Introduction

Ductal carcinoma in situ (DCIS) is a neoplastic proliferation of epithelial cells confined to the ductal system without tumor invasion through the basement membrane [1]. DCIS is described as a non-obligate precursor of invasive carcinoma [2,3,4]. The proportion of DCIS among diagnosed breast cancers went from less than 5% to about 20% after the dissemination of screening mammography [5]. Among the drawbacks of screening mammography is the plausible detection and treatment of some DCIS that would have never progress toward invasiveness [6, 7]. Why and how often DCIS progress to invasive disease remains to be clarified [5].

Given the potential role of DCIS as a precursor, it is expected that DCIS and IBC would share common risk factors [8,9,10,11]. Factors more strongly associated with DCIS compared to IBC can represent factors affecting disease initiation or factors associated with DCIS which do not progress to IBC [8, 11]. Factors affecting progression would only be associated with IBC risk [8, 11]. Understanding up to what point DCIS and IBC share the same risk factors could help better distinguish DCIS that will progress toward IBC from those that will not. This will also help to avoid overtreatment by tailoring treatment according to risk of progression [11].

To our knowledge, only one study [12] reported the difference in DCIS detection rates and IBC detection rates according to several characteristics among only screened women. This study, based on 39,542 women aged 30 years or older, concluded that risk factors for DCIS are similar to those for IBC. However, this study included few screen-detected DCIS (n = 102) diagnosed in 1995 or earlier and did not take into account potentially important confounding variables such as HRT or breast density.

The aim of our study is to examine, in our population-based screening program, a wide array of women’s characteristics in relation to DCIS and IBC detection rates. We also aimed to assess whether some characteristics have a different effect on the DCIS detection rate compared to IBC detection rate. Finally, we want to verify whether these associations are maintained with the change in technology from film to digital mammograms.

Methods

Study population

The Quebec Breast Cancer Screening Program (Programme Québécois de Dépistage du Cancer du Sein, PQDCS) is an organized population-based mammography screening program launched in 1998 that actively invites women 50–69 years of age to receive biennially 2-views screening mammography in accredited facilities. The study is based on screening mammograms performed in the PQDCS from January 1, 2002, to September 30, 2015. Among the 3,888,262 screening mammograms performed during the study period, 278,693 (7.2%) were excluded (Fig. 1). A total of 3,609,569 screening mammograms, performed in 1,105,824 women in 102 facilities and interpreted by 469 radiologists, were used for the analysis.

Fig. 1
figure 1

Study population from the Quebec Breast Cancer Screening Program

Full size image

Characteristics of women

Information on women’s characteristics was obtained from a self-administered questionnaire completed at each screening examination and captured in the PQDCS information system. Size and weight of women were self-reported, and the body mass index (BMI) was then calculated (weight in kilogram divided by the height in meters squared). Breast density in 4 categories (< 25%, 25–49%, 50–75%, > 75%) was assessed by the radiologist who read the screening mammogram.

Technology used

Since 2007, digital mammography (computed radiography (CR) and direct radiography (DR)) gradually replaced screen-film mammography. In 2014, 58% of PQDCS mammograms were performed in CR, 41% in DR and 0.7% in film. The information on mammography technology was obtained from the Laboratoire de Santé Publique du Québec, including the date of change in the mammography unit from film to CR or DR systems.

Ascertainment of breast cancer

A screen-detected breast cancer is a DCIS or invasive carcinoma diagnosed within 6 months following a positive mammogram. A screening mammogram was classified as positive (abnormal) if the patient was referred for assessment; otherwise, it was considered negative (normal). Diagnoses of breast cancer were identified by validated algorithms that linked PQDCS data with other provincial databases [13, 14]. A breast cancer case is defined as DCIS if the pathology report from the PQDCS information system is completed for an ‘in situ ductal carcinoma (intraductal noninvasive)’ or if the International Classification of Diseases (ICD) recorded in a provincial database (MedEcho) is DCIS (ICD-9: code 233.0,ICD-10: code D05.1, D05.7, D05.9). A breast cancer case is defined as IBC if the pathology report from the PQDCS information system is completed for an ‘invasive carcinoma’ or if the ICD in MedEcho is IBC (ICD-9: 174, ICD-10: C50). For bilateral breast cancers, only the one with the most aggressive histopathological features was considered. Otherwise, it was chosen randomly.

The DCIS detection rate was the number of screen-detected DCIS over the total number of screening mammograms. The IBC detection rate was the number of invasive screen-detected breast cancers over the total number of screening mammograms. The percentage of DCIS among screen-detected breast cancer was calculated as the number of screen-detected DCIS over the total number of screen-detected breast cancer whose type is known.

