Clinicopathological Risk Factors for an Invasive Breast Cancer Recurrence after Ductal Carcinoma In Situ—A Nested Case–Control Study

Ductal carcinoma in situ (DCIS) is a potential precursor of invasive breast cancer (IBC). It is characterized by proliferation of ductal epithelial cells confined within the ductal–lobular system. Most women (80%–85%) are diagnosed with DCIS by screening mammography in which breast abnormalities are found without the women having clinical symptoms (1). In the western world, the incidence of DCIS has increased almost 6-fold with the introduction of population-based breast cancer screening and accounts for about 20% to 30% of all newly diagnosed breast neoplasms (2–7).

Although DCIS is not life threatening, it does increase a woman's risk of developing IBC later in life, which subsequently could lead to a breast cancer–specific death (8). However, we are currently unable to distinguish DCIS lesions that will progress to IBC from those that will not, because there is only limited information on the long-term natural history of DCIS (9). As a consequence, almost all DCIS is treated by mastectomy or breast-conserving surgery (BCS) with or without radiotherapy. This is done under the assumption that this will prevent IBCs and subsequently breast cancer–specific deaths, despite the fact that breast cancer–specific mortality after DCIS is uncommon: <2% (10, 11). On top of that, the long-term benefit of treatment of asymptomatic DCIS that may or may not progress to IBC is difficult to quantify (12). As a result, screening programs are nowadays criticized for being associated with overdiagnosis and overtreatment (13, 14).

Distinguishing, at diagnosis, DCIS that might cause life-threatening disease from indolent DCIS is therefore of great importance. A multitude of studies have tried to find markers that could predict local recurrence or progression of DCIS (15). In a few studies, investigators showed that various histopathologic characteristics of DCIS, such as lesion size, marginal status, histologic grade, architectural patterns, and presence of necrosis were associated with recurrence (16, 17). However, these studies did not discriminate between invasive and in situ recurrences. Furthermore, due to limited patient numbers and lack of validation studies, none of the markers studied to date show sufficiently strong evidence for an association with subsequent ipsilateral IBC (iIBC).

The primary objective of this study was to identify clinical and histologic characteristics of the initial DCIS lesions that are associated with the development of subsequent iIBC. Here, we report the results of our case–control study nested within a large nation-wide population-based cohort of Dutch women with DCIS treated by BCS alone between 1989 and 2005.

The study population has been described previously (18). In brief, we used a nation-wide population-based patient cohort derived from the Netherlands Cancer Registry (NCR), in which we included all women diagnosed with primary DCIS and treated with BCS alone within the Netherlands from January 1, 1989, to December 31, 2004. Patients did not receive tamoxifen or other antihormonal adjuvant treatment. According to the Dutch guidelines, patients diagnosed and treated for DCIS were followed by undergoing annual mammograms for at least 5 years. If no recurrence occurred, women could participate in population-based screening again, if applicable regarding age group. Patients with adjacent invasive disease or a prior cancer diagnosis except for nonmelanoma skin cancer were not included. This resulted in 2,658 eligible female participants.

Data provided by the NCR included information on age at and date of diagnosis, histology and treatment for DCIS, and any subsequent IBCs. Follow-up for subsequent iIBC and vital status were complete until at least January 1, 2011. The median follow-up was 12.0 years (interquartile range, 9.0–15.3). A total of 374 of the 2,658 women developed subsequent iIBC, as first invasive cancer, after a primary diagnosis of DCIS (18). At the start of this study, the first 316 women with a subsequent iIBC were identified and the remaining 58 cases were identified when data collection of this study was completed. These 316 women were included in this study and were considered “cases.” Controls were matched to cases based on age in years at DCIS diagnosis (exact) using a variable matching ratio. Controls had to have remained free from ipsilateral and contralateral IBC for at least as long as the initial DCIS diagnosis to iIBC development of the case they were matched to. Controls were selected with replacement, so some individuals were a control for more than one case.

Cases and controls originated from 58 hospitals within the Netherlands. We could not obtain formalin-fixed paraffin-embedded (FFPE) tissue blocks from some hospitals, either because tissue blocks were unavailable or because the hospital refused to provide tissue for research (61 cases and 388 controls; Supplementary Table S1). Furthermore, some patients were excluded during pathology review because no DCIS component was found, a (micro)invasive component or LCIS was present, or because the specimen was not assessable (53 cases and 173 controls). Finally, some patients were excluded because no matched case or control was available for case–control sets they belonged to (2 cases and 268 controls).

