The distribution of ductal carcinoma in situ (DCIS) grade in 4232 women and its impact on overdiagnosis in breast cancer screening

We obtained 17,744 excerpts from 12,301 women with DCIS from the years 2007, 2008 and 2009 from the ‘Nationwide network and registry of histopathology and cytopathology in the Netherlands’ (PALGA). PALGA is a national database containing the excerpts and coded diagnoses of all pathological and cytological examinations performed in the Netherlands [20]. The mass screening status of these women was established by linking the database to the databases of the screening organisations by an independent third party, with the permission of the screening organisations. Our database contained anonymised records of mass screening status (positive, negative, year of last mass screening and number of mass screening examinations), age, year of diagnosis, and a short summary of the conclusion of the original pathology report.

From the 12,301 women, we excluded those who also had a concurrent invasive breast cancer (ipsilateral or contralateral, N = 7089), those who had a lobular carcinoma in situ and no DCIS (N = 6), those who turned out after excision biopsy or ablation not to have any malignancy (N = 131), those who had recently been treated for invasive breast cancer (N = 247), those who had no grade mentioned in the excerpt (N = 17), those who had an inconclusive excerpt on invasion or otherwise (N = 242), and women who had a prevalent DCIS, rather than a new diagnosis in the study period (N = 354). We excluded contralateral disease because our model does not include bilateral disease.

DCIS were assumed to be ‘detected by mass screening’ when a woman had had a positive screening examination between 2007 and 2009. Women who had participated in the screening programme, and did not have any positive screens, but who did have a DCIS diagnosis in 2007, 2008, or 2009 were assumed to have an interval DCIS. The number of interval DCIS increased across the study period due to the cumulative effect of interval DCIS diagnosed in women screened in the previous year (2007) or in the 2 previous years (2007 and 2008). Interval DCIS were rare in 2007 because of the low frequency of interval carcinomas within the same calendar year in which the screening examination took place. In 2008, interval DCIS were diagnosed in women screened in 2007 or 2008, and in 2009, interval DCIS were diagnosed in women screened in 2007, 2008 or 2009.

Women who were not known to the screening organisations may have been under clinical surveillance because of high familial risk, frequent (benign) breast anomalies, or because of personal preference. Diagnoses in this group may be the result of screening, but are not the result of the mass screening programme. Therefore we cannot conclude that DCIS not detected by mass screening, were not detected by screening. To compare the distribution of DCIS detected by mass screening to DCIS not detected by mass screening, we, therefore, chose to compare the DCIS detected by mass screening to the interval DCIS.

In line with the Dutch guidelines, the classification by Holland et al. is almost exclusively used [21]. At the start of the mass screening programme in the early 1990s, pathologists were instructed on how to uniformly classify each DCIS.

DCIS grade was determined using the information in the short summary of the pathology report by description, i.e. high, moderate, or low differentiation; low, intermediate, or high malignancy potential; or grade I, II, or III. If the summary contained more than one grade, this case was graded according to the highest grade mentioned. If there was a discrepancy between grades in different specimens of the same patient, the grade was based on the most representative specimen, i.e. resection is more representative than biopsy, but biopsy is more representative than cytology.

Proportions of DCIS grades were calculated by year, age group, and screening status. We compared these proportions between screening groups using the Pearson chi-square test. Multivariate analyses on age groups were performed with a logistic regression model. The statistically significant parameters were identified by the introduction of variables in a stepwise manner. All calculations were performed using IBM SPSS version 20.0 (IBM Corp., Armonk, NY, USA).

The MISCAN model is a microsimulation model that simulates the individual life histories of women [22]. The probability of each woman to have an onset of breast cancer is determined by calibrating the model to the incidence rate in 1989 (the year before screening was introduced), adjusted with an annual percentage change of 1.4 % to account for the rising background breast cancer incidence [23]. The natural history of breast cancer is modelled as a Markov-like progression through the successive preclinical stages of the disease. Details of the model have been described previously [4]. For this analysis we added the three DCIS grades to the model, using the age-dependent grade distribution found in this study (Fig. 1).

Following onset, breast cancer in a preclinical stage can progress to the next preclinical stage (dependent on the duration of the previous state), or become clinically detected. In addition, the DCIS stages may also regress to normal [24, 25]. Screening is superimposed on this life history.

The transition probabilities, duration of tumour stages, and test sensitivities were calibrated using data from the Dutch population and Dutch breast cancer screening from 1975 to 2010 on breast cancer incidence by stage, age, and detection mode. The Dutch nationwide breast-cancer screening programme has invited all women aged 50–69 since 1990 and women aged 50–75 since 1998 biennially for a mammographic screening examination, free of charge. The attendance rate is approximately 80 % [26].

We chose to look at model outcomes for the years 2000–2009 because there was a steady state situation in these years, more than 10 years after the start of the screening programme. We evaluated the following output: incidence rate by detection mode (screen detected or clinically detected), age, and year of diagnosis. The model compares women in the situation with screening, to the same women in the situation without screening; if a woman has a screen-detected cancer, but would not have had a diagnosis in the situation without screening, this case is regarded as overdiagnosed (Fig. 2).

The estimates and definitions of overdiagnosis vary widely among international publications [4]. To minimise confusion, we used the definitions of overdiagnosis which were deemed most useful by an independent review panel in the UK; from a population perspective: the proportion of all cancers ever diagnosed in women of the screening age and over (50–100 years) that are overdiagnosed; and from an individual perspective: the proportion of all cancers ever diagnosed in women of the screening age (50–75 years) that are overdiagnosed [27].

In the original model a 2 % regression rate, an 11 % progression rate, and a 5 % clinical detection rate was assumed for all DCIS, resulting in a proper fit of incidence [28]. Little is known about the natural history of DCIS without treatment. Small studies were published, indicating a progression rate of one in two to one in three for low-grade DCIS, one in three for intermediate-grade DCIS and two in three in high-grade DCIS [29, 30]. Progression rate may differ from the rate assumed in the original model. In the new model we assumed that intermediate-grade DCIS has the same transition probabilities as all DCIS had in the original model. We lowered the regression rate to 1 % for high-grade DCIS, and increased the regression rate to 4 % for low-grade DCIS, based on the findings of Sanders et al. [30]. The probability for a DCIS to be clinically detected was assumed independent of grade. The probability of progression: 16 % for low-grade DCIS, 31 % for intermediate-grade DCIS, and 53 % for high-grade DCIS, was estimated by correcting the probabilities of low-grade DCIS and high-grade DCIS by the progression found in literature [29, 30]. Adjusting the progression rate and therefore the duration of the state, influences all successive states, because the progression of each successive state is dependent on the duration of the previous state. High-grade invasive breast cancer follows high-grade DCIS and low-grade invasive breast cancer follows low-grade DCIS. We calibrated DCIS incidence rate to observed data for the period 1990–2010.