The prognostic significance of multiple morphologic features and biologic markers in ductal carcinoma in situ of the breast
A study of a large cohort of patients treated with surgery alone
Abstract
BACKGROUND
A number of conventional histopathologic features have been associated with recurrence of ductal carcinoma in situ (DCIS) after surgery alone and are included in the Van Nuys Pathologic Classification and Prognostic Index. To the authors' knowledge, very little is known regarding the prognostic significance of the many biologic markers that have been studied in DCIS in the past decade.
METHODS
Clinical and pathologic data were analyzed from 151 patients who underwent wide local excision alone for DCIS that was diagnosed by mammography or as an incidental finding between 1982 and 2000. Using local disease recurrence as an endpoint, the authors sought to determine the prognostic significance of a large number of histopathologic parameters as well as biologic markers (estrogen receptor [ER], progesterone receptor [PR], p53, HER-2/neu, Ki-67, p21, and bcl-2), as determined by immunohistochemical staining of contemporary or archival tissue.
RESULTS
With a median follow-up of 65 months, 42 recurrences were reported to occur between 11 months and 97 months after definitive surgery. In a univariate analysis, tumor size, Van Nuys pathologic classification, and degree of necrosis demonstrated significant correlations with the rate of recurrence. Tumor size, necrosis, nuclear grade, and comedonecrosis were found to be associated significantly with the time to disease recurrence. None of the biologic markers demonstrated a significant association with the rate of recurrence or the time to disease recurrence. In a multivariate analysis, only large tumor size (Van Nuys 2 or 3) and higher degrees of necrosis (Van Nuys 2 or 3) were found to be associated significantly with both the rate of recurrence and the time to recurrence. No biologic marker showed a significant correlation with recurrence. Using Classification and Regression-Tree Analysis and Tree-Structured Survival Analysis, PR > 3.5% and bcl-2 < 97.5% were associated with a higher recurrence rate in the subgroup of patients with small tumor size (Van Nuys size 1) and higher degrees of tumor necrosis (Van Nuys 2 or 3).
CONCLUSIONS
The current results confirmed the value of conventional histopathologic parameters, as outlined in the Van Nuys classification system, in predicting local recurrence of DCIS. Using traditional logistic analyses, no significant correlation was found between a variety of biologic markers and disease recurrence. Cancer 2004. © 2004 American Cancer Society.
Because breast conservation has been adopted as a goal of treatment for patients with ductal carcinoma in situ (DCIS) of the breast, and because wide local excision alone without radiation has been championed as appropriate treatment for selected patients with DCIS, the correct treatment choice for patients with DCIS has become something of a dilemma.1 To date, a precise algorithm to categorize patients with DCIS so that an appropriate therapeutic recommendation can be made has proved elusive.
A series of publications over the last 20 years by the Nottingham Group in England has established generally accepted major pathologic prognostic factors in invasive breast carcinoma.2 These include tumor size, histologic type and grade, lymph node status, and hormone receptor status.2, 3 Conventional prognostic histopathologic factors also have been studied in DCIS, with varying results. In the National Surgical Adjuvant Breast and Bowel Project (NSABP) study B-17, which examined the effect of adding radiation therapy to excision alone in patients with DCIS, moderate-to-marked comedonecrosis was the single histologic feature that was found to predict for recurrent disease.4 In a 15-year follow-up of patients who were treated with excision and irradiation for DCIS, Solin et al. found that no unique histopathologic parameter was associated with local recurrence.5 Ottesen et al. found that comedonecrosis, large nuclei, and tumor size > 1cm were significant predictors of local recurrence after excision alone.6 Silverstein et al. and the Van Nuys group created a DCIS pathologic classification based on nuclear grade and the presence or absence of comedonecrosis.7 The Van Nuys (VN) Prognostic Index builds on this classification with the addition of two parameters with prognostic significance, as identified in multivariate analysis, namely, tumor size and, more recently, margin width.8, 9
The last decade has witnessed the emergence of increasing numbers of biologic markers to characterize breast neoplasms. These markers typically reflect alterations in genes that regulate cell growth, development, and proliferation. HER-2/neu and p53, for example, have been applied widely to invasive breast carcinoma; their overexpression appears to correlate with a poor prognosis.10, 11 Numerous reports also describe the biologic marker profiles of patients with DCIS and the correlations between and among markers such as HER-2/neu, p53, bcl-2, and hormone receptors.12-14 However, to our knowledge there is very little information available concerning the prognostic significance of biologic markers in DCIS. In the current study, in addition to conventional histopathologic parameters, we chose to examine the following biologic markers in patients with DCIS: estrogen receptor (ER), progesterone receptor (PR), p53, Ki-67, HER-2/neu, bcl-2, and p21.