Together with the tissue blocks, pathology reports were retrieved from the participating hospitals. Pathology reports were reviewed for measurements of lesion size and margin status. Notable, data on lesion size were often not routinely described for DCIS in these old retrospective series.

We categorized year of DCIS diagnosis into two time periods: 1989 to 1998 (screening implementation phase) and 1999 to 2004 (full nation-wide coverage phase). Clinical presentation of DCIS was subdivided into screen-detected (mean age, 59 years; range, 50–74) and nonscreening-related (mean age, 54 years; range, 30–89). This information was available for 91% of the included patient group.

The study was approved by the review boards of the NCR (request K12.281; January 3, 2013) and PALGA (LZV990; April 16, 2013). The secondary use of tissue and data under an opt-out regime in this study conform Dutch regulations and the Code of Conduct of Federa-COREON (19).

New hematoxylin and eosin–stained whole slides were prepared for all tissue specimens and subsequently histopathologic reexamined by three consultant breast pathologists (J. Wesseling, E.J. Groen, and K. van de Vijver). Slides were assessed on morphologic characteristics, including DCIS architecture and nuclear grade, the presence of calcifications and necrosis, and microenvironmental characteristics like stromal features and the presence of lymphocytes. This was done blinded of case or control status. In addition, for every patient, a representative part of the lesion was selected for further evaluation. Clinical characteristics of patients in- and excluded in this study are presented in Supplementary Table S2. Pathology review data were available for 200 case–control sets, including at least one control per case, resulting in a case–control series of 200 cases and 474 controls, which was representative of the original case–control selection (Supplementary Table S3).

IHC staining was used to identify DCIS phenotypes using slides from FFPE tissue blocks of 185 DCIS cases and 420 DCIS controls. Some tissue specimens were excluded because insufficient tissue material was available for IHC (12 cases and 26 controls; Supplementary Table S1). DCIS lesions were scored for estrogen receptor (ER) and progesterone receptor (PR) status, overexpression of HER2, Ki67, and expression of the tumor suppressor proteins p16 and p53, and COX-2. These markers were selected because the antigens have been associated with subsequent IBC after DCIS, based on multivariable analyses, and these results had been reported previously in at least two articles (15, 20). Moreover, the ability to perform good-quality IHC on FFPE material was decisive. Details about the used antibodies, IHC staining procedure, and scoring criteria can be found in Supplementary Materials and Methods. All antibodies used in this study were previously tested in our laboratory using normal tissue and tumor samples known to contain the antigens.

Logistic regression models, conditional on matched sets, were used to assess associations of various clinical and histopathologic characteristics with subsequent iIBC risk after primary DCIS. Wald-based 95% confidence intervals (95%CI) and P values are reported for overall effect for factors with more than two categories.

Variables were selected for inclusion in multivariable models based on a P value ≤0.1 in univariate analyses. Because of the amount of missing data, margin status and lesion size were excluded from the multivariable models. In addition, because histologic grade was correlated (r > 0.4) with necrosis and periductal fibrosis, this variable was excluded from the multivariable models. Subsequently, the likelihood ratio (LR) χ² was used to identify the models with the strongest association with subsequent iIBC development.

Approximate cumulative incidence of subsequent iIBC by HER2 and COX-2 status was estimated using the iIBC ORs for HER2 and COX-2 status and cumulative risk of iIBC in the entire cohort. Death due to causes other than breast cancer was considered as a competing risk in this analysis. The expected cumulative incidence of breast cancer for our study population was derived from age-specific breast cancer incidence and all-cause mortality rates in the Dutch female population using the Hakulinen method (21).

All statistical analyses were performed using Stata/SE (version 13.1, StataCorp). P values ≤0.05 were considered statistically significant.

Clinical characteristics of cases and controls were comparable (Table 1): For both cases and controls, median age was 57 years, main period of DCIS diagnosis was 1989 to 1998, and around 50% of DCIS was screen-detected. For cases, the median time to iIBC was 6.2 years (range, 0.5–19.2). Ninety-five percent of all subsequent iIBC lesions recurred at or near the side of the DCIS excision (data not shown). Furthermore, of 141 cases (71%), we were able to assess the ER status of the matched subsequent IBC, which showed an agreement of 89% between the primary DCIS and matched IBC (data not shown).

