A prevalence peak is expected in breast cancer incidence when mammography screening begins, but afterward the incidence still may be elevated compared with prescreening levels. It is important to determine whether this is due to overdiagnosis (ie, the detection of asymptomatic disease that would otherwise not have arisen clinically). In the current study, the authors examined breast cancer incidence after the introduction of mammography screening in Denmark.
Denmark has 2 regional screening programs targeting women ages 50 years to 69 years. The programs were initiated in 1991 and 1993, respectively. No screening takes place in the 13 other Danish regions. Data regarding incident breast cancers detected between 1979 and 2001 were retrieved from the Danish Cancer Registry for each screening region and for the rest of Denmark, and time trends in rates for women ages 50 years to 69 years were compared. From 1 program, individual screening data were used to analyze breast cancer incidence in women who were never screened, those who were screened for the first time, or those who previously were screened.
The incidence of breast cancer was found to have increased regardless of screening. In the screening regions, a marked prevalence peak was observed, and the incidence hereafter was compatible with the level indicated by the 95% confidence limits for the regression curves for the rates in the prescreening period, taking into account the artificial ageing in the program, the influx of newcomers, and variations in the data. Women who had undergone previous screening were found to have the same incidence of breast cancer as women who were never screened.
Detecting disease before it becomes symptomatic most likely means diagnosing disease at an earlier stage. Breast cancer patients diagnosed at an early stage are known to have a better survival than other patients.1 Therefore, the purpose of breast cancer screening programs is to bring the diagnosis forward in time, and thereby reduce mortality from the disease. However, the introduction of a screening program also affects the incidence of the disease.
When a screening program is first introduced, a pronounced increase in incidence rates is noted. This is the result of the simultaneous detection of both prevalent and incident cases. Depending on the lead time (the time from when a disease is screen detected until it would otherwise become clinically detected), screening detects those cases that would otherwise have shown up within the near/or not-so-near future. Therefore, the significant increase in incidence rates noted after the introduction of a screening program is well understood. However, what happens afterward? What should be expected? If screening stops, a decrease is expected in the incidence rate as a compensation for the prevalence peak, and the incidence rates will thereafter return to the level without screening. However, screening programs are rarely stopped. Usually, once a program has been introduced, screening will continue to be offered at regular intervals to women in the target age groups. In this situation, a drop in the incidence rates is not expected in the target age groups after the prevalence peak. Because each screening round brings forward in time the diagnosis of the pool of new incident cases, we would expect the incidence rates to return to the prescreening level. However, inherent in advancing the time of diagnosis is the risk of overdiagnosis. By overdiagnosis, we mean that as a result of screening, a person is diagnosed with a disease that, without screening, would in all likelihood never have emerged as a clinical disease during the person's lifespan. In other words, the disease would never have become symptomatic or the person would have died of another cause before the disease became symptomatic.
Overdiagnosis is difficult to estimate because there are several factors influencing incidence rates when screening is introduced and even after the prevalence peak is reached. First, there is a general tendency in the Western world for breast cancer incidence rates to increase with time. Second, lead time is responsible for the initial prevalence peak, and newcomers to the program will have their prevalence round when entering the program, thereby also pushing the incidence rates upward later in the program. Third, incidence rates of breast cancer are age dependent, meaning that because screening advances the time of diagnosis, those cases that are detected by screening are those that otherwise would have appeared clinically when the woman was slightly older (more accurately, when she was of an age equal to her own age plus the lead time). Finally, cases of carcinoma in situ (CIS) are also detected by screening. Therefore, the detection of cases of CIS in 1 screening round should lower the incidence rates of invasive breast cancer in subsequent rounds. In the current study, we investigated the risk of the overdiagnosis of breast cancer after the introduction of mammography screening in Copenhagen and Fyn County, Denmark.