Statistical analysis

The analyses were conducted in a threefold manner. First we measured the association between women characteristics and the risk of screen-detected DCIS. Second, we assessed the association between women characteristics and the risk of screen-detected IBC. Third, we compared the odds of DCIS detection against the odds of IBC detection according to women characteristics.

All analyses were conducted using the screening mammograms as the unit of analysis. We used logistic regression and generalized estimating equations (GEE) with independent correlation matrix and sandwich estimator to account for correlation between mammograms. We considered radiologist-level and center-level correlations and estimated empirical variance matrix with the three-step method developed by Miglioretti and Heagerty [15]. Adjusted detection rate ratios were estimated using exponential of the model parameters. Wald 95% confidence intervals and p values were derived from the empirical variance estimator. We also estimated DCIS and IBC detection rates according to women’s characteristics from the adjusted GEE model using marginal standardization [16]. Thus, adjusted rates represent the average of the model predicted probabilities assuming all units would possess the characteristic of interest, but keeping other covariates as observed.

Models were adjusted for all women’s characteristics, year of screening mammogram (2002–2006, 2007–2011, 2012–2015) and technology used (film, CR, DR). A complementary analysis was carried out in order to assess whether restricting our analysis to digital mammograms changed results.

The GENMOD procedure of the SAS software (version 9.4, Copyright © 2016 by SAS Institute Inc., Cary, NC, USA) was used. Statistical significance was tested at 5% for all tests (2-sided).

Results

Characteristics of the women studied

Among the 3,609,569 screening mammograms in the study, 362,817 were positive (recall rate = 10.0%) and 19,384 breast cancers were screen-detected (4173 DCIS, 15,136 IBC and 75 with unknown type). Among the screen-detected breast cancer, 21.5% were DCIS. The detection rate was 1.2/1000 screens for DCIS and 4.2/1000 screens for IBC.

The distributions of the screening mammograms according to characteristics of the women and mammograms by breast cancer status are presented in Table 1. About half of screening mammograms were made on film (51%), while 35% were made on CR and 13% on DR. About 11% and 24% of screening mammograms were performed in, respectively, women who had a previous breast aspiration or biopsy and women who currently use HRT. The proportion of mammograms performed in women with an elevated BMI (≥ 30 kg/m2) is higher in women with IBC screen-detected (27%) compared to women with DCIS screen-detected (21%) or all screening mammograms (23%). Finally, the proportion of women with breast density > 50% is higher in women with DCIS screen-detected (49%) compared to women with IBC screen-detected (40%) and all screening mammograms (36%).

Table 1 Characteristics of women and mammograms for the screen-detected breast cancer (DCIS and invasive) and all screening mammograms
Full size table

Univariate analysis of DCIS and IBC detection according to women’s characteristics

The DCIS detection rates and the percentages of DCIS among screen-detected breast cancer according to women’s characteristics are presented in Fig. 2. The highest DCIS detection rate was observed in mammograms from women without previous breast aspiration or biopsy, with a DCIS detection rate of 1.8/1000 screens. The lowest DCIS detection rate, 0.6/1000 screens, was observed in mammograms from women with a breast density < 25%. Otherwise, the proportion of DCIS among screen-detected breast cancer was lower in mammograms done on older women. It went from 26% in mammograms done on women aged 50–54 to 19% in mammograms done on women aged 60–64 and 65–69 years. The highest proportions of DCIS among screen-detected breast cancers were observed in mammograms done on pre-menopausal women (27%), women with BMI < 20 kg/m2 (31%) and women with breast density > 75% (28%).

Fig. 2
figure 2

DCIS detection rate and proportion of DCIS among screen-detected breast cancer according to characteristics of women. DCIS ductal carcinoma in situ, HRT hormone replacement therapy, BMI body mass index, Ini. initial, mam. mammogram

Full size image

Multivariate analysis of DCIS and IBC detection according to women’s characteristics

Relationship of characteristics of women and mammograms with DCIS and IBC detection rate is presented in Table 2. Compared to mammograms from women without breast clinical examination in the last year, mammograms from women with clinical breast examination in the last year had a similar DCIS detection rate (adjusted cancer detection rate (CDR) ratio = 1.04, 95% CI 0.96–1.12), but had a lower IBC detection rate (adjusted CDR ratio = 0.96, 95% CI 0.93–0.99).