ER and PR were among the first biologic markers studied in breast carcinoma. Although early studies15 suggested that ER status was of significant prognostic value in invasive disease, more recent thinking is that hormone receptor status is less of an independent prognostic factor than it is a predictor of response to hormonal manipulation.2, 3
The p53 gene is a tumor suppressor gene with a protein product that regulates the cell cycle negatively by causing arrest in the G1 phase.16 Gene mutation of p53 is inferred by immunohistochemical expression of the p53 protein.
Ki-67 is a nuclear protein that is expressed from the late G1 phase through the M-phase of the cell cycle, but not in resting (G0) cells. It generally correlates with the S (synthetic)-phase fraction17 and is useful as a proliferation marker.
HER-2/neu is a protooncogene with a protein gene product, p185, which has structural similarities to the epidermal growth factor family and participates in signal transduction and cellular proliferation. Overexpression of HER-2/neu is reported to occur at a significantly higher rate (≈ 60%) in DCIS compared with invasive breast carcinoma (20–30%).18
The bcl-2 protein inhibits programmed cell death without stimulating cellular proliferation.19 bcl-2 protein expression, as demonstrated by immunohistochemical staining in invasive and intraductal breast carcinoma, has been correlated directly with favorable prognostic features such as well differentiated histology and positive ER status,20-22 and has been correlated inversely with unfavorable features such as HER-2/neu and p53 expression.14
p21 is a nuclear-based protein that modulates the cell cycle by interrupting the transition from G1 to S-phase. Expression of p21 is induced by the gene product of wild-type p53 in response to DNA damage.23, 24 Loss of normal p53 function results in loss of checkpoint control mechanisms mediated by p21, thereby contributing to instability.25 It is believed that overexpression of the p21 protein results primarily from induction through either a p53-dependent or p53-independent pathway.26
For more than 20 years, one of us (G.F.S.) has been treating patients with subclinical DCIS with wide excision alone and close clinical follow-up. A report on the first 70 patients treated between 1978 and 1990 appeared in 1992.27 In general, the major criterion for treatment by excision alone was a greatest tumor dimension ≤ 2.5 cm, as measured on a mammogram in the region of abnormal calcifications, or calcifications measuring < 6 cm2 if a single dimension could not be measured accurately. In a few patients, 3.5 cm was accepted as the greatest dimension of calcifications if the breast was large enough to accommodate a wider excision without a major cosmetic defect (1 patient in this series had a greatest tumor dimension that measured 5.0 cm). High nuclear grade or so-called comedonecrosis did not exclude patients from undergoing local excision alone, although, in each of these patients, care was taken to ensure a wider, clear margin. Macroscopically and microscopically clear margins were achieved in every patient. In the current study, we sought to determine the prognostic significance of a large number of histopathologic features and biologic markers as detected by immunohistochemistry. The number of patients available for analysis (n = 151 patients) has more than doubled since the initial report, making the current study, to the best of our knowledge, one of the largest with adequate follow-up in which the natural history of DCIS could be evaluated after wide excision alone and unaffected by radiation or systemic therapy. The results obtained concerning these 151 patients form the basis of this report.
MATERIALS AND METHODS
Patients
Data regarding 151 patients treated between February 1982 and August 2000 were collected and analyzed. The age of patients at the time of the initial diagnosis ranged from 31–88 years. All patients in the current study underwent primary wide surgical excision, reexcision, or both by one of the authors (G.F.S.). To be eligible for inclusion in this study, all the following criteria had to be met: 1) DCIS detected by mammography or as an incidental pathologic finding related to surgery for benign breast disease; 2) negative specimen margins at the conclusion of surgical excision or reexcision (no hormonal therapy, chemotherapy, or radiation therapy); 3) no evidence of concurrent malignancy elsewhere; and 4) a minimum follow-up of 15 months.
Definitions
The date of diagnosis and the date of recurrence were defined as the date of the surgery that established the pathologic diagnosis. Local recurrence of DCIS or invasive breast carcinoma constituted an endpoint with respect to data analysis, so that radiation therapy or systemic therapy administered after disease recurrence did not affect the results. Tumor size was determined by mammographic, macroscopic pathologic, and (when available) microscopic measurements. If the area of suspicious calcification on the mammogram was roughly circular or spherical, then the greatest dimension was used. If the suspicious area was rectangular, then the area was calculated using length × width, and the best approximation to the VN Prognostic Index was made. Tumor size also was derived from the number of pathology slides containing DCIS. In some patients, direct tumor measurement was made microscopically. Size was categorized according to the VN Prognostic Index8: category 1, ≤ 15 mm; category 2, 16–40 mm; category 3: ≥ 41 mm.
