CANCER 33 CYTOPATHOLOGY Cytologic Features of Proliferative Breast Disease A Study Designed to Minimize Sampling Error Andra R. Frost, M.D.1 Sana O. Tabbara, M.D.2 Linda A. Poprocky, C.T.2 Heidi Weiss, Ph.D.1 Mary K. Sidawy, M.D.2 1 University of Alabama at Birmingham, Birmingham, Alabama. 2 The George Washington University, Washington, DC. BACKGROUND. Assessment of cytologic features that allow accurate classification of proliferative breast disease has been hampered by sampling errors when fineneedle aspirations have been compared with their corresponding histologic sections. METHODS. To allow for optimal cytohistologic correlation, 2 smears (1 hematoxylin and eosin–stained and 1 Diff-Quik-stained) were prepared from each of 98 breast biopsies without mass lesions and compared with the corresponding histologic sections of the scraped area. Each smear was reviewed in a blinded fashion and assessed for cellularity, background elements, cytoarchitectural features of cell groups, and nuclear features by 2 reviewers. Smears were then classified as nonproliferative breast disease (NPBD), proliferative breast disease without atypia (PBD) or with atypia (PBDA), or DCIS, based on review of the corresponding histologic sections. RESULTS. When comparing NPBD/PBD (n ⫽ 86) with PBDA/DCIS (n ⫽ 12), smears from PBDA/DCIS were significantly (by the Fisher exact test or Wilcoxon rank sum P values with adjustment for multiple comparisons) more likely to be cellular; contain single cells and necrosis; exhibit nuclear overlap and cytoplasmic vacuoles; have large nuclei, macronucleoli, pleomorphism, clumped chromatin, and hyperchromasia; and were less likely to have complex cell groups, monolayers, swirling, cohesion, and myoepithelial cells in epithelial sheets and the smear background. When NPBD (n ⫽ 53) and PBD (n ⫽ 33) were similarly compared, smears from PBD were more likely to exhibit larger and more complex cell groups, but they were otherwise similar to smears from NPBD. CONCLUSIONS. There are many cytologic features that will allow a distinction of NPBD/PBD from PBDA/DCIS, but relatively few that can aid in separating NPBD from PBD. Cancer (Cancer Cytopathol) 2000;90:33– 40. © 2000 American Cancer Society. KEYWORDS: cytology, breast, nonproliferative breast disease, proliferative breast disease, ductal carcinoma in situ. Abstract presented in part at the 87th Annual Meeting of the United States and Canadian Academy of Pathology, Boston, Massachusetts, March 1998. Address for reprints: Andra R. Frost, M.D., Department of Pathology, University of Alabama at Birmingham, 619 South 19th Street, KB 506 Birmingham, Alabama 35233. Received April 19, 1999; revision received July 28, 1999; accepted August 6, 1999. © 2000 American Cancer Society F ine-needle aspiration (FNA) of the breast has proven to be a reliable method of identifying breast carcinomas prior to therapy, and the diagnostic cytologic features of breast carcinoma are well established.1– 4 It is also recognized that the cytologic features of invasive and in situ carcinomas, especially high grade intraductal carcinomas, are often similar and cannot be reliably distinguished by FNA.3,5 In the cytologic assessment of proliferative breast disease, most authors claim good success in distinguishing benign from malignant intraductal proliferations, with most identifying more than 90% of aspirates of ductal carcinoma in situ (DCIS) as either atypical, probable carcinoma or carcinoma.6 –9 However, a definitive cytologic 34 CANCER (CANCER CYTOPATHOLOGY) February 25, 2000 / Volume 90 / Number 1 diagnosis of carcinoma may be difficult in some cases of DCIS, particularly low grade DCIS. Cytologic criteria for identifying and classifying benign proliferative breast lesions, including ductal hyperplasia with and without atypia, and for distinguishing proliferative from nonproliferative breast lesions in FNA, have been proposed.7,8,10 –12 Some authors have also proposed a combination of cytologic and architectural features of epithelial cell groups.8,10,11,13 Masood et al.7 advocated a scoring system based on the evaluation of individual cytologic features to classify nonproliferative breast disease (NPBD), proliferative breast disease without atypia (PBD), proliferative breast disease with atypia (PBDA), and DCIS. Most authors have reported some degree of success in identifying and correctly classifying proliferative