE TIME PERIODS Period Number observed Number prevented Percentage reduction1 1978–1982 1983–1987 1988–1992 112 105 104 37 38 62 24.7 26.8 37.3 1Estimated percentage incidence reduction due to intervention. delay until patients treated for CIN III would have developed invasive cancer was around 16 years for the 2 last periods. In the first period 1977 to 1982, the mean time delay was significantly shorter, about 10 years. The estimated probability ( p) that a patient treated for CIN III would later have developed cancer if not treated was found to be approximately 20% in the 3 study periods. The confidence limits were, however, rather broad in the 2 first periods (Table III). The constant term c measures the change in incidence from the reference to the study period that would have occurred if no CIN-III cases had been treated. Only in the last period does c attain a statistically significant negative value, which indicates that an increase in expected incidence would have taken place if no screening or treatment of CIN III had been performed. Results from the first period should be interpreted bearing in mind that the incidence of cervical cancer in these years had been less influenced by pre-invasive treatment than in the later periods. The minor effect of prevention is probably partly disguised in random incidence variations. We consider the results from the 2 last periods, and especially the last one, to be the most valid. For the last period (1988–92), the estimates had narrow confidence limits, and the coefficient of determination (R2) improved markedly. This indicates that the model better fits the data in this period than in previous periods. Results based on all 3 periods pooled are not presented, as the results from the periods analyzed individually differed significantly. Table IV presents the estimated number of cervical-cancer cases prevented by treatment of CIN III and the percentage incidence reduction for women aged 25 to 79 years during the 3 study periods between 1977 and 1992. In the first period, prevention was most pronounced in younger age groups, with maximum effect around 45 years (data not shown). In the next periods (1983–92), maximum effect was skewed towards older age groups, and the greatest total effect was observed in the last period, with maximum incidence reduction in women between 40 and 59 years. Parameter estimates from the period 1988 to 1992 were used to predict the number of cancer cases prevented in the future, assuming the same incidence of CIN III and treatment rates as in 1990. Maximum effect of pre-invasive treatment on cervicalcancer incidence will be reached around the year 2005, with almost 70% reduction in the incidence that would be expected without early intervention. DISCUSSION The results of the regression model indicate that opportunistic screening and treatment of CIN III have led to reduced cervicalcancer incidence in the county of Sør-Trøndelag over the last 15 years. Because of the time delay before the results of pre-invasive treatment on cervical-cancer incidence can be observed, only part of the total improvement in incidence reduction has been revealed so far. Method The study is based on a small (6%) and delimited population in Norway, and random incidence variations of pre-malignant conditions and cervical cancer may influence the outcome of the regression model. If similar data were available for the whole nation, we believe our method would be valid for evaluation of secondary-cervical-cancer prevention on a national scale. Registration of carcinoma in situ may not have been complete in the Cancer Registry, especially in the early 1960s. Thus, true treatment rates may have been higher than those applied in the model. Increased treatment rates would have given lower probability ( p) of cancer progression and increased time delay before clinical cancer. It can be argued that an exponential or Erlang-family distribution of the time delay would seem more appropriate than the Poisson distribution. These distributions all came out with consistently poorer results than the Poisson distribution. EFFECTIVENESS OF TREATING CERVICAL PRE-CANCER As indicated by the registration rates of carcinoma in situ, there was some opportunistic screening during the last years (1968–72) of the reference period. This may have led to increased detection of invasive cervical cancer, hence a larger difference in incidence between the reference period and subsequent time periods than without screening. However, cervical-cancer incidence in SørTrøndelag during this period was actually lower than during the previous 5-year period, 1963 to 1967 (Cancer Registry of Norway, 1995). Although it remains a possibility, screening detected cancer cases during the reference period do not seem to represent a major bias to the results. For the analyses, we assumed that previous treatment of pre-malignant disease had not yet had any major influence on the cervical-cancer incidence rates in the reference period (1965–77). This is not necessarily true, especially in the last years of the reference period. Although treatment rates were very low in the years prior to this, some prevalent pre-malignant conditions destined to surface as invasive cancer during the reference years might have been eliminated. These would have been prevalent cases close to the point of malignant transition or cases of fast-progressing pre-invasive disease. The last category is rare (IARC Working Group, 1986). Probably less than 10% of invasive cases develop within 5 years (Gustafsson and Adami, 1989; Albert, 1981), and we do not think they represent a major