Pathologic findings from the National Surgical Adjuvant Breast Project (NSABP) eight-year update of Protocol B-17

Intraductal carcinoma


  • Edwin R. Fisher M.D.,

    Corresponding author
    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Pathology Center, Pittsburgh, Pennsylvania
    • Allegheny University, Allegheny General Hospital, 5th Floor, Continuing Care Center, 320 East North Avenue, Pittsburgh, PA 15212
    Search for more papers by this author
  • James Dignam Ph.D.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Statistical Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • Elizabeth Tan-Chiu M.D.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Operations Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • Joseph Costantino Dr.P.H.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Statistical Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • Bernard Fisher M.D.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Operations Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • Soonmyung Paik M.D.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Pathology Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • Norman Wolmark M.D.,

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Operations Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
  • for the National Surgical Adjuvant Breast Bowel Project (NSABP) Collaborating Investigators

    1. National Surgical Adjuvant Breast and Bowel Project (NSABP) Pathology Center, Pittsburgh, Pennsylvania
    Search for more papers by this author
    • The following are the institutions (and principal pathologists) who contributed 10 or more patients to the study: Baptist Medical Center, Oklahoma City, OK (S. Shrago); Boston University, Boston, MA (D. Faller); British Columbia Cancer Agency, Vancouver, British Columbia, Canada (A. J. Worth); Community Clinical Oncology Program (CCOP), Allegheny General Hospital, Pittsburgh, PA (R. J. Hartsock); Alton Ochsner Medical Foundation, New Orleans, LA (G. Farr); Columbia River Oncology Program, Portland, OR (P. W. Kohnen); CCOP, Marshfield Clinic, Marshfield, WI (C. Reyes); City of Hope Medical Center, Duarte, CA (H. Battifora); Good Samaritan Hospital,, Cincinnati, OH (T. Wessler); Hartford Hospital, Hartford, CT (A. Ricci, Jr.); Hotel-Dieu, Montreal, Quebec, Canada (Y. Boivin); Jewish General Hospital, Montreal, Quebec, Canada (L. Begin, M. Brisson); Kaiser Permanente, Portland, OR (N. Olson); Medical College of Wisconsin, Milwaukee, WI (D. Rothwell); Michigan State University, East Lansing, MI (H. Bowman, R. Edminister); Mt. Sinai Medical Center, Cleveland, OH (R. Lash, G. Mendelsohn); Pennsylvania Hospital, Philadelphia, PA (M. Cunnane); St. Michael's Hospital, Toronto, Ontario, Canada (A. Chalvardjian); Tufts University/New England Medical Center, Boston, MA (H. Safaii); University of Hawaii, Honolulu, HI (D. Hono); University of Michigan, Ann Arbor, MI (H. A. Oberman); University of Pittsburgh, Pittsburgh, PA (S. Yousem).



This report is an 8-year update of the authors' previous findings from National Surgical Adjuvant Breast Project (NSABP) Protocol B-17, which relates to the influence of pathologic characteristics on the natural history and treatment of intraductal carcinoma (DCIS).


Nine pathologic features observed in a pathologic subset of 623 of 814 evaluable women enrolled in this randomized clinical trial were assessed for their role in the prediction of second ipsilateral breast tumors (IBT), other events, and selection of breast irradiation (XRT) following lumpectomy.


The frequency of subsequent IBT was reduced from 31% to 13% (P = 0.0001) by XRT. The average annual hazard rates for IBT were reduced by XRT for all pathologic features examined. Four characteristics were individually noted to be significantly related to IBT, but only moderate-to-marked and absent-to-slight comedo necrosis were found to be independent high and low risk predictors, respectively, for such an event in patients of both treatment groups. XRT effected a 7% absolute reduction at 8 years in the low risk group. Despite a relatively high incidence (≈40%) of IBT consisting of invasive cancer, mortality due to breast carcinoma after DCIS for the entire cohort was found to be only 1.6% at 8 years.