breast lesions, but acknowledge that a significant cytologic overlap exists in the continuum of proliferative lesions. Underscoring the diagnostic difficulty created by this degree of cytologic overlap, a recent assessment of the applicability of the published criteria for identifying nonproliferative and proliferative breast lesions reported poor diagnostic accuracy and reproducibility when six cytopathologists with expertise in breast cytopathology were asked to distinguish NPBD, PBD, PBDA, and DCIS.14 In an attempt to identify cytologic features that can reliably distinguish PBD from NPBD, an evaluation of 12 cytologic features in 51 benign breast aspirates indicated that only the frequency of a swirling pattern of epithelial cells was significantly different in PBD and NPBD.15 The development and assessment of cytologic criteria for classifying proliferative breast disease has been complicated by the heterogeneity of morphologic changes within a given lesion and the lack of exact histologic correlation related to sampling. This problem is difficult to overcome completely even when strict standards of sampling are applied. Sampling errors due to the small size of a proliferative lesion, or due to the fact that benign PBD may be present in the midst of nonproliferative fibrocystic change or carcinoma in situ, may have hindered attempts to identify criteria that reliably distinguish proliferative from nonproliferative breast lesions. In an effort to overcome the problem of sampling, we have assessed the cytologic features of proliferative and nonproliferative breast lesions in cytologic smears prepared from scraping surgically excised breast tissue. MATERIALS AND METHODS Specimen Collection Ninety-eight excisional breast biopsy specimens without macroscopic mass lesions were included. During intraoperative consultation, one alcohol-fixed, H & E–stained and one air-dried, Diff-Quik–stained direct smear were prepared by scraping a portion of the specimen consisting macroscopically of fibrous breast parenchyma. The hematoxylin and eosin (H & E) stain was chosen because this is the stain routinely used at the George Washington University Hospital for intraoperative evaluation of alcohol-fixed smears. Although the Papanicolaou stain is more commonly used in cytology laboratories for alcohol-fixed specimens, some laboratories chose to use the H & E stain for FNA specimens. Both stains allow good evaluation of nuclear detail in alcohol-fixed smears. The area scraped was separated and placed in a designated cassette. Subsequently, an H & E–stained histologic section was prepared from the scraped area, providing optimal cytohistologic correlation. Slide Review The air-dried smear, alcohol-fixed smear, and histologic section for each case were coded, separated from each other, and reviewed in a masked fashion. Each slide was reviewed by two cytopathologists (A.R.F., S.O.T., or M.K.S.). The smears were assessed and graded for the following features: Cellularity was scored as 1 for low cellularity (⬍10 epithelial groups/slide), 2 for moderate cellularity (10 –30 epithelial groups/slide), and 3 for high cellularity (⬎30 epithelial groups/slide). A dual population of epithelial cells, apocrine and nonapocrine, was evaluated as present or absent. Elements in the smear background, including a cystic or proteinaceous background, naked bipolar nuclei (myoepithelial cells), and necrosis, were evaluated as present or absent. The size of the epithelial groups was graded as 1 if predominantly small (less than half of a 10⫻ field or less than approximately 3000 cells), 2 if a nearly equal mixture of small and large (more than half of a 10⫻ field or more than approximately 3000 cells), and 3 if predominantly large. The 10⫻ objective used in this assessment had an ocular field number of FN 20. The number of single epithelial cells were graded as 1 if they constituted less than 20%, 2 if they constituted 20 –70%, and 3 if they constituted more than 70% of epithelial cells. The degree of cohesion of the epithelial cell groups was graded as 1 if loose, 2 if a mixture of tight and loose groups was present, and 3 if tight. The presence or absence of myoepithelial cells within the epithelial groups was assessed as present or absent. Myoepithelial cells were recognized as having spindled to ovoid nuclei that were smaller and darker that the epithelial cell nuclei. Cytology of Proliferative