bias in our results. Because of very low carcinoma-in situ registration and treatment rates in the early 1960s, we believe that the number of prevalent CIN-III cases prevented from emerging as invasive cervical cancer during the reference period was very small. In order to minimize the effect of general trends in cervical-cancer incidence, we preferred to keep the baseline and study periods close in time. Test sensitivity and screening frequency may have varied during the study periods, and this would affect estimation of the parameters, primarily of p.The actual results indicate that such variation is small compared with other sources of variation. The mean age of carcinoma-in situ patients registered was virtually equal throughout the study periods, which is one argument against important fluctuations in test validity. Screening for cervical cancer affects cancer incidence in 2 ways: (i) reduction is caused by detection and treatment of pre-invasive conditions; (ii) screening contributes to early detection of invasive cases. Early detection implies a shift in the age distribution towards younger age. As the amount of testing was much higher in the study periods than in the reference period, estimation of time delay between treatment and diagnosis will be biased by earlier diagnosis. This new parameter was included in an elaborate model. However, because of limited data in a model that included 2 more parameters, we were not able to produce precise estimates. We believe that this factor must be taken into account when evaluating the time delay. Natural history Estimations in mathematical models of the duration of premalignant conditions have yielded values ranging from 5 to nearly 30 years (Prorok, 1986; Eddy, 1987). In more recent studies (Gustafsson and Adami, 1989; Van Oortmarssen and Habbema, 1995), a mean duration of 16 or 17 years from the onset of carcinoma in situ until the eventual development of clinical invasive cancer seems to fit the data on the natural history of cervical cancer. Our results, comparable to these estimates, represent the mean duration from the diagnosis of CIN III until clinical cancer. As CIN III includes both severe dysplasia and carcinoma in situ, it is possible that the average time delay before clinical cancer is longer than for carcinoma in situ only. The ratio between prevalent and incident cases treated for CIN III was presumably higher in the first years of opportunistic testing than later. The cases might have spent longer in the pre-malignant phase than cases diagnosed later. One indication is the mean age of patients with carcinoma in situ, which was higher in the 1960s than 7 a decade later. We do not know the proportion of CIN-III cases that were diagnosed based on genital symptoms. Spontaneous testing of symptom-free women was infrequent in the 1960s compared with 15 to 20 years later, when the intensity of opportunistic screening was high, especially in the age groups that empirically show a high incidence of CIN III. We believe that the relative number of women tested on indication was much higher in the early years. This represents a confounder in the estimation of time delay and may explain, although the estimates were less reliable, the shorter time delay in the first study period. The elimination of cases with a longer pre-malignant history may explain the pronounced incidence reduction in the first study period. In the last period, which produced the most valid results, we found that 18 of 100 women treated for CIN III would have developed cervical cancer if left untreated. Treatment affected prevalent and also incident cases. The likelihood of preventing cancer when treating a new case of CIN III will be lower, as demonstrated by Gustafsson and Adami (1989). Where there were no screening measures, they found that progression to cancer will occur in 15 to 23% of prevalent in situ cases and in about 12% of new in situ cases. Age should be included as a possible factor of progression from CIN to cancer in the study of the natural history (Morrison et al., 1996). In a mathematical model based on British Columbia data (Van Oortmarssen and Habbema, 1991), the best fit was achieved in a model with lower regression rates in women aged 34 years or more than in women of younger age. Owing to the organization of our data and a small population in our study, no conclusive results could be procured when assuming agedependency for the parameter p in our model, which was then interpreted as an average. Our results appears to confirm that secondary prevention of cervical cancer involves substantial overtreatment, as only a proportion of the patients will develop cancer if left untreated (Gustafsson and Adami, 1989; Bergström et al., 1993). Besides the economic aspects of unnecessary treatment, it also represents an ethical dilemma (Anonymous, 1985; Skegg, 1995). Reduction in cervical-cancer incidence According to our study, an incidence reduction of 65 to 70% is expected with present treatment rates. However, a possible 90% incidence reduction with screening intervals of 3 years has been estimated (IARC Working Group, 1986). This must imply increased detection and treatment rates of pre-invasive conditions, given equal association between the diagnostic criteria of CIN and the specificity in detecting lesions likely to develop into cancer as today. The main difference between the automatical recall system in the event of positive smears and an organized screening programme, was that a woman was not invited unless she had had a positive smear registered. Despite the absence of an organized programme, a substantial