In the vast majority of patients, the surgical margins were shaved after the primary area of concern was excised from the breast. Five marginal specimens were submitted, namely, the shaved superior, inferior, medial, lateral, and deep margins.28 Each of these specimens was separated from the others and examined microscopically. DCIS in any area of a shaved margin was considered “positive;” in such patients; reoperation was employed until the margins were clear. If this could not be accomplished, then the treatment plans were altered. All patients who were included in the current study were found to have negative surgical margins at the conclusion of treatment. When this study began, the prognostic significance of margin width had not been established, and specific measurement of margin width was not performed. Lack of a protocol for handling tissue for measurement of margins precluded retrospective measurements.
Histopathologic Features
Hematoxylin and eosin-stained slides and immunohistochemical slides were reviewed by two of the four pathologist coauthors (D.B.C., S.A.G., J.P.P., or R.S.). When a significant difference of opinion occurred, the problem was referred to a third pathologist for resolution. With regard to microscopic histopathologic features, the guidelines of the Consensus Conference of the Classification of Ductal Carcinoma in Situ28 were used to assess nuclear grade. Briefly, low-grade (Grade 1) nuclei are monotonous with finely dispersed chromatin and occasional nucleoli; high-grade (Grade 3) nuclei are larger, pleomorphic nuclei with vesicular chromatin and prominent nucleoli; and intermediate-grade (Grade 2) nuclei are neither low-grade nor high-grade. The following histologic patterns were assessed and graded according to the degree to which they were represented in each tumor: cribriform, solid, papillary, and micropapillary (Grade 0, absent; Grade 1, minimal; Grade 2, moderate; and Grade 3, marked). For example, a tumor that showed a pure cribriform pattern would be graded as cribriform (Grade 3), solid (Grade 0), papillary (Grade 0), and micropapillary (Grade 0). A tumor with approximately equal amounts of solid and cribriform patterns would be graded as solid (Grade 2), cribriform (Grade 2), papillary (Grade 0), and micropapillary (Grade 0). A similar grading system also was applied to the following microscopic features: calcification within the tumor, calcification within the breast but outside the tumor, inflammation, tumor necrosis, and comedonecrosis. At the time of the current study, the Department of Pathology at Thomas Jefferson University Hospital used a definition of comedonecrosis that required high nuclear grade of solid intraepithelial growth along with central zone necrosis (the presence of ghost cells and karyorrhectic debris28) within a duct.
The VN pathologic classification system employs a definition of comedonecrosis (“central lumina containing necrotic debris surrounded by large pleomorphic viable cells in solid masses”7) that does not require a specific amount of high nuclear grade. To translate our data into the VN pathologic classification for the current study, we used the VN definition of comedonecrosis. Minimal individual tumor cell necrosis was ignored. The VN pathology classification was calculated for each tumor according to Silverstein et al.7: VN 1, nuclear Grade 1 or 2 without comedonecrosis; VN 2, nuclear Grade 1 or 2 with comedonecrosis; and VN 3, nuclear Grade 3 with or without comedonecrosis.
Biologic Markers
The following biologic markers were evaluated (listed as antibody, dilution, and manufacturer, respectively): ER, 1:50 dilution (Dako Corporation USA, Carpinteria, CA); PR, 1:80 dilution (NovoCastra Laboratories, New castle-upon-Tyne, U.K.); p53, 1:160 dilution, (NovoCastra); HER-2/neu (c-erb-B2), 1:100 dilution (Signet Laboratories, Dedham, MA); Ki-67, 1:100 dilution (Immunotech, Westbrook, ME); bcl-2, 1:40 dilution (Dako Corporation USA); and p21, 1:100 dilution (Oncogene Sciences, Cambridge, MA). Five-micron sections of formalin fixed, paraffin embedded tissue were cut onto neoprene-coated slides (Aldrich Chemical, Milwaukee, WI). Routine deparaffinization and rehydration were performed on the Leica Autostainer (Leica, Inc., Deerfield, IL). Antigen recovery was performed in a Panasonic microwave oven (model NN-5602A, 800 Watts) (Panasonic, Franklin Park, IL). Slides were submersed in 200 mL of ChemMate H.I.E.R. buffer (pH 5.5–5.7) (Ventana Medical Systems, Tucson, AZ) and microwaved on the high setting for 5 minutes. Then, 50 mL distilled water were added to replenish evaporative loss, and an additional 5 minutes of high-setting microwaving was performed, followed by 20 minutes of cooling outside the oven. Slides then were washed in distilled water and placed in ChemMate buffer 1 (phosphate-buffered saline [PBS] containing carrier protein and sodium azide). Immunostaining was performed on the Techmate 1000 instrument (Ventana Medical Systems) according to the manufacturer's suggested techniques. A 20-minute incubation in normal serum was followed by overnight incubation with the chosen primary antibodies. The next day, a PBS wash was followed by a 30-minute incubation with a biotinylated secondary antibody, another wash, and 3 7-minute applications of 3,3′ diaminobenzidine tetrahydrochloride (DAB)/peroxide. The slides were then washed and counterstained with hematoxylin.