MATERIALS AND METHODS
The first Danish mammography screening program was initiated in the municipality of Copenhagen on April 1, 1991.2 In November 1993, another program was introduced in the county of Fyn.3 A third small program began in the municipality of Frederiksberg in June 1994, but this program later was merged with the Copenhagen program and was not included in the current analyses. Both programs have screening rounds of approximately 2 years in length, but the target groups are slightly different. In Copenhagen, the target population is comprised entirely of women who are between ages 50 years and 69 years on the start date of the screening round, whereas the program in Fyn County invites women to a screening round if they are age 50 years but have not yet turned 70 years on the date of the invitation. For the estimation of the person years at risk we used the number of women in the age group in the middle of the round and multiplied this with the length of the round. We called this the full time. However, for the target group of the Copenhagen program, the age groups 50, 51, 70 and 71 were special. For the group of women aged 50 and 71, we therefore made the same calculation as for the other age groups but multiplied by 25%. For the group of women aged 51 and 70 we multiplied by 75%. For each program, we defined a comparison group consisting of all women in the appropriate age group living in Denmark (excluding the 3 screening regions of Copenhagen, Fyn County, and Frederiksberg), referred to as “the rest of Denmark.” Opportunistic screening is used to a very limited extent in Denmark,4 and therefore “the rest of Denmark” was considered to be a good proxy for nonscreening regions.
At birth, every person in Denmark is issued a unique personal identification number. We used these numbers to link information from the different registries. From the Danish Cancer Registry, we retrieved the number of cases of invasive breast cancer (International Classification of Diseases and Causes of Death, 7th edition [ICD-7] code 170 and behavior code 3) and from the Central Population Registry, we retrieved information regarding places and periods of residence for the women. Before the initiation of screening, we considered 2-year calendar periods. After screening began, we considered the actual calendar period of each screening round for Copenhagen, which was approximately 2 years. For Fyn County, we considered 2-year calendar periods because the screening rounds overlapped (ie, some women invited in the second round were screened at the same time as some women invited in the third round, etc.). In a given period, we included cases of invasive breast cancer among women in the relevant age groups living in the region at the date of diagnosis. We computed age-specific incidence rates by 1-year age groups (in which age is the age at the date of diagnosis) using population data from Statistics Denmark to estimate person-years at risk. Finally, we computed age-standardized incidence rates, standardized to the Danish female population in 1990 in the appropriate age group. Moreover, using regression analysis, we estimated the time trend in the age-standardized rates in both screening regions before the initiation of screening and in their control groups.
From the screening program in Copenhagen, we obtained information regarding each screened woman, and by linking this information with the Central Population Registry we computed actual time at risk for each woman by age group and invitation round. We divided both cases and risk times according to the screening status of the woman (ie. whether she had never been screened [no screen], was being screened for the first time in the round [first screen], or had been screened in an earlier round [after first screen]). For first screen, women contributed person-years at risk from the round in which they had their first screen. For after first screen, women contributed person-years at risk from all rounds after the round in which they underwent their first screen. For no screen, women contributed all their person-years at risk to never screened. For women diagnosed as a result of screening, we used the date of mammography screening as the date of diagnosis. Finally, to obtain an estimate of the effect of the artificial aging of women by the lead time, we computed the number of expected breast cancer cases for women ages 52 years to 69 years who were previously screened based on the age-specific incidence rates in the last 2-year period before screening was introduced, and compared this with the number of expected cases for the same women if they had all been 2 years older. SAS software (version 9.2; SAS Institute Inc., Cary, NC) was used for the analysis.
Only invasive breast cancers are systematically recorded in the Danish Cancer Registry. Therefore, data regarding CIS cases were available only from the screening participants in Copenhagen.
During the study period between 1979 and 2001, a total of 3199 cases of invasive breast cancer occurred in Copenhagen, compared with 22,539 in the rest of Denmark; in Fyn County, a total of 3162 cases of invasive breast cancer occurred, compared with 23,160 in the rest of Denmark (Table 2).