The degree of comedo necrosis in patients with DCIS appears to be sufficient for discriminating between high and low risks for IBT following lumpectomy for DCIS. Although margin status, unlike in our previous report, was found to have only a slight or borderline influence on the frequency of IBT at 8 years, excision of DCIS with free margins is advised. The low risk group exhibits a statistically significant reduction of IBT from XRT. The decision to forgo XRT in the treatment of this singular subset of patients would appear to depend on clinical considerations and the input of informed patients rather than being standard practice. [See editorial on pages 375–7, this issue.] Cancer 1999;86:429–38. © 1999 American Cancer Society.

Increased detection of ductal carcinoma in situ (DCIS) continues to provoke a spate of reports concerning its pathologic features, natural history, and treatment. In 1993, the initial 5-year clinical findings from what we believe was the only prospectively randomized clinical trial addressing some of these issues, Protocol B-17 of the National Surgical Adjuvant Breast Project (NSABP), disclosed a significant reduction in second ipsilateral breast tumors (IBT) in patients who received local breast irradiation (XRT) following lumpectomy, as opposed to those who were treated with only the latter (16.4% vs. 7.0%).1 Parenthetically, it should be noted that although “lumpectomy” is the term used for the surgical procedure, in approximately 80% of the cohort no lump or nodule was appreciated either clinically or pathologically. The beneficial effect of XRT in reducing IBT has been observed to persist into the eighth year of follow-up.2 It was not influenced by any mammographic characteristics of the lesions that were examined.

Results of our attempt to identify pathologic discriminants predictive of IBT at 4-1/2 years in a pathologic subset of 573 of the 790 women evaluated in the clinical report appeared in 1995.3 Several of the nine pathologic features individually analyzed for their relation to IBT were found to be significant in this regard. However, when jointly evaluated, only the presence of moderate-to-marked comedo necrosis and uncertain or involved margins of resection were found to be independent high risk factors for such an event. The highest hazard rates occurred in both treatment arms when both high risk parameters were present, and the lowest occurred when margins were free and comedo necrosis was absent or slight. The difference in average annual hazard rates between the two treatment arms for absent-to-slight comedo necrosis and free margins was not great (1.97 vs. 1.18 per 100 patients). Although the difference in the relative frequency of IBT was 40%, the absolute difference was only 3%. This suggested that patients whose tumors exhibited such features might not require XRT following lumpectomy. However, the number of events were too few to warrant such a proposal at that time, a view subsequently expressed by others as well.4

The current report is an update describing 8 years of our pathologic observations of the B-17 cohort of patients with DCIS. The accumulation of a greater number of events during the past several years prompted us to perform these analyses.


Details regarding patient selection, eligibility, study design, randomization, stratification, follow-up procedures, and identification of endpoints for NSABP Protocol B-17 have been presented elsewhere.1, 2 Patients were randomly assigned to receive either XRT or no further treatment after lumpectomy for “pure” DCIS. Although initially all women underwent axillary lymph node dissection, this procedure became optional and was rarely performed after 2 years from the time the protocol was activated because of the absence of lymph node metastases as well as the historic rarity of such an event among patients with DCIS. XRT consisted of 50 Gray (Gy) at a rate of 10 Gy per week for 5 weeks. A boost to the tumor bed with either external or interstitial technique was not used, nor was lymph node irradiation administered.

The recent clinical study of follow-up for 8 years for B-17 recorded in 19982 was based on the evaluation of 814 patients instead of the 790 evaluated previously.1 This increase resulted from the addition of 24 patients from one institution who had been excluded because of concerns about the quality of their data. However, subsequent audits failed to confirm this suspicion. A central pathology review excluded 2 of the 24 patients from this study. The remaining 22, as well as 28 who had been excluded from our initial pathologic analyses because slides representative of IBT had not been submitted for central review, were added to the original subset of 573 patients. Careful evaluation of the surgical pathology reports and other information that was subsequently supplied for these 28 cases indicated that IBT did occur. Except for some instances in which data was regarded and tabulated as unknown, all analyses were performed on 623 patients, or 77% of the entire B-17 cohort. Three hundred three were treated by lumpectomy only and 320 by lumpectomy and subsequent XRT. The findings in this report are based on an average time since entry into the study of 102 months.