Staining of non-Techmate slides was performed on the Cadenza Automated Immunostainer (Shandon, Inc., Pittsburgh, PA). Ten minutes of incubation with normal horse serum (1:50 dilution; Vector Laboratories, Burlingame, CA) was followed by overnight primary antibody incubation. The next day, a buffer wash was followed by a 30-minute application of biotinylated horse-antimouse antibody (1:200 dilution; Vector Laboratories), another wash, a 30-minute incubation of avidin-biotin complex (Vector Laboratories), a subsequent wash, and 3 consecutive 8-minute applications of DAB/peroxide. Slides then were placed on the Leica Autostainer for counterstaining with Harris hematoxylin (Surgipath, Richmond, IL), dehydration to xylene, and bluing with 3% lithium carbonate. All immunohistochemical stains were accompanied by appropriate positive and negative controls. The percentage of positive cells in the tumor population was determined by visual estimation for ER, PR, p53, Ki-67, bcl-2, and p21. Only nuclear staining was considered in the evaluation of ER, PR, p53, Ki-67, and p21. HER-2/neu tended to exhibit either complete membrane negativity or complete positivity and was scored according to the intensity of staining of tumor cells as follows: 0, negative; 1, weak staining; 2, moderate staining; and 3, strong staining. bcl-2 was evaluated for cytoplasmic positivity. Staining for p21 often exhibited marked heterogeneity from field to field within tumors; thus, scoring for p21 represented an estimate of overall positivity.
Statistical Methods
Correlation between the biologic markers was estimated using the Spearman correlation coefficient. Association between the histopathologic variables, all of which were ordinal, was measured using the exact version of the Goodman and Kruskal γ statistic.29 This measure has the same meaning as a correlation coefficient, with values close to 1 implying a nearly perfect positive association. Key biologic marker variables and histopathologic variables, except for the VN pathology classification, were dichotomized in all analyses except for the Classification and Regression Tree (CART) analysis and the Tree-Structured Survival Analysis (Tables 1 and 2). Associations between the recurrence rate with dichotomous variables were compared univariately using the Fisher exact test. The Mantel–Hanszel test for trend was used to assess whether higher recurrence rates were associated with higher VN pathology classification. The time to disease recurrence was modeled using the Kaplan–Meier estimator, and the differences in survival rates were tested using the log-rank test.
| Variable | No. of patients (%) | P valuea | |
|---|---|---|---|
| Without recurrence | With recurrence | ||
| VN size | |||
| 1 | 92 (78.6) | 25 (21.4) | 0.002 |
| 2 or 3b | 17 (50.0) | 17 (50.0) | — |
| VN pathology | |||
| 1 | 37 (82.2) | 8 (17.8) | 0.035c |
| 2 | 44 (72.1) | 17 (27.9) | 0.255d |
| 3 | 28 (62.2) | 17 (37.8) | 0.059e |
| Necrosis | |||
| 0 or 1 | 62 (82.7) | 13 (17.3) | 0.006 |
| 2 or 3 | 47 (61.8) | 29 (38.2) | — |
| Nuclear grade | |||
| 1 or 2 | 81 (76.4) | 25 (23.6) | 0.111 |
| 3 | 28 (62.2) | 17 (37.8) | — |
| Comedonecrosis | |||
| 0 | 83 (76.1) | 26 (23.9) | 0.105 |
| 1–3 | 26 (61.9) | 16 (38.1) | — |
| Cribriform | |||
| 0 | 40 (75.5) | 13 (24.5) | 0.571 |
| 1–3 | 69 (70.4) | 29 (29.6) | — |
| Solid | |||
| 0 | 28 (82.4) | 6 (17.6) | 0.191 |
| 1–3 | 81 (69.2) | 36 (30.8) | — |
| Papillary | |||
| 0 | 94 (74.0) | 33 (26.0) | 0.447f |
| 1–3 | 15 (65.2) | 8 (34.8) | — |
| Micropapillary | |||
| 0 | 92 (74.2) | 32 (25.8) | 0.245 |
| 1–3 | 17 (63.0) | 10 (37.0) | — |
| Other calcification | |||
| 0 | 73 (68.9) | 33 (31.1) | 0.235 |
| 1–3 | 36 (80.0) | 9 (20.0) | — |
| Inflammation | |||
| 0 | 68 (78.2) | 19 (21.8) | 0.156g |
| 1–3 | 41 (89.1) | 5 (10.9) | — |
- VN: Van Nuys pathologic classification and prognostic index.