Table 1. Time Periods for Mammography Screening Programs in Copenhagen and Fyn County
Dates of the Round
Date Invited Women were Born
Round 7 (1) was initiated on 4/4/1991, but we included the first 3 days of April so that it matches the previous round. Some invitation rounds were longer than 2 years, so 2 women in Round 8 (2) were diagnosed with breast cancer at age 72 years. These women were not included in the analysis.
For Fyn County: The first round took place between 1/1/1980 and 12/31/1981, both dates included. The next round occurred between 1/1/1982 and 12/31/1983, both dates included, and onward until the last round, which took place between 1/1/2000 and 12/31/2001.
Table 2. Cases of Invasive Breast Cancer and Person-Years at Risk for Copenhagen, Fyn County, and Their Control Regions from the Rest of Denmark
Time period refers to the approximately 2-year calendar period studies, see Table 1 for details.
Copenhagen invitation rounds are in parentheses.
Fyn County invitation rounds are in squared parentheses.
The Danish Cancer Registry only registers cases by month, but we chose to use the first of the month as the date of diagnosis, and contribute the case to where the woman lived on that date. If she moved on the date of diagnosis, the case was attributed to the place from which she moved.
Incidence rates increased with time. When screening was introduced, a remarkable prevalence peak was observed, followed by a drop to a more moderate level. The quadratic regression curves for the rest of Denmark demonstrated a stable increase in incidence over time with a slight upward curvature, and the regression curves fit well with the data (Fig. 1). With the highly varied data from Copenhagen before the initiation of screening, the quadratic regression analysis provided a somewhat unrealistic curve with very broad 95% confidence limits. Therefore, we have added the linear regression line with more narrow confidence limits. For Fyn County prior to the initiation of screening, a quadratic regression analysis was performed. The prevalence peak for Copenhagen was found to be more pronounced than the prevalence peak for Fyn County. One reason for this was that the screening rounds in Fyn County overlapped; therefore, some of the women were screened for the first time in Round 2, thereby defusing the prevalence peak.
Throughout the first 5 screening rounds in Copenhagen, there were a total of 1507 cases of invasive breast cancer. Of these, 788 cases were detected by mammography screening and 381 cases were detected later in time in screened women. Table 3 shows the distribution of cases and person-years at risk by rounds and screening status of the women (ie, whether the woman had never been screened, was screened for the first time, or had previously been screened). The distributions of person-years at risk were unique for Round 1 but similar for Rounds 2 through 5; however, the distributions varied more for the cases (Fig. 2). From Round 2 onward, approximately 65% of the total time was accumulated by women after the first screen. After Round 1, women screened for the first time were found to have a clear overrepresentation in the younger age groups because such women were offered screening for the first time. The older age groups were represented only in the first screen group if they declined a previous invitation or had just moved into Copenhagen. The estimated person-years at risk in Table 2 were found to be in good accordance with the computed person-years at risk shown in Table 3. Women screened for the first time in general had a high incidence rate due to the detection of prevalent cases. The breast cancer incidence rate of women previously screened was found to be close to the incidence rate of the women who were never screened. Women who were never screened were found to have a surprisingly high incidence rate in Round 1.
Table 3. Cases of Invasive Breast Cancer and Person-Years at Risk for Copenhagen by Invitation Round and Screening Status
After First Screen
PY: person-years at risk.
The artificial aging was estimated to result in a mean increase of 4.6% in the number of expected cases (Table 4). After Round 1, approximately 65% of the target population were previously screened women and, on this basis, the expected number of cases in the total population would be increased by 3%. It should be noted that some of the previously screened women might not have been screened for a significant period; therefore they might not appear artificially aged, which makes the calculated increase too high. Conversely, we have assumed a lead time of 2 years, which might be too short, thereby contributing to an underestimation of the increase. Assuming the same increase for women first screened would result in a total increase of 3.5% in the number of expected breast cancer cases.
Table 4. Expected Cases of Invasive Breast Cancer among Women Previously Screened Using Actual Age and Actual Age +2 Years, Respectively, in Copenhagen, Denmark*
Age-specific incidence rates for Copenhagen for the last prescreening round were used for the calculation of expected cases. Expected cases for actual age were computed for women previously screened and who were ages 52 years to 69 years. Expected cases for actual age +2 years were computed for the same population, now ages 54 years to 71 years.