The 9 tumor characteristics reviewed by NSABP central pathologists are found in Table 1. These have been described in detail previously.3 However, several characteristics that are not generally used by pathologists warrant brief explanation. In the context of this study, the designation “multifocal” indicates the presence of DCIS in sections prepared from more than one block of the specimen. Tumor in only one block of several was regarded as “unifocal.” If only one block was available for evaluation, focality was considered “unknown.” Comedo necrosis was assessed as an independent feature of DCIS rather than a representative feature of a specific histologic tumor type. When this alteration occurred in more than one-third of ducts exhibiting DCIS, it was subjectively graded as “moderate/marked”; fewer or none were graded as “absent/slight.”

Table 1. Average Annual Hazard Rates per 100 Patients and Rate Ratios of IBT According to Pathologic Characteristics and Treatment
CharacteristicLumpectomyLumpectomy + XRTRRa95% CIP value
  • IBT: ipsilateral breast tumor; XRT: radiotherapy; HR: hazard rate; RR: rate ratio; CI: confidence interval.

  • a

    Risk of IBT for patients with category of pathologic characteristics compared with reference (baseline 1.00) (e.g., relation of moderate/marked to absent/slight comedo necrosis adjusted for treatment).

  • b

    Macroscopic estimates.

  • c

    Based on 302 tumors microscopically <10 mm.

Comedo necrosis
Histologic type
Lymphoid infiltrate
Nuclear grade
Tumor size (mm)

Although the protocol required tumor margins to be free, i.e., not transected, central review revealed instances in which the tumor margin was violated. In some instances in which the margins were not inked, their status was regarded as uncertain because it could not be determined whether transected tumor represented “tumor” or “patient” margin. Because a preliminary investigation disclosed residual tumor in specimens from completion lumpectomies to be similar when margins were considered uncertain or involved and there were only a few examples of the latter, they were combined and designated as “uncertain/involved.”

Tumor size was recorded from information provided by institutional pathologists on their pathology reports. Approximately 80% were detected mammographically without any distinctly measurable lesion. In these instances, it was stated in the reports that blocks were obtained from areas indicated by a guide wire or stain. Some examples were macroscopically measurable, particularly when ≥1.0 cm. This size was considered an appropriate cutoff, and consequently tumors were described as either ≥1.0 cm or <1.0 cm. Lesions were also measured microscopically in 553 patients as described previously,7, 8 without prior knowledge of the macroscopic estimates. The microscopic size was represented by the greatest dimension of tumor encountered in any of the slides examined. This included intervening normal tissue or in situ cancer when several foci were encountered in the same section. This mimicked the method for macroscopic mensuration. There was 94% agreement between macroscopic and microscopic sizes for tumors <1.0 cm but only 19% for those ≥1.0 cm. Accordingly, microscopic sizes were used only to delineate tumors <10 mm and macroscopic estimates for those ≥10 mm. As a result, only 302 of the 553 microscopic measurements were utilized.

The site of IBT was obtained from pertinent clinical records and compared with that recorded for the index tumor. Because some lesions appeared to be “on the line” of two or more quadrants or in a subareolar location, it was considered most appropriate to regard IBTs found in the same quadrant or close to the site of the index cancer as being at the “same” site, as opposed to IBTs that were distinctly in a remote quadrant, which were designated as being at a “different” site.

Microscopic sections of material from IBT was available for central review in 90% of instances. The type of cancer, i.e., “pure” DCIS, invasive, or that comprised of both elements, was noted. The histologic type of DCIS when present in IBT was compared with that of the index lesion and designated as “same” or “different.”

Chi-square tests were utilized for comparisons of selected characteristics of the pathologic subset of 623 patients and the clinical cohort of 814. A log rank test9 was performed to compare the patterns of disease outcome in these two populations.