Table 2 Multivariate association of women characteristics with DCIS and IBC detection rate, and with the odds ratio of DCIS among screen-detected breast cancers
Full size table

Increasing women’s age, current HRT use and higher BMI are associated with higher IBC detection rates than DCIS detection rates (Table 2). For example, women with a BMI ≥ 35 kg/m2 showed DCIS and IBC detection rates of, respectively, 1.2 times (adjusted CDRratio = 1.21, 95% CI 1.07–1.36) and 1.7 times (adjusted CDR ratio = 1.73, 95% CI 1.63–1.84) higher than that of mammograms from women with a BMI between 20.0 and 24.9 kg/m2. This translated into an adjusted odds of DCIS on IBC among screen-detected cancers lower for mammograms from women with a BMI ≥ 35 kg/m2 compared with mammograms from women with a BMI between 20.0 and 24.9 kg/m2 [adjusted odds ratio (OR) 0.70, 95% CI 0.60–0.81].

In contrast, having a previous breast aspiration or biopsy and increasing breast density were more strongly associated with DCIS detection rates than with IBC detection rates. Mammograms from women with a previous aspiration or biopsy had a 49% higher DCIS detection rate (adjusted CDR = 1.49, 95% CI 1.38–1.61) and a 29% higher IBC detection rate (adjusted CDR = 1.29, 95% CI 1.22–1.35) compared to mammograms from women without this antecedent. This translated into an OR of DCIS among screen-detected of 1.15 (95% CI 1.04–1.27). Also, mammograms from women with breast density > 75% had a 2.8 times higher DCIS detection rate (adjusted CDR = 2.79, 95% CI 2.43–3.20) and a 1.8 times higher IBC detection rate (adjusted CDR = 1.83, 95% CI 1.67–2.01) compared to mammograms from women with breast density < 25%. Then, the odds of DCIS on IBC among screen-detected cancers are higher in mammograms from women with breast density > 75% compared to mammograms from women with breast density < 25% (adjusted OR = 1.53, 95% CI 1.32–1.77).

Similar patterns of association for detection of DCIS and IBC were observed according to screening history, women family history of breast cancer, age of the first birth and menopausal status (Table 2).

We observed no association between technologies used for mammogram and either DCIS or IBC detection rates (Table 2).

Multivariate analysis of DCIS and IBC detection according to women’s characteristics for digital mammograms only

Associations between DCIS and IBC detection rates according to women’s characteristics for digital mammograms are presented in Table 3. Compared to the whole cohort (main analysis), the same pattern of associations was observed between DCIS or IBC detection rates for age, screening history, breast clinical examination, family history, age at first birth, BMI and breast density when only digital mammograms were considered. The odds of DCIS among screen detected cancers were no longer statistically significant according to previous breast aspiration of biopsy (p value = 0.3562) and HRT use (p value = 0.2253) with the digital mammograms only (these p values were, respectively, 0.0050 and 0.0244 in the whole cohort).

Table 3 Multivariate association of women characteristics with DCIS and IBC detection rate and with the odds ratio of DCIS among screen-detected breast cancers, for digital screening mammograms only
Full size table

Discussion

Among the screening mammograms of the PQDCS between 2002 and 2015, the DCIS detection rate was 1.2/1000 screens, which is comparable with other breast cancer screening programs [17,18,19,20]. In our breast cancer screening program, clinical breast examination in the last year was not associated with the DCIS detection rate, but was associated with a decrease in the IBC detection rate. The age of the women at screening, the use of HRT and the BMI were less associated with the DCIS detection rate than with IBC detection rate. On the opposite, previous breast aspiration or biopsy and breast density were more strongly associated with the DCIS detection rate rather than the IBC detection rate.

Kerlikowske et al. [12] found, in their cross-sectional study of screened women, that the magnitude of the associations were similar between their studied risk factors and DCIS or IBC screen-detected, except for increasing age and the presence of a palpable mass. These factors were more strongly associated with IBC rather than DCIS screen-detected. In our analysis, we also observed the same results for age, but we also found other characteristics for which the strength of the associations varies with DCIS or IBC detection rate.

Other studies have examined the association between risk factors and DCIS or IBC diagnoses [9, 10, 21,22,23,24,25,26,27,28,29]. Results of these studies are divergent. Some studies concluded that, in general, risk factors for DCIS are similar to those for IBC [10, 12, 21,22,23,24,25,26,27,28,, 24–27, 29, 30]. However, some studies have observed differences in associations for DCIS and IBC, particularly for HRT use, BMI and breast density [10, 12, 21,22,23,24,25,26,27,28,24, 26–28, 31,32,, 32]. For example, Ko et al. [21,22,23,24,25,26,27,28,] and Reeves et al. [10] have found similar associations between the use of HRT and DCIS or invasive ductal cancer, whereas Trentham-Dietz et al. [26] observed that HRT use was more strongly associated with IBC compared to in situ.