The difference in % was the difference in the percentage of the expected cases for actual age.
During the 5 screening rounds in Copenhagen, a total of 119 cases of CIS were detected. After the diagnosis of CIS, 4 women were diagnosed with invasive breast cancer in the same round and 4 women were diagnosed with invasive cancer later in time.
The incidence of invasive breast cancer increased in Denmark from the beginning of the 1980s to the beginning of the 2000s, irrespective of screening. A slight curvature trend was observed for women ages 50 years to 71 years in the nonscreening regions. Similar trends were noted in the 2 screening regions, Copenhagen and Fyn County, before the initiation of the screening programs, and both regions had slightly higher prescreening rates than the nonscreening regions. However, the 95% confidence limits for the curves were very broad, indicating that although we had a well-defined population and a very reliable cancer registration, there still was much variation in the data.
The observed pattern implies that in the hypothetical case in which mammography screening had not been introduced, we would have expected to observe incidence rates for Copenhagen and Fyn County continuing to be slightly higher than those for the rest of Denmark, and these rates should have followed the generally increasing time trend demonstrated by the regression curves for Copenhagen/Fyn County or for the rest of Denmark, or a combination of these, although still with much variation. However, screening was introduced, and for both programs we noted a pronounced prevalence peak in the first screening round. The prevalence peak in Copenhagen was approximately twice the incidence of breast cancer in the last period before screening was initiated. This increase is affected by both the lead time, the participation rate, and the sensitivity. The overlapping screening rounds diluted the prevalence peak in Fyn County. However, the 2 programs also differed in other ways. The participation rate in the first round was 71% in Copenhagen and 84% in Fyn County, whereas the sensitivity estimated from the interval cancers was higher in Copenhagen than in Fyn County (87% and 82%, respectively).2, 3 Both a high participation rate and a high sensitivity will contribute to a high prevalence peak; therefore, the 2 programs were favored in different ways. Given the known participation rate and sensitivity, a lead time of approximately 3 years would be compatible with no overdiagnosis occurring during the first screening round.
Because screening continued to be offered every second year to women ages 50 years to 69 years in both Copenhagen and Fyn County, we would not expect the prevalence peak to be compensated by a subsequent drop in the incidence to less than the prescreening level. As expected with continued screening, the incidence was, with the variation in the data, compatible with the level expected based on the confidence limits for the rates in the prescreening period. However, the observed rates were in the upper end of this range, which could be due to overdiagnosis. In addition, for Copenhagen, it should be taken into account that there was a slight (approximately 3.5%) increase in the incidence due to the artificial ageing. This means that, for example, the age-standardized rate of 304 per 100,000 noted during the second screening round in Copenhagen would have been approximately 294 without the ageing factor. There is also even in subsequent rounds a steady supply of newcomers having prevalent screens, but in Copenhagen they are concentrated within the younger age groups (those mainly ages 50–53 yrs) (Fig. 2). Therefore, the contribution of newcomers in subsequent rounds to the total age-standardized incidence rate is very small.
The incidence rate of never-screened women in Copenhagen was very high in the first screening round, despite the finding that opportunistic screening was very limited.4 The never-screened women could be a selected group; however, studies from the Scandinavian countries have not indicated that nonparticipants in mammography screening programs should be particularly burdened by known breast cancer risk factors.5, 6 However, it is possible that receiving their first invitation to screening prompted some women with symptoms of breast cancer to take action and contact their general practitioner. A similar mechanism is not expected to play a role in subsequent screening rounds in which the majority of the never-screened women have previously received a screening invitation.
The incidence rate among women screened for the first time was, as expected, high during the first round due to the screen-detected prevalent cases. However, the age-standardized incidence rate in this small group of women was found to be lower in the fourth and fifth rounds, in which nearly all cases occurred among the young newcomers to the program (Fig. 3).