The presence of periductal fibrosis was associated with increased risk of subsequent iIBC (OR = 1.44; 95% CI, 1.01–2.06) compared with women who did not develop iIBC (Table 2). Furthermore, there was a trend toward a larger lesion size (P = 0.08) and more frequent positive resection margins (P = 0.06) among cases as compared with controls. However, it should be stressed that around 65% of data on lesion size and around 15% of margin data are missing within our case–control series. There was a trend for necrosis (P = 0.06), periductal lymphocytes (P = 0.13), and high histologic grade (P = 0.08) to be more often present in DCIS lesions of women who subsequently developed iIBC. DCIS architecture, calcifications, and periductal lymphocytes were not associated with subsequent invasive disease. The IHC markers HER2 (OR = 1.56; 95% CI, 1.05–2.31) and COX-2 (OR = 2.97; 95% CI, 1.72–5.10) were associated with subsequent iIBC risk, but ER, PR, p16, and p53 expression and IHC subtype were not associated with subsequent iIBC risk (Table 3). Ki67 was excluded from the analysis because of unreliable staining results.

COX-2 was also significantly associated with the risk of subsequent iIBC in multivariable analyses (Table 4; Supplementary Table S4). Subsequent invasive disease was significantly associated with high COX-2 expression in combination with: (i) overexpression of HER2 (LR χ² = 6.47; OR = 3.98); (ii) the presence of periductal fibrosis (LR χ² = 6.34; OR = 4.87); or (iii) the presence of necrosis (LR χ² = 5.18; OR = 5.76). Combination of periductal fibrosis, HER2, and COX-2 was also significantly associated with subsequent iIBC, although achieving a lower LR χ² ratio (LR χ² = 4.98; OR = 4.63; Table 4). When combining COX-2, HER2, periductal fibrosis, and necrosis, the addition of necrosis deteriorated the performance of the model, because necrosis is positively correlated with HER2 (data not shown). Eighty-seven percent of DCIS lesions associated with subsequent iIBC (DCIS cases) showed high expression of COX-2. Of this subset of COX-2^high DCIS lesions, 34% was HER2 positive, and in 37%, periductal fibrosis was present (Table 2; ref. 3). HER2 overexpression was most frequently accompanied with high COX-2 expression. In contrast, high COX-2 expression was independent of HER2 overexpression. The cumulative risk of subsequent iIBC of HER2⁺/COX-2^high DCIS was almost 4 times higher as compared with the risk for HER2⁻/COX-2^low DCIS lesions (Table 4). Furthermore, analysis of IHC data of 141 DCIS and matched IBC pairs showed that of the 43 patients with HER2⁺/COX-2^high DCIS that subsequently developed an iIBC, 35% developed an ER⁻ invasive recurrence (Table 5).

In our study group, the estimated overall 10-year and 15-year cumulative incidence of iIBC were 10.9% and 13.8%, respectively. For patients with HER2⁻/COX-2^low DCIS, the estimated 10-year and 15-year cumulative incidence of iIBC were 4.8% and 6.0%, respectively; for patients with HER2⁺/COX-2^low DCIS, 4.5% and 5.6%; for patient with HER2⁻/COX-2^high DCIS, 11.3% and 14.3%; and for patients with HER2⁺/COX-2^high DCIS, 18.1% and 22.8%, respectively (Fig. 1). Within our study group, 29.7% of cases and 18.3% of controls had this unfavorable DCIS subtype of HER2⁺/COX-2^high DCIS.

The positive predictive value of the HER2⁺/COX-2^high DCIS subtype is 42% (sensitivity, 31%; specificity, 81%), indicating that more than half of the HER2⁺/COX-2^high DCIS lesions are not associated with invasive recurrence and thus are false positives. The strength of this marker combination is most likely in its negative predictive value (NPV = 73%), indicating that the risk of subsequent iIBC after DCIS is low in the non-HER2⁺/COX-2^high subgroup.

In this study, we identified promising risk factors for progression of DCIS into iIBC, by conducting a nested case–control study comparing women who did and did not develop invasive disease after primary DCIS. To avoid confounding by radiation effects, we analyzed all women treated by BCS alone, within our well-characterized nation-wide population-based cohort of women diagnosed with primary DCIS between 1989 and 2005 in the Netherlands with a median follow-up time of 12.0 years. The large size of our series, the design applied, the comprehensive data, and the long-term follow-up are essential to overcome limitations due to bias and lack of power of small sample series with a relatively short follow-up.

We found that initial DCIS lesions with HER2⁺/COX-2^high expression were associated with increased risk of subsequent iIBC, which are often ER⁻ (35%). This is an important finding, as ER⁻ tumors have in general a worse prognosis than ER⁺ breast cancers. Women diagnosed with this unfavorable DCIS subtype had a 4-fold higher risk and an estimated 22.8% cumulative 15-year risk of developing subsequent iIBC. This was higher than the overall cumulative incidence of iIBC in this patient cohort, that is, 13.8% at 15 years. The estimated cumulative risk for patients with HER2⁻/COX-2^low and HER2⁺/COX-2^low DCIS was comparable with the risk for the general population, that is, 4–6% at 15 years. The positive and negative predictive value of the HER2/COX-2 marker combination is 42% and 73%, respectively. For clinical purposes, the predictive values should definitely be improved.