Frequency distributions for the nine pathologic features were compared by treatment groups using chi-square tests of association. Average annual IBT rates according to pathologic features were computed as ratios of the number of events to total person-years of observation.10 Rates were also estimated for each follow-up year through 8 years.

The Cox proportional hazards model was used to evaluate further the prognostic significance of each pathologic feature in relation to risk of IBT.11, 12 Tests of interactions of each factor with treatment were conducted to detect a differential magnitude of the treatment benefit for XRT according to pathologic features. Variables associated with IBT at α = 0.10 or less in univariate analyses were evaluated jointly to determine which were independently associated with the risk of IBT. From these models, IBT rates and the cumulative probability of IBT were computed for each treatment group. This provided an estimate of the effect of treatment within each risk group. The cumulative probability of IBT, taking into account competing risks, was determined using cumulative incidence functions.13

There have been a number of classifications of DCIS purported to define risk groups for IBT and to have a possible impact on the treatment of DCIS. Two popular schemes in particular were considered for analysis with our data. One was described in 199514 and the other the following year15 by Silverstein et al. Our material closely conformed to the three groups of DCIS described in their earlier publication, which represented a pathologic classification. We found 218 patients whose tumors qualified as their Group 1 because of “non-high grade” nuclei (our good nuclear grade) and no necrosis (our absent/slight comedo necrosis). One hundred three Group 2 cases also appeared to be characterized by “non-high grade” nuclei but with necrosis (our good nuclear grade and moderate/marked comedo necrosis). Three hundred two cases in Group 3 had “high grade nuclei”(our poor nuclear grade) with or without comedo necrosis, as defined by the authors.

The second and more recent classification, designated as the Van Nuys Prognostic Index (VNPI), is more complicated and was difficult to adapt to our material. Our patient population contained much fewer tumors measuring >15 mm. Further, the degree of freedom of the margins used in the VNPI was not required in NSABP Protocol B-17. Because of this, adaptation of our material to this scheme might have been regarded as arbitrary, and such analyses were not performed.


The frequency of such characteristics as age, race, tumor size, presence of lobular carcinoma in situ, method of detection (mammographic, clinical, or both), and disease free survival at 8 years was comparable in the pathologic subset (623) and the cohort of 814 evaluated for the recent clinical report of NSABP Protocol B-17 published in 1998.2 The distribution of all pathologic features except histologic tumor type was similar in the two treatment arms. The “pure” cribriform type was frequent and the “pure” solid form less so in the lumpectomy-and-XRT group than in the lumpectomy-only arm.

The cumulative frequency of IBT was 137 (22%) for all 623 patients. Ninety-four of 303 (31%) occurred in the lumpectomy-only group and 43 of 320 (13%) in those also receiving XRT. The average annual hazard rate per 100 patients for the former was 5.10 and the latter 1.97, resulting in a relative risk of 0.39 (95% CI, 0.27–0.56). This represented a 61% relative reduction in IBT for patients receiving XRT (log rank test, P < 0.0001). The time of occurrence of IBT ranged from 2 to 123 months (median time, 36 months) after surgery. Forty-nine percent and 17% of all IBT occurred within the first 4 years in the lumpectomy-only and lumpectomy-and-XRT groups, respectively. The frequency of IBT in the lumpectomy-and-XRT group was relatively constant throughout: 17% in the first 4 years and 12% in the later periods. Sixty-seven percent of all invasive and 72% of all noninvasive IBTs occurred in the first 4 years (Fig. 1).

Figure 1.

The biennial frequency of ipsilateral breast tumors (IBTs) is shown.