proportion of the female population aged ,50 years is already regularly tested, about 73% at least once in a 3-year period (Forsmo et al., 1994). Older women have Pap smears taken less frequently. Increased detection of pre-invasive disease relies on higher attendance rates than today, if the prevalence of CIN in women rarely or never tested is similar to or higher than in women who regularly have smears taken. We believe that organized screening will lead to an incidence reduction beyond the predictions in this study, though the greatest reduction will be due to previous opportunistic screening. The Pap test has a low level of accuracy (Fahey et al., 1995), and a negative smear is no guarantee against development of cervical cancer (Morrison et al., 1996). It has been demonstrated that the eagerness to trace all cases and avoid false-negative results in organized screening programs may lead to reduced test specificity and to increased detection and treatment of minor abnormalities and conditions at lower risk for malignant transition (Bergström et 8 FORSMO ET AL. al., 1993; Skegg, 1995; Raffle et al., 1995). This is of particular concern, since our study indicates that 4 of 5 women treated for pre-malignant cervical conditions subsequent to opportunistic screening would not develop invasive cancer. ACKNOWLEDGEMENTS We thank Dr. B.K. Jacobsen, Institute of Community Medicine, University of Tromsø, Norway for valuable comments. REFERENCES ALBERT, A., Estimated cervical-cancer-disease state incidence and transition rates. J. nat. Cancer Inst., 67, 571–576 (1981). ANONYMOUS, Cancer of the cervix: death by incompetence. Lancet, II, 363– 364 (1985). BERGSTRÖM, R., ADAMI, H.O., GUSTAFSSON, L., PONTÉN, J. and SPARÉN, P., Detection of pre-invasive cancer of the cervix and the subsequent reduction in invasive cancer. J. Nat. Cancer Inst., 85, 1050–1057 (1993). BJØRGE, T., THORESEN, S. and SKARE, G., Incidence, survival and mortality in cervical cancer in Norway, 1956–1990. Europ. J. Cancer, 29A, 2291– 2297 (1993). BOYES, D.A., FIDLER, H.K. and LOCK, D.R., Significance of in situ carcinoma of the uterine cervix. Brit. Med. J., 244, 203–206 (1962). CANCER REGISTRY OF NORWAY, Cancer in Norway 1992. Kreftregisteret, Oslo (1995). EDDY, D.M., The frequency of cervical cancer screening. Comparison of a mathematical model with empirical data. Cancer, 60, 1117–1122 (1987). FAHEY, M.T., IRWIG, L. and MACASKILL, P., Meta-analysis of Pap test accuracy. Amer. J. Epidemiol., 141, 680–689 (1995). FORSMO, S., BUHAUG, H. and STALSBERG, H., Use of Pap-smears in a population without a mass screening program. Acta Obstet. Gynecol. Scand., 73, 824–828 (1994). GUSTAFSSON, L. and ADAMI, H.O., Natural history of cervical neoplasia: consistent results obtained by an identification technique. Brit. J. Cancer, 60, 132–141 (1989). GUSTAFSSON, L. and ADAMI, H.O., Optimization of cervical-cancer screening. Cancer Causes Control, 3, 125–136 (1992). HAKAMA, M., Trends in incidence in Nordic countries. In: K. Magnus (ed.), Trends in cancer incidence. Causes and practical implications, pp. 279–292, Hemisphere Publishing Corporation, Washington (1982). HAKAMA, M., CHAMBERLAIN, J., DAY, N.E., MILLER, A.B. and PROROK, P., Evaluation of screening programmes for gynaecological cancer. Brit. J. Cancer, 52, 669–673 (1985). IARC WORKING GROUP ON EVALUATION OF CERVICAL-CANCER SCREENING PROGRAMMES, Screening for squamous cervical cancer: duration of low risk after negative results of cervical cytology and its implication for screening policies. Brit. Med. J., 293, 659–664 (1986). MILLER, A.B., Evaluation of the impact of screening for cancer of the cervix. In: M. Hakama, A.B. Miller and N.E. Day (eds.), Screening for cancer of the uterine cervix, pp. 149–160, IARC Scientific Publication IARC, 76, Lyon (1986). MORRISON, B.J., COLDMAN, A.J., BOYES, D.A. and ANDERSON, G.H., Forty years of repeated screening: the significance of carcinoma in situ. Brit. J. Cancer, 74, 814–819 (1996). NORGES OFFENTLIGE UTREDNINGER, Masseundersøkelse for kreft i livmorhalsen (Mass screening for cancer of the uterine cervix). NOU, 8, Universitetsforlaget, Oslo (1987). (In Norwegian) PARKIN, D.M., A computer simulation model for the practical planning of cervical-cancer screening programmes. Brit. J. Cancer, 51, 551–568 (1985). PARKIN, D.M., PISANI, P. and FERLAY, J., Estimates of the worldwide incidence of eighteen major cancers in 1985. Int. J. Cancer, 54, 594–606 (1993). PRESS, W.H., FLANNERY, B.P., TEUKOLSKY, S.A. and VETTERLING, W.T., Confidence limits on estimated model parameters. In: W.H. Press, B.P. Flannery, S.A. Teukolsky and W.T. Vetterling (eds.) Numerical recipes, pp. 529–538, Cambridge University Press, Cambridge (1988). PROROK, P.C., Mathematical models and natural history in cervical-cancer screening. In: M. Hakama, A.B. Miller and N.E. Day (eds.), Screening for cancer of the uterine cervix, pp. 185–196, IARC Scientific Publication IARC, Lyon (1986). RAFFLE, A.E., ALDEN, B. and MACKENZIE, E.F.D., Detection rates for abnormal smears: what are we screening for? Lancet, 345, 1469–1473 (1995). RICHART, R.M., Natural history of cervical intraepithelial neoplasia. Clin. Obstet. Gynecol., 10, 748–784 (1968). SKEGG, D.C.G., Cervical screening blues. Lancet, 345, 1451–1452 (1995). VAN OORTMARSSEN, G.J. and HABBEMA, J.D.F., Epidemiological evidence for age-dependent regression of pre-invasive cervical cancer. Brit. J. Cancer, 64, 559–565 (1991). VAN OORTMARSSEN, G.J. and HABBEMA, J.D.F., Duration of pre-clinical cervical cancer and reduction in incidence of invasive negative Pap smears. Int. J. Epidemiol., 24, 300–307 (1995).