- a Results of Fisher tests of equality of rates are given in the “With recurrence” column.
- b Only 1 patient had Van Nuys size = 3, and 33 patients had Van Nuys size = 2.
- c Mantel–Hanszel test of trend.
- d Testing equality of recurrence rates for Van Nuys pathology 1 and Van Nuys pathology 2.
- e Testing equality of recurrence rates for Van Nuys pathology 1 and Van Nuys pathology 3.
- f Data were not available concerning one patient with recurrent disease.
- g Data were not available concerning 18 patients with recurrent disease.
| Variable | Time to recurrence (mos) | P valuea | ||
|---|---|---|---|---|
| First quartile | 95% CI | |||
| LL | UL | |||
| VN size | ||||
| 1 | 97 | 64 | < 0.001 | |
| 2 or 3 | 27 | 19 | 62 | — |
| Nuclear grade | ||||
| 1 or 2 | 85 | 65 | NE | 0.029 |
| 3 | 27 | 23 | — | |
| Necrosis | ||||
| 0 or 1 | 112 | 71 | 0.004 | |
| 2 or 3 | 39 | 24 | 66 | — |
| Comedonecrosis | ||||
| 0 | 85 | 64 | NE | 0.025 |
| 1–3 | 26 | 23 | — | |
| VN pathology | ||||
| 1 | NR | 62 | 0.056 | |
| 2 | 71 | 64 | NE | — |
| 3 | 27 | 23 | — | |
| Cribriform | ||||
| 0 | 65 | 24 | 0.800 | |
| 1–3 | 66 | 37 | 112 | — |
| Solid | ||||
| 0 | NR | 39 | 0.182 | |
| 1–3 | 64 | 35 | 97 | — |
| Papillary | ||||
| 0 | 66 | 44 | NE | 0.362 |
| 1–3 | 26 | 16 | — | |
| Micropapillary | ||||
| 0 | 71 | 48 | NE | 0.163 |
| 1–3 | 28 | 19 | — | |
| Other calcification | ||||
| 0 | 48 | 29 | 112 | 0.142 |
| 1–3 | 95 | 48 | — | |
| Inflammation | ||||
| 0 | 97 | |||
| 1–3 | NR | 66 | NE | 0.122 |
- 95% CI: 95% confidence interval; LL: lower limit; UL: upper limit; VN: Van Nuys pathologic classification and prognostic index; NE: not evaluable; NR: not reached.
- a Log-rank test.
Multivariate analyses were performed with all markers and histopathologic variables. For these analyses, we initially considered those possible subgroups of markers and histopathologic variables that were not associated highly (Spearman correlation coefficient ≤ 0.3 or Goodman and Kruskal γ ≤ 0.5) and then excluded nonsignificant variables.
We modeled the rates of disease recurrence using logistic regression models. Multivariate models for the time to recurrence were fit as Cox proportional hazards models. Tests of the proportional hazards assumptions, based on scaled Schoenfeld residuals,30 were used to verify the validity of the Cox models.
To define potentially important marker cut-off points associated with disease recurrence, CART analysis was performed on the data from all patients, including morphologic variables and nondichotomized biologic markers. For the time-to-recurrence analysis, we used the Tree-Structured Survival Analysis functions written in S-Plus by Wagner and Segal. These functions implement statistical methods described by Segal.31 In the CART analysis and the Tree-Structured Survival Analysis, we allowed the model to select the best predictors of recurrence from the pool of covariates that contained the morphologic variables and the seven biologic markers. In the context of these models, sensitivity measures the percent of correctly classified recurrences, and specificity measures the percent of correctly classified nonrecurrences. A patient is classified with recurrent disease if the rate of recurrence in her subgroup is higher than the overall rate of recurrence in the data set and is classified as nonrecurrent if the rate of recurrence in her subgroup is lower than the overall rate of recurrence in the data set. The data were analyzed using SAS software (version 8.0; SAS Institute, Inc., Cary, NC), S-Plus software (Insightful, Seattle, WA), StatXact (version 4.0.1) and LogXact (version 4.0) software (both from Cytel Software Corporation, Cambridge, MA), and CART (Salford Systems, San Diego, CA).
RESULTS
The median follow-up duration was 65 months for all patients, with a mean follow-up duration of 73.1 months. Follow-up for patients without recurrent disease ranged from 15–201 months, with a median of 86 months.