The average annual hazard rates for IBT were lower for all 9 pathologic characteristics in the lumpectomy-and-XRT group than in the lumpectomy-only group (Table 1). The relative risks for this event, shown in Table 1, were adjusted for treatment; they pertained to the pathologic characteristics and not the effects of treatment. Only moderate/marked comedo necrosis, solid tumor type, moderate/marked lymphoid infiltrate, and multifocality were found to be significant predictors for IBT in both treatment groups. Uncertain/involved margins and poor nuclear grade were borderline (0.05 < P < 0.10). The absolute reduction in IBT due to XRT for patients with free margins was 16% and 22% when the margins were uncertain/involved. Only comedo necrosis exhibited a suggestion of a differential treatment effect or interaction; the degree of benefit from XRT was less when comedo necrosis was absent/slight than when it was moderate/marked. The relative reduction in IBT risk resulting from XRT when comedo necrosis was absent/slight was 48%, compared with 70% when a moderate/marked degree was present (Table 2). The differences in treatment groups at 8 years were 7% for the low risk group and 27% for the high risk group. Thus, absent/slight comedo necrosis might be regarded as a low risk predictor for IBT and moderate/marked comedo necrosis as the converse.

Table 2. Average Annual Hazard Rates per 100 Patients, Rate Ratios, and Absolute Reductions of IBT According to Comedo Necrosis and Margin Status
FeatureLumpectomy onlyLympectomy + XRTRRaARe
No. (%)HRNo. (%)HR
  • IBT: ipsilateral breast tumor; XRT: radiotherapy; HR: hazard rate; RR: rate ratio; ARe: absolute reduction.

  • a

    Treatment comparison among patients with the degree of pathologic features indicated (e.g., 0.52 represents relative risk for lumpectomy + XRT/lumpectomy among patients with absent/slight comedo necrosis only).

Comedo necrosis
 Absent/slight16438 (23%)3.517823 (13%)1.80.5210
 Moderate/marked13956 (40%)7.514220 (14%)2.10.3026
Status margins
 Free24973 (29%)4.726734 (13%)1.90.4016
 Uncertain/involved5421 (39%)7.2539 (17%)2.50.3622

Many of these pathologic features are interrelated. Comedo necrosis was associated with solid tumor type, multifocality, and nuclear grade. This latter was also interrelated with solid tumor type. Multifocality was more frequent among patients with positive margins.

When all nine pathologic features were examined jointly for prognostic significance, only comedo necrosis remained as a significant predictor for IBT. When other individual characteristics that exhibited an increased risk for IBT (α = 0.10 or less) were assessed jointly, tumor type and multifocality were found to be only marginally associated with IBT (0.05 < P < 0.10).

Including degrees of comedo necrosis and status of margins into a model with treatment failed to reveal any significant modification of the effect of the latter (interaction P = 0.42). After cross-classifying categories of two features and treatment, it was observed that patients subjected to lumpectomy whose tumors exhibited two favorable features, such as absent/slight comedo necrosis and free margins, had the lowest rates of IBT (3.0 per 100 patients per year), whereas those with both unfavorable characteristics had the highest (8.2 per 100 patients per year). Yet those with moderate/marked comedo necrosis and free margins exhibited the same incidence of IBT as the group with both unfavorable features. The incidence was intermediate when margins were uncertain/involved and comedo necrosis absent/slight. A benefit due to XRT was observed in each risk group. Similar results were obtained when either multifocality or tumor type were cross-classified with comedo necrosis. The combination of multifocality, tumor type, and status of margins each with comedo necrosis revealed only a marginally greater risk for each pair than comedo necrosis alone (0.05 < P < 0.10).

Classifying our data according to the 1995 Van Nuys pathologic classification14 revealed that both of their Groups 2 and 3 represented high risk categories for IBT in patients treated with lumpectomy but not lumpectomy and XRT. All three groups had a significant benefit from the latter (Table 3).

Table 3. Analysis of NSABP B-17 Pathologic Data According to the 1995 Van Nuys Pathologic Classification
GroupLumpectomyLumpectomy + XRTRRb95% CI
  • XRT: radiotherapy; IBT: ipsilateral breast tumor; HR: hazard ratio; RR: risk ratio; CI: confidence interval.

  • a

    Average annual hazard rate/100 patients.

  • b

    Rate ratio (lumpectomy + XRT/lumpectomy) within the risk group.