The incidence rate of women previously screened was the same as the incidence rate of never-screened women. The finding that screened women, once the pool of prevalent cases had been emptied, behaved in the same way as the never-screened women indicated that overdiagnosis did not play a role.
CIS cases constitute only 12% of the detected cases in the Copenhagen screening program, reflecting a deliberately conservative attitude toward supposedly benign microcalcifications,7 and therefore we did not have enough data to investigate the effect of CIS detection on the incidence rate of invasive breast cancer in subsequent rounds.
In randomized controlled trials of breast cancer screening in which the control arm had been screened at the end of the study period, no difference was noted with regard to the cumulative incidence rates between the intervention and control arms.8 The incidence of both invasive and in situ breast cancer has increased in the U.S. since the beginning of the1980s, in parallel with an increase in mammography use.9 This could well indicate overdiagnosis, but it is not possible with the gradual introduction of screening to separate the prevalence peak from the underlying incidence trend. However, it is clear that the majority of in situ cancers would have remained undetected without screening. For the Dutch screening program, Boer et al. estimated overdiagnosis to amount to less than 2% of the total incidence.10 A similar conclusion was reached by Paci et al. for the screening program in Florence, Italy, by modeling the expected cases that would have arisen in the absence of screening, taking lead time into account.11 Including prevalent cases and CIS cases, Anttila et al. found an increase in breast cancer cases of 18% during the first 10 years after the introduction of screening in Helsinki, Finland.12 From studying the incidence rates of invasive breast cancer in Norway and Sweden before and after the initiation of screening, Zahl et al. concluded that approximately one-third of all cases were overdiagnosed.13 However, their article had some methodological errors, as pointed out in the correspondence regarding their report.14 A detailed analysis of the effect of the introduction of mammography screening on incidence rates of invasive breast cancer was made by Moller et al.15 They used age period-cohort modeling of the national population from the 5 Nordic countries. They identified 3 distinct effects of screening: 1) a large increase in incidence rates in the first screening round, 2) a persistently higher incidence rate for the artificially aged women who remained in the program, and 3) a marked decrease in the rate when the women left the program. The Danish screening programs began too late and covered too small a proportion of the national population for this method to provide reliable results.
The strength of the current study was that the screening regions covered only approximately 20% of Danish women in the targeted age group of 50 years to 69 years. This made Denmark a good natural experiment because we could compare the incidence rates after screening was introduced with both the incidence rates before the introduction of screening and with the incidence rates in the nonscreening regions. Therefore, we had a well-defined population, and we had comprehensive information regarding population movements from the Central Population Registry and on incident breast cancer cases from the Danish Cancer Registry. Another advantage was that we had detailed information concerning women's screening participation over time from the Copenhagen screening program, allowing us to analyze incidence rates by the screening status of each individual woman. The person-years at risk accumulated in a given period of approximately 2 years varied from 153,000 to 78,000 in Copenhagen, in which the size of the population decreased over time, and it remained fairly stable at approximately 100,000 in Fyn County. However, even with populations of this size, the data demonstrated much variation. This was observed particularly when the Copenhagen data were divided by screening status. It also should be noted that the analysis was based on only 5 screening rounds from Copenhagen and 4 screening rounds from Fyn County, and that only cases of invasive breast cancer were included in the analysis.
The data from the current study demonstrated a prevalence peak during the first invitation round that was approximately twice the incidence rate before screening was initiated. Whether this represents overdiagnosis depends on the participation rate, the sensitivity, and the unknown lead time. From the second invitation round onward, the results of the current study demonstrated that the incidence of breast cancer was compatible with the incidence that could have been expected in the absence of screening, taking into account the artificial ageing of the women, the influx of newcomers, and the variation within the data. Furthermore, we found that women who were previously screened had an incidence rate that was in keeping with that of never-screened women. Therefore, the data do not provide evidence of the overdiagnosis of invasive breast cancer in Rounds 2-5 of the 2 Danish screening programs, or if overdiagnosis occurs, it is only of limited magnitude.