Extensive granular cytoplasmic expression of COX-2 was the marker that had the strongest association with subsequent iIBC both in univariate and multivariable analysis. This is in line with one previous study that showed an association between COX-2 and subsequent IBC in univariate analysis (22). We also found an association of subsequent iIBC with HER2⁺ primary DCIS, and presence of periductal fibrosis. HER2 overexpression was associated with subsequent iIBC in univariate analysis, which was also found in two previous studies (23, 24). Interestingly, results from our multivariable analysis showed that HER2 overexpression is not predictive for subsequent iIBC in the absence of high COX-2 expression. This indicates that the prognostic value of HER2 overexpression in the risk of subsequent iIBC is probably limited and the increase in risk is primarily driven by COX-2 overexpression. Another prognostic factor we found was periductal fibrosis. This microenvironmental factor was associated with subsequent iIBC in both univariate and multivariable analysis and is supported by a previous study (24). So far, diagnosis of breast disease has been limited to the morphologic interpretation of epithelial cells and the assessment of epithelial tissue architecture, in which the stromal compartment is largely ignored. The role of periductal fibrosis described here underlines the importance of assessment of the DCIS stromal compartment.

There is biological support for a role of COX-2 in IBC recurrence. COX-2 is a cytoplasmic enzyme involved in prostaglandin synthesis. It is induced rapidly in response to growth factors, tumor promotors, hormones, and cytokines (25). It has been shown that overexpression of COX-2 leads to an increased prostaglandin E2 (PGE2) level, which can potentially affect most of the key processes in cancer development, including proliferation, resistance to apoptosis, angiogenesis, immunosuppression, and invasion (26). Furthermore, overexpression of COX-2 has been shown to result in p16-mediated cell-cycle arrest through the p16/Rb signaling pathway. When the p16/Rb signaling pathway is disrupted, cellular proliferation continues, resulting in additional high Ki67 expression in the presence of high p16 and high COX-2 expression (27). It has previously been shown that p16⁺/COX-2⁺/Ki67⁺ DCIS is associated with subsequent iIBC (28). Unfortunately, in our study, we were unable to assess p16/COX-2/Ki67 protein expression, because Ki67 suffers from loss of antigenicity over time and is very sensitive to improper formalin fixation (29, 30).

Next to the factors described above, a wide range of other prognostic factors have been reported in literature related to recurrent disease, albeit with small effect sizes (1, 15). In contrast to our study, most studies did not discriminate between invasive and in situ recurrences as a primary endpoint (31–33). As invasive recurrences may lead to breast cancer mortality, it is of utmost importance to make a distinction between in situ and invasive recurrences in risk prediction.

Our study group comprised of patients with DCIS diagnosed between 1989 and 2005 and treated by BCS alone. Regarding this time period, we have to keep in mind that treatment strategies and screening techniques for DCIS have evolved over the years, which may have impacted treatment or other care for these patients. The main period of DCIS diagnosis of our patient group was 1989 to 1998 and about 50% of DCIS was screen-detected. Within the time period 1989 to 1998, guidelines for DCIS treatment in the Netherlands recommended mastectomy or BCS alone. From 1999, the addition of radiotherapy after BCS was included. Clinical trials have shown that adjuvant radiotherapy reduces the risk of both in situ and invasive recurrence with about 50% (34, 35). Moreover, our group has shown that women diagnosed with DCIS between 1999 and 2004 were less likely to develop iIBC than women diagnosed between 1989 and 1998, regardless of treatment and age (18). With the introduction of digital mammography, the coverage and sensitivity of screening improved significantly and led to an increase in the percentage of screen-detected DCIS (36).

In our study group, the estimated overall 10-year and 15-year cumulative incidence of iIBC were 10.9% and 13.8%, respectively. This is comparable with the 10-year cumulative incidence reported in two nonrandomized prospective studies of women with DCIS treated by BCS alone (37, 38).

Our study has several strengths. First, our study is nested in a large, unique population-based study of women with DCIS treated by BCS alone that provides information on clinical, histopathologic, and IHC characteristics and focuses specifically on subsequent ipsilateral IBC with a median follow-up time of 12.0 years. This well-annotated nature of our patient series enabled us to prevent a great amount of bias often seen in previous studies. Second, we were able to collect 85% of the requested tissue blocks, from which new whole slides were developed for reassessment by specialized breast pathologists. Third, our large sample allowed us to assess the combinations of clinical, histopathologic, and IHC characteristics, that were independently associated with subsequent invasive disease, by using multivariable models.