1. NG* good+ absent/slight necrosis10525243.511312111.50.420.21–0.83
2. NG good+ moderate/marked necrosis5621386.4477152.20.350.15–0.81
3. NG poor+ any degree of necrosis14248346.116024152.30.380.24–0.63

The frequency of first events is presented in Table 4. The most frequent of these, other than IBT, was the occurrence of contralateral breast carcinoma (4.3%) and second primary cancers (3.2%). These second primary cancers presented at a variety of sites, including the colon, pancreas, lung, endometrium, uterine cervix, and thyroid. There was one instance of lymphoma. There were 36 deaths among patients of this cohort, and they were equally distributed between the two treatment arms. Ten of these were attributable to breast carcinoma, four in the lumpectomy-only treatment arm and six in the lumpectomy-and-XRT arm. Only two in each group had prior IBT. There were no statistically significant differences in the incidences of these events other than IBT in the two treatment arms.

Table 4. Frequency of Events after DCIS According to Treatment
LumpectomyLumpectomy + XRTAll patients
  • DCIS: ductal carcinoma in situ; XRT: radiotherapy; IBT: ipsilateral breast tumor.

  • a

    Two had prior IBTs.

First events
 Other logoregional10.330.940.6
 Contralateral breast carcinoma103.0175.0274.3
 Second primary cancer103.0103.0203.2
 Other causes142.0122.0264.2
 Breast carcinoma4a1.06a2.0101.6

Eighty-four (61%) of the 137 IBTs were “pure” DCIS (Table 5). The remainder were either a combination of the latter and invasive cancer (26%) or “pure” invasive cancer (13%). The presence of invasive IBT was not specifically associated with any pathologic feature of the index DCIS. The relative and absolute reductions for the IBT comprised of invasive cancer after XRT were not significantly different from the reductions observed for all IBTs according to degree of comedo necrosis or margin status noted in Table 2. The percentages of types of IBT were not influenced by treatment. Seventy-six percent of all IBTs were found to be at the same site as the index DCIS. Although this difference was not statistically significant, the invasive form of IBT occurred at different sites more frequently than the noninvasive type. The DCIS component in 90% of IBTs was morphologically similar to that of the index lesion. The average size of IBT containing invasive cancer was 1.3 cm (range, 0.3–4.5 cm). Fifty percent were ≤1.0 cm.

Table 5. Frequency of Types of IBT According to Treatment
TypeLumpectomyLumpectomy + XRTAll patients
  • IBT: ipsilateral breast tumor; XRT: radiotherapy; DCIS: ductal carcinoma in situ.

  • a

    Combined to represent invasive cancer.

Pure DCIS576127638461
DCIS+ invasive26 } 37a28 } 409  } 1621 } 3735 } 5326 } 39
Pure invasive11127 161813
 94  43 137

None of the 24 instances of DCIS that were independently associated with LCIS or noted to arise in fibroadenoma (2) or that presented as intracystic DCIS (6) have as yet been followed by an IBT.


The increase in the cumulative incidence of IBT in both treatment arms at 8 years, shown in Figures 2 and 3, closely mimicked that observed in our recent clinical report. We did not observe any increase in these events, particularly in the lumpectomy-and-XRT group, over later years of follow-up. These observations fail to support the notion that XRT might only delay the occurrence of IBT following lumpectomy.16

Figure 2.

The cumulative incidence of ipsilateral breast tumors (IBTs) is shown according to comedo necrosis and treatment.

Figure 3.

The cumulative incidence of ipsilateral breast tumors (IBTs) is shown according to margin status and treatment.