Univariate Results
Rate of recurrence.
Forty-two recurrences occurred among 151 patients between 11–97 months after initial surgery (median time to recurrence, 28.5 months). The median time to recurrence for the entire group had not been reached in the current study by the time of last follow-up because these 42 recurrences correspond to only 27.8% of patients.
Among the dichotomized histopathologic variables examined, three were associated univariately with recurrence: tumor size, VN pathology classification, and necrosis (Table 1). 1) The disease recurrence rate of 50% for patients with a VN tumor size of 2 or 3 (> 15 mm) was significantly higher (P = 0.002) compared with the recurrence rate of 21.4% for patients with a VN tumor size of 1 (≤ 15 mm). Only one patient had a DCIS with a VN tumor size of 3; thus, the comparison essentially was between VN tumor sizes of 1 and 2. The mean size of VN 2 or 3 tumors was 26.0 mm, and the mean size of VN 1 tumors was 8.0 mm. 2) Higher recurrence rates were associated with higher VN pathology classification (P = 0.035). For patients with a VN pathology classification of 1, the recurrence rate was 17.8%. For patients with a VN pathology classification of 2, the recurrence rate was 27.9%. For patients with a VN pathology classification of 3, the recurrence rate was 37.8%. 3) The recurrence rate of 38.2% for patients with a necrosis score 2 or 3 was significantly higher (P = 0.006) compared with the recurrence rate of 17.3% for patients with necrosis scores of 0 or 1. None of the dichotomized biologic markers demonstrated a significant association with disease recurrence in the univariate analysis.
Time to disease recurrence
Analysis of the time to recurrence indicated significant differences for the following four histopathologic variables: VN tumor size (P < 0.0001), necrosis (P = 0.004), nuclear grade (P = 0.029 for Grade 1–2 vs. Grade 3), and comedonecrosis (P =; 0.025) (Table 2) (Fig. 1). For example, the time to recurrence was found to be significantly shorter for patients who had VN size 2–3 tumors (> 15 mm) compared with patients who had VN size 1 tumors (≤15 mm). The difference in the time to disease recurrence among the classifications of VN pathology was of borderline significance (P = 0.056). Among the dichotomized biologic markers, the difference in time to recurrence only approached significance for bcl-2 (P = 0.087). Among patients with bcl-2 levels < 100%, 25% developed recurrent disease within 26 months (95% confidence interval [95% CI], 19–48 months); among patients with bcl-2 levels = 100%, 25% developed recurrent disease by 66 months (95% CI, 43–97 months).
The time to recurrence in patients with ductal carcinoma in situ of the breast based on dichotomized morphologic variables (A–D) and a multivariate Cox model (E). Tumor size (P < 0.001) and the degree necrosis (P = 0.005) based on the Van Nuys (VN) Prognostic Index, were found to be predictors of time to disease recurrence in the multivariate Cox model, as illustrated in plot E, which shows the following Kaplan–Meier curves: 1) tumor size ≤ 15 mm, necrosis = VN 0 or 1; 2) tumor size ≤ 15 mm, necrosis = VN 2 or 3; 3) tumor size > 15 mm, necrosis = VN 0 or 1; and 4) tumor size > 15 mm, necrosis = VN 2 or 3. Comedo: comedonecrosis.
Correlations between biologic markers
Correlations between the biologic markers were found to be negative and significant between HER-2/neu and ER (Spearman correlation coefficient, − 0.37; P < 0.001) and between HER-2/neu and bcl-2 (Spearman correlation coefficient, − 0.32; P = 0.001). bcl-2 was found to be correlated positively with both ER (Spearman correlation coefficient, 0.44; P < 0.001) and PR (Spearman correlation coefficient, 0.35; P < 0.001). p21 did not demonstrate a significant correlation with any other biologic marker.
Correlations between histopathologic variables
Associations between histopathologic variables were very high and were significant (γ > 0.70; P < 0.0001) for any two of the following: VN pathology classification, nuclear grade, necrosis, and comedonecrosis. Therefore, we considered only multivariate models that included one of those four variables at a time.
Multivariate Results
Rate of recurrence
The best logistic regression model with histopathologic variables included VN tumor size and necrosis. There were 147 observations among 151 patients that were included in this model. 1) The odds of recurrence were 4.1 times higher (P = 0.001; 95% CI, 1.8–9.5) for patients who had VN size 2 or 3 tumors (> 15 mm) compared with patients who had VN size 1 tumors (≤ 15 mm). 2) The odds of recurrence were 3.3 times higher (P = 0.004; 95% CI, 1.5–7.2) for patients who had 2 or 3 degrees of necrosis compared with patients who had 0 or 1 degree of necrosis. No biologic marker was found to be significant, either alone or together with other markers, when added to this logistic regression model.