Such inconsistencies in findings may be explained, at least in part, by variation in the population studied and the women characteristics considered. Some studies are based on a selected sample of women and the women’s screening history or whether or not the breast cancer was likely to have been screen-detected are not take into account [10, 23, 29]. When interpreting the findings of etiological breast cancer studies, the mix of screen-detected and women/clinician-detected cases in the study population will influence the results [10, 33]. Furthermore, each study does not consider the same confounding factors, such as breast density [8,9,10, 12, 25–27, 29]. Breast density is an important variable since it was recognized as a risk factor for breast cancer [34, 35,36,], a factor that also influences the screening sensitivity of mammography [35,36,37] and is correlated with the BMI [35,36,] and the use of HRT [35,36,, 37, 38]. Only three studies [21,22,23] have considered together breast density, HRT use and BMI in their analysis.

In our data, we observed that being older, HRT use and higher BMI were more strongly associated with IBC detection rate than DCIS detection rate. These results are consistent with the hypothesis that these characteristics may have an effect on cancer progression from the in situ phase to the infiltrating phase [8, 9, 11]. Some authors have already discussed this possibility, especially concerning HRT use [39–41]. Gapstur et al. [28] show that there was no association between ever HRT use and the incidence of DCIS, while exposure to HRT was associated with an increased risk of IBC with a favorable histology. The study by Marshall et al. [31] also observed that the discontinuation of HRT reduced the incidence of IBC, but not the incidence of DCIS.

Like others studies [22, 26, 27], our data also suggest that having a previous breast aspiration or biopsy and higher breast density were more strongly associated with DCIS detection rate than IBC detection rate. These characteristics may play a greater role in cancer initiation rather than progression of cancer and could play a role on screening sensitivity, or both. For example, the decrease in screening sensitivity according to breast density may be less important for DCIS compared to IBC. A large proportion of DCIS is screen-detected due to the presence of microcalcifications [42, 43], and they could remain more visible on mammograms even in the presence of higher breast density [22]. Moreover, the biological properties of the breast tissue components associated with breast density may increase the probability of the transition of normal epithelium to malignant cells [44]. Hence, breast density can create an environment that promotes the initiation of breast cancer. Thus, the DCIS pool would be higher in women with previous breast aspiration or biopsy and in women with higher breast density. Given these larger reservoir of DCIS, the risk of overdiagnosis in these women can be higher. Further studies will be needed to determine the individually effect of these characteristics on the initiation of the disease and the screening mammography sensitivity.

This study had some limitations. DCIS or IBC is determined according to provincial databases and not by a revision of the pathology reports; thus, some cases may have been misclassified. Moreover, we studied all invasive breast cancer and we cannot restrict our analysis on invasive ductal carcinoma. However, the invasive ductal carcinoma is the commonest type of invasive breast cancer [28, 45]. We could not adjust for some behavioral women’s characteristics such as alcohol consumption or smoking habit. Also, we do not have detailed information about such as specific regimens of HRT used as well as duration of the exposition to HRT. Moreover, this study included film and digital screening mammograms. We have checked the robustness of our results in a complementary analysis restricted to digital mammograms. Although the analysis had lower statistical power (they were based on about half of the DCIS and IBC), we found the same associations. These results reassure us that the findings of this study are still valid even in the era of digital mammography.

Our study also had several strengths. We used a large population-based cohort of women participating in an organized screening program, avoiding potential selection bias due to differential participation. Compared to previous studies, we have, to our knowledge, the largest number of screen-detected DCIS. We also have a wide array of women characteristics, including HRT use, BMI and breast density, reducing concerns about residual confounding.

In conclusion, our study shows that women’s age, HRT use and BMI appear to be more strongly associated with IBC than DCIS. These results suggest that these characteristics seem to play a role in the progression of breast cancer from in situ to invasive stage. On the other hand, having a previous breast aspiration or biopsy and breast density seems to be more strongly associated with DCIS rather than IBC detection by mammography. These findings suggest that these characteristics could be playing a role in the initiation of the breast cancer. However, we must not forget that cases studied are all screen-detected cancers by mammography. All these characteristics can also have an effect on the screening sensitivity. This effect on sensitivity may be different depending on whether the screen-detected cases were DCIS or IBS. Although these findings do not provide direct evidence regarding the mechanisms underlying the development of DCIS and IBC, they deepen our understanding of the characteristics that affect DCIS and IBC detection.