A recent review concerning the pathology of DCIS emphasized that tumor size, status of margins, and histologic tumor type represent the most important characteristics related to what the authors refer to as the “control of DCIS” after conservation therapy.17 However, histologic type is rarely included in such evaluations, and then generally refers only to comedo and noncomedo types of DCIS. Most investigators have thus far regarded comedo necrosis as an independent pathologic variable rather than a specific histologic type, a view expressed by us more than a decade previously.4 The authors of the review also indicate that most classifications of DCIS, of which we have found at least eight,5, 6, 17–23 have utilized nuclear differentiation, presence or absence of necrosis (apparently delineating between a comedo type and another ill-defined form simply referred to as “necrosis”), and growth pattern. In any event, nosologically, these characteristics are presented individually or in groups of varying numbers. Tumor size and status of margins appear in only a few,6, 19 and in those they are also incorporated with other features, resulting in a complex categorization. One group of investigators expanded their classification within l year.5, 6 This brief overview reflects the lack of unanimity and consternation attendant with efforts to classify DCIS. This difficulty is further compounded by the sparsity of studies revealing their biologic significance vis-à-vis the actual development of IBT as well as the natural history and treatment of DCIS. Some of those that have been correlated with IBT reveal contradictions.6, 19 It is not surprising that a recent panel concerned with these issues failed to endorse any particular algorithm that might be predictive for IBT.24

Classifying our material according to the criteria of the Van Nuys pathologic method described in 19955 failed to reveal complete agreement regarding predictive value for IBT or response to XRT in the 3 groups they evaluated. When their method of stratification was applied to our material, it revealed only two, rather than three, predictive groups for IBT. Furthermore, XRT was found to be of benefit to all patient subgroups (rather than just one, as noted by those investigators14). Our findings also indicate that estimation of comedo necrosis alone might have provided similar results in their analyses. This is in accord with the association between comedo necrosis and nuclear grade observed in this study. Yet only the former was found to be an independent predictor for IBT. The basis of this dichotomy is unclear. However, the sample sizes utilized in the Van Nuys presentation were relatively small, and some possibly important statistical analyses were either not available for review or not performed.

Recent efforts to define the extent of free margins necessary to minimize or prevent the development of IBT without the administration of XRT might appear to render moot both the simple and the complex pathologic classifications purported for this purpose. However, there is currently no unanimity regarding the best way to demonstrate margin status, let alone regarding how free it should be to accomplish this.25 Unfortunately, most information regarding an optimum extent of excision is derived from estimates of residual disease in mastectomy specimens or in the tissue of completion lumpectomies following removal of DCIS. These have revealed residual disease ranging from 19% to 43% for free margins <1 mm,26, 27 and an optimum width for possibly avoiding IBT has been variably suggested to be 5 mm28 or 10 mm.29–31 There were 243 two- and three-step procedures performed in this study that allowed for estimation of residual disease. The presence of the latter was found in 25% of instances in which the margins were free, and there was no quid pro quo relation between the detection of residual disease in anatomic specimens and IBT, the latter being much less than expected not only for DCIS but also for invasive cancer.32 Further, IBT was not found to be related to the various treatments administered to our cohort. It remains to be demonstrated whether a 10 mm margin is sufficient or necessary for the omission of XRT as prophylaxis against IBT.

Our own material offers no information concerning an optimum width, because this protocol, like other NSABP protocols that utilize lumpectomy, requires only enough tissue to ensure complete removal with a satisfactory cosmetic result. Pathologically, margins are regarded as free when the tumor is not transected, and no quantitative width is prescribed. Yet, despite claims that the width of margins utilized by the NSABP may be inadequate,27, 33 the annual rate of incidence of IBT (≈4%) in our cohort of women who did not receive XRT was almost identical to the rates reported by others for nonrandomized studies in which follow-up was even shorter, as was reported by Marks and Prosnitz.4 Also, the rate of overall incidence of IBT following lumpectomy and XRT was also similar to the rates in 2 other studies, each of which contained more than 100 cases and a 7-year follow-up period (13% vs. 9%33 and 10%34).

Even more important in this regard, and unlike our earlier pathologic analyses of protocol B-17,3 uncertain/involved or free margins were not independent high or low risk predictors of IBT at the eighth year of follow-up; they were only borderline for this latter period. The biologic implication of this inconsistency is unclear, but it may be related to the predominantly early appearance of IBT in our cohort. In our experience, changes in the significance of prognostic discriminants occasionally occur, e.g., tumor necrosis represents a prognostic indicator for invasive cancer at 5 years but not at 10 years.35 Although margin status was found to be borderline as a predictor for IBT, it does exhibit some additional but slight effect in this regard when examined together with comedo necrosis. Despite this limited effect at 8 years, its early importance signifies the need for histologically free margins in the treatment of DCIS patients with lumpectomy.