Time to recurrence
Two morphologic variables were found to be significant when considered in the multivariate Cox proportional hazards model: VN tumor size (P < 0.001) and degree of necrosis (P = 0.005) (Fig. 1). Similar to the results noted with the best logistic regression model, no one biologic marker demonstrated a significant association with disease recurrence when added to this proportional hazards model.
Results for patients with at least 5 years of follow-up
Because DCIS may not recur for prolonged intervals after initial treatment, we identified 116 patients with at least 5 years of follow-up and independently analyzed the data on these patients. The results essentially were identical to the results from the larger group of patients. Three histopathologic parameters were associated univariately with recurrence, namely, VN tumor size (P = 0.007), VN pathology classification (P = 0.048), and degree of necrosis (P = 0.012). The best logistic regression model also included tumor size and degree of necrosis in this smaller group. Similar to the larger group, no biologic marker demonstrated a significant association with recurrence of disease.
CART and Tree-Structured Survival Analysis
Both the CART analysis and the Tree-Structured Survival Analysis yielded consistent results. The model obtained with Tree-Structured Survival Analysis is presented (Fig. 2) because it provided greater sensitivity (83.8% vs. 66.7%) and specificity (77.9% vs. 72.5%) compared with the CART model. Tumor size (VN size 1 tumors vs. VN size 2 or 3 tumors) separated the study group into two distinct populations, and the degree of necrosis (0 or 1 vs. 2 or 3) and nuclear grade (Grade 1 or Grade 2 vs. Grade 3) subdivided these major groups into four separate groups. When the effect of biologic markers was evaluated for each of these groups, positive PR status (PR > 3.5%) was found to contribute strongly to recurrence of disease in the subgroup with small tumor size (1–15 mm) and 2 or 3 degrees of necrosis. Approximately 56% of patients in this subgroup developed recurrent disease at a median of 84 months, compared with an 8% recurrence rate at a median of 39 months in the subgroup with PR < 3.5%. Furthermore, the model suggests that, in the subgroup with higher PR values, bcl-2 expression in < 97.5% of neoplastic epithelial cells may be associated with a higher recurrence rate (8 of 11 patients; 73%) compared with bcl-2 expression in > 97.5% of cells (7 of 16 patients; 44%).
Survival regression tree for 104 patients who had both bcl-2 and progesterone receptor (PR) levels measured. The percentages depicted in bold numbers highlight the major differences in recurrence rates. VN size: tumor size according to the Van Nuys pathologic classification system.
DISCUSSION
This study of 151 patients with DCIS who were treated with wide excision and observation alone represents 1 of what to our knowledge are very few large studies (i.e., > 100 patients)6, 32-34 in which a second event (namely, recurrence as either DCIS or invasive carcinoma) is unaffected by either radiation therapy, hormone manipulation, or chemotherapy. It has afforded us the opportunity to assess the prognostic impact of a large number of histopathologic and biologic markers, free of any potentially confounding treatment-related influences, in a series of patients who were treated by one surgeon at one institution. In this group of 151 patients, 27.8% had a second event with a median follow-up duration of 65 months, which is nearly 2 years longer than reported in the NSABP study by Fisher et al.,32 1 year longer than the study by Ottesen et al.,6 and 3 months longer than the study by Ringberg et al.33 (the follow-up duration is not mentioned in the series by Douglas-Jones et al.34). The NSABP study reported by Fisher et al.32 included 391 patients with DCIS who underwent wide excision alone. The recurrence rate in that study was 16%, and the median follow-up was 43 months. The study by Ottesen et al.6 on 112 patients with DCIS who underwent wide excision showed a recurrence rate of 22% and a median follow-up of 53 months. The Swedish study reported by Ringberg et al.33 included 121 patients with DCIS who underwent breast-conserving surgery and received no postoperative irradiation. In that study, at a median follow-up time of 62 months, there were 31 local recurrences (26%). The series by Douglas-Jones et al. from the University of Wales34 included only women ages 50 years and 65 years with DCIS that was detected by screening mammography between 1989 and 1998. Of 115 women who underwent wide local excision alone, 15 women (13%) experienced recurrent disease.