Our study also has some limitations. First, we were only able to assess prognostic factors for a subset of all women included in the case–control study because this depended on the availability of FFPE tissue blocks in the participating hospitals. Fortunately, this did not cause significant bias, as these samples were missing randomly (hospital participation) and the group for which we could successfully collect the tissue blocks was an excellent representation of the complete case–control selection based on the patients' clinical characteristics. Second, the interpretation of IHC markers can be challenging because of the heterogeneous expression of certain proteins. Nevertheless, we succeeded in minimizing the interobserver variability by our well-designed scoring method. Third, margin status and size of the DCIS lesion were not known in 15% and 65% of the cases, respectively, as these data were not always routinely described in these old retrospective series. As far as margin status and DCIS lesions size were known, no statistically significant differences were present between the two groups, that is, cases and controls. Missing data were also equally randomly distributed among these groups (Table 2) and thus were not related to the outcome of interest. Therefore, resulting over- or underrepresentation of some factors is unlikely. Fourth, we did not have data on BRCA status or family history of breast cancer of our patient group. Yet, it has to be taken into account that BRCA status could be a confounding factor in subsequent iIBC development.

In summary, identification of prognostic factors has the potential to improve the clinical management of women diagnosed with DCIS. We found a prognostic role for COX-2, HER2, and periductal fibrosis. In addition, women diagnosed with HER2⁺/COX-2^high DCIS and treated by BCS alone had a 4-fold higher prevalence of subsequent iIBC than women with HER2⁻/COX-2^low DCIS lesions. Furthermore, HER2⁺/COX-2^high DCIS was associated with ER⁻ invasive recurrences. Our results underline the importance of assessment of the DCIS stromal compartment and protein expression of HER2 and COX-2 to estimate the risk of subsequent invasive disease after a diagnosis of DCIS. Patients with COX-2^low DCIS are at lowest risk of iIBC as their risk is comparable with the general population. Our study design and unique retrospective patient series provided us with excellent candidate prognostic markers for use in personalized patient risk stratification. As a next step, these prognostic markers will need to be validated in independent datasets, as a major step to distinguish harmless from potentially hazardous DCIS.

No potential conflicts of interest were disclosed.

Conception and design: L.L. Visser, L.E. Elshof, M. Schaapveld, K. van de Vijver, F.E. van Leeuwen, E.J. Rutgers, M.K. Schmidt, E.H. Lips, J. Wesseling

Development of methodology: L.L. Visser, L.E. Elshof, E.J. Groen, F.E. van Leeuwen, M.K. Schmidt, E.H. Lips, J. Wesseling

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.L. Visser, L.E. Elshof, K. van de Vijver, E.J. Groen, M.M. Almekinders, J. Wesseling

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.L. Visser, M. Schaapveld, K. van de Vijver, C. Bierman, E.J. Rutgers, M.K. Schmidt, E.H. Lips, J. Wesseling

Writing, review, and/or revision of the manuscript: L.L. Visser, L.E. Elshof, M. Schaapveld, K. van de Vijver, E.J. Groen, F.E. van Leeuwen, E.J. Rutgers, M.K. Schmidt, E.H. Lips, J. Wesseling

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.L. Visser, L.E. Elshof, K. van de Vijver, J. Wesseling

Study supervision: L.L. Visser, E.J. Rutgers, M.K. Schmidt, E.H. Lips, J. Wesseling

We thank all patients whose stored residual material was used in this study and their treating physicians, all the collaborating hospitals and especially the pathology departments, and all persons who helped in the data collection and analysis. Specifically, we acknowledge Otto Visser, Annemarie Eeltink, and the registration teams of the Netherlands Comprehensive Cancer Organization for the collection of data for the Netherlands Cancer Registry; Lucy Overbeek and PALGA, the nationwide histopathology and cytopathology data network and archive, for providing pathology data; the pathologists of the PALGA-group: Drs. P. Bult, C.H.M. van Deurzen, E. Geuken, and C.M. van Dijk for their cooperation; Frank Nieboer, Danah Kamphuis, and Chaima Loukili for their assistance; and the NKI-AVL Core Facility Molecular Pathology & Biobanking (CFMPB) for their lab support. This work was supported by Pink Ribbon (grant number 2013.WO29) and A Sisters Hope (grant number 2011.WO19.C88). The funders had no role in the study design, data collection, analysis and data interpretation, or the writing of the report.