Tumor size is another important factor that might be expected to be related to the extent of tumor excision and IBT. Indeed, one might consider one of the important attributes of mammography to be its ability to detect truly small (<10 mm) tumors, allowing for wide excision without compromising cosmesis. Yet we continue to fail to discern any significant or consistent relation between tumor size and IBT, whether assessed macroscopically or microscopically with cutoff points of 10 mm or even 5 mm. Furthermore, tumors of all sizes according to our groupings exhibit a beneficial effect of XRT. This suggests that even very small foci of DCIS may represent a marker of risk for subsequent IBT, in keeping with the recognized multifocality of DCIS (vide infra), as we have suggested previously.7, 8 The appearance of the majority of IBTs at or close to the site of the index cancer, as well as the discovery that the histologic identity of the DCIS component of the IBT (when present) is the same as that of the index cancer in 90% of instances, also confirms our view that most IBTs following DCIS—as well as some invasive cancers—represent residual disease or incomplete removal of the index lesion.3, 7, 8 In this light, IBT is not necessarily related to any surgical inadequacy per se, but is instead more closely related to the multifocal nature of DCIS (multiple foci of tumor within the same quadrant or close to the index lesion32), which cannot be perceived at the time of surgery.

One of the conundrums related to DCIS is the origin and biologic nature of the invasive cancer, which we have encountered in ≈40% of IBTs. Such invasive cancers, found alone or with DCIS in some IBTs, are conventionally considered to arise from preexisting DCIS. However, there is no unequivocal evidence to support such a contention. Our recognition of a trend revealing a higher incidence of invasive IBT at sites different from the index lesion strongly suggests that at least some may arise de novo or not necessarily from preexisting DCIS. Despite the relatively high incidence of invasive cancer in IBT following DCIS, as well as in 11 of 21 (52%) of the contralateral breast carcinomas associated with ipsilateral DCIS, we continue to find the rate of death related to breast carcinoma to be exceedingly low; overall, it is only 1.6% at 8 years. This event showed no predilection for either treatment group, and it was preceded by an IBT in one-half of the deaths in the lumpectomy-only group and one-third of the deaths in the lumpectomy-and-XRT group. One apparent reason for this dichotomy in mortality may be the very good prognosis associated with the small size of the recurrent invasive cancers detected in our material. Our analyses continue to reveal an overall benefit for XRT in reducing IBTs for all patients. In the low risk group, the 59% relative risk reduction provided a 7% absolute difference in IBT at 8 years. This estimate is more than twice that noted in our previous pathologic report concerning IBT at 4-1/2; years.3 The decision to forgo XRT in the treatment of DCIS patients would now appear to depend on clinical considerations, such as patient age, health, and informed patient input, rather than standard practice.

It is surprising that despite the long-standing recognition and importance of comedo necrosis in DCIS, there is little or no consistent information concerning its pathogenesis. It has been found to be associated with cells exhibiting either high or low proliferative rates.36 Its relation to the identification of such biologic markers as c-erb B-2, p 53, and hormone receptors is unclear.23, 37 Although some32 have observed an increased frequency of vascularity in its vicinity, suggesting a relation to angiogenesis factor,38 others have not confirmed this observation.39 One might hypothesize that if comedo necrosis were related to a vascular phenomenon, the contrary or vascular obliteration would be a more likely cause. Recognition of “regressive” changes associated with comedo necrosis3, 7–9 resulting in scars is in keeping with such a suggestion. Yet the association of such a vascular change with comedo necrosis has apparently not been noted.

Finally, as emphasized in our most recent clinical report,2 there is no information provided in these analyses to support the need for mastectomy among patients with DCIS, at least among those who exhibit the characteristics of the women who comprised this cohort.