In the current study, tumor size, degree of necrosis, and Van Nuys pathologic classification all demonstrated significant associations with recurrence of DCIS by univariate analysis. These results are in keeping with findings by the NSABP,32 the Van Nuys group of investigators,8 Ottesen et al.,6 and others; and they validate the use of conventional histopathologic parameters to predict recurrence. Boland et al. recently reported on 237 patients who underwent breast-conserving surgery (some patients also received tamoxifen and radiation therapy) and were followed for a median of 47 months.35 Those authors found that margin width and tumor grade were independent risk factors for local recurrence, with the former carrying three times more power than the latter in predicting recurrence. Margin width was not measured specifically in our mostly retrospective study, because it was not an established risk factor for recurrence until 1999,9 several years after our study began. When the time to disease recurrence was examined, the results paralleled the incidence of recurrence for the conventional prognostic histopathologic markers (i.e., patients with poor prognostic histopathologic markers developed recurrent disease significantly earlier compared with patients who did not have such markers).
In the current study, we found no significant association, by univariate or multivariate analysis, between any of the biologic markers evaluated and recurrence of disease. This result may be related in part to the distribution of some markers that ordinarily would be considered unfavorable in patients with invasive breast carcinoma. For example, < 25% of the patients in the current series had negative ER profiles, and < 33% of patients showed strong expression of the HER-2/neu protein. We are aware of only two other studies that examined the prognostic value of biologic markers in DCIS. In the study reported by Ringberg et al.33, 3 biologic markers were associated significantly (or almost significantly) with disease recurrence in a univariate analysis: high Ki-67 levels (defined as > 10%; P = 0.048), p53 levels (P = 0.052), and bcl-2 levels (P = 0.061). However, when they were included in a model that contained tumor grade and growth patterns, none of the biologic markers reached statistical significance in the multivariate analysis. In another series that described 49 patients who were treated with either mastectomy or conservative surgery with or without radiotherapy, Perin et al. found no significant association between a wide variety of biologic markers, including ER, PR, HER-2/neu, and p53, and the rate of disease recurrence.36
One interesting finding was a tendency toward a reduced time to recurrence, despite an absence of effect on the rate of recurrence, among patients with disease that showed < 100% expression of bcl-2. bcl-2 protein expression has been correlated, in invasive and in intraductal breast carcinoma, with well differentiated histology and positive ER status,20, 21 two acknowledged beneficial prognostic factors in invasive breast carcinoma. A similar correlation was found in the current study.
Because we found no significant correlation between biologic markers and recurrence of DCIS by using traditional logistic analysis approaches, we attempted an alternate approach using CART analysis and Tree-Structured Survival Analysis (Fig. 2). High bcl-2 levels (> 97.5%) and low PR levels (< 3.5%) were selected in the model that differentiated patients with recurrent disease from patients without recurrent disease. The beneficial effect of 100% positive bcl-2 status may have been anticipated by the results obtained by Ringberg et al.33 and the association of bcl-2 expression with prognostically favorable histopathologic features. However, the adverse effect of positive PR status appears contrary to the collected literature regarding hormone receptor status and breast carcinoma behavior. Currently, whether the association observed in this subgroup of patients with small DCIS lesions and higher grades of necrosis is significant or fortuitous remains unclear.
The current study disclosed a multitude of correlations between and among the biologic markers evaluated. In general, higher levels of hormone receptors, typically associated with more favorable clinical features in invasive breast carcinoma, showed a significant, direct correlation with the expression of bcl-2 (a marker of low-grade breast neoplasms) and an inverse correlation with p53, HER-2/neu, and Ki-67, all markers of more aggressive disease. These are very similar to findings of other authors, as reviewed by Ravdin in Silverstein's text on DCIS.37 It has been suggested that down-regulation of bcl-2 by mutant p53 accounts for their reciprocal correlation38 and that hormonal regulation of bcl-2 expression explains the positive correlation between ER and bcl-2.14
p21 expression demonstrated no significant correlation with any of the other biologic markers in the current study. p21 expression in invasive breast carcinoma is extremely variable and has been correlated directly,39, 40 inversely,41, 42 and not at all43, 44 with survival. To our knowledge, no information exists regarding its prognostic value in DCIS. The wide variety of stimuli that affect p21 expression through both p53-mediated and non-p53 pathways may account for the disparate observations that have been reported and the lack of a unifying concept regarding its significance.
In a retrospective study such as ours, we acknowledge that immunohistochemical results may be affected by certain technical factors that are beyond our control. These include the effect of prolonged periods of time on paraffin embedded tissue and the possibility of discrepant results related to the use of Techmate (capillary gap) slides in some tumors and non-Techmate slides in others. The experience in our immunohistochemistry laboratory has been that Techmate and non-Techmate slides generally yield the same immunohistochemical results.
The results of the current study confirm the value of conventional histopathologic factors, and especially the Van Nuys pathologic classification, in predicting local recurrence of DCIS. In the group of patients studied, no significant correlation was found between disease recurrence and expression of ER, PR, p53, HER-2/neu, Ki-67, p21, or bcl-2.