Breast Lesions/Frost et al. Various architectural features of the cell groups were evaluated. The overall degree of complexity of the cellular arrangement in the groups was assessed and scored as 1 if composed of simple sheets with little folding or branching; 2 if the most complex sheets exhibited a mild to moderate degree of folding, branching, lumen formation, or three-dimensionality; and 3 if the most complex sheets exhibited a marked degree of folding, branching, lumen formation, or three-dimensionality. Specific architectural patterns (i.e., monolayers, cell swirling/streaming, papillae, slitlike lumens, and cribriform sheets) were assessed as absent or present. The degree of overlap of nuclei was scored as 1 for no overlap, and 2 for mild, 3 for moderate, and 4 for marked overlap. The nuclear size was graded as 1 if approximately 2 times or less than the greatest dimension of an RBC, 2 if 3– 4 times the greatest dimension of an RBC, and 3 if 5 times or more than the greatest dimension of an RBC. Nuclear pleomorphism was graded as 1 if the nuclei were round or oval and uniform in size and 2 for mild, 3 for moderate, and 4 for marked pleomorphism. Nucleoli were graded as 1 if absent, 2 if micronucleoli were present, and 3 if macronucleoli (nucleolus one-fourth of nuclear greatest dimension or more) were present. The degrees of chromatin clumping and hyperchromasia were assessed only on alcohol-fixed smears and graded as 1 if absent, and 2 if mild or moderate, and 3 if marked in degree. Cytoplasmic vacuolization was assessed as present or absent. A diagnosis of unsatisfactory, NPBD, PBD, PBDA, or carcinoma was also made for each slide by the reviewer, using criteria published in the literature.7,8,10 No equivocal diagnoses (e.g., PBDA vs. carcinoma) were allowed. Histologic sections were also reviewed by two pathologists (A.R.F. and M.K.S.) and classified as NPBD, PBD (i.e., moderate and florid ductal hyperplasia), PBDA (atypical ductal hyperplasia), and DCIS, using criteria presented by Page et al.16,17 DCIS was classified as high, intermediate, or low grade using the criteria presented by Silverstein et al.18 No cases of invasive carcinoma or lobular neoplasia were included. Data Analysis Direct smears and histologic sections were unmasked and smear diagnoses were compared with the histologic diagnoses. Based on the histologic diagnoses, the direct smears were then classified as NPBD, PBD, or PBDA/DCIS, and cytologic features were compared among these three groups. Because two different reviewers evaluated each direct smear and the review- 35 ers’ assessments were not always concordant, numeric scores were averaged and a feature was considered to be present even if identified by only one reviewer. The proportions of cases exhibiting a cytologic feature were compared among diagnostic groups using the Fisher exact test. Features measured on an ordinal scale were compared between diagnostic groups using the Wilcoxon rank sum test. A significance level of P ⱕ 0.05 was required. Analysis was done separately for air-dried and alcohol-fixed smears. Because multiple cytologic features were being compared, the sequential rejective bonferroni procedure19 was compared with the Fisher exact test or Wilcoxon rank sum P values in order to control for the experimentwise error rate. Because multiple testing of the different cytologic features results in inflated Type I error rates, this procedure allows the use of more conservative P values before declaring significant differences. The result of a Wilcoxon rank sum or Fisher exact test is significant only if it is less than the sequential bonferroni P value. RESULTS Based on the histologic review, the study included smears from 53 NPBD, 33 PBD, 1 PBDA, and 11 DCIS. Of the 11 DCIS, 3 were high grade, 6 were intermediate grade, and 2 were low grade. Diagnostic Accuracy The reviewers’ cytologic diagnoses (on air-dried and alcohol-fixed smears) agreed with the corresponding histologic diagnosis of NPBD in 58%, PBD in 47%, and DCIS or PBDA in 82% of the cases. Only low and intermediate grade DCIS were misclassified cytologically. Comparison of Cytologic Features When individual cytologic features of benign breast disease (NPBD and PBD) were compared with those of atypical or malignant breast disease (PBDA and DCIS), there were many features that differed significantly in degree or frequency on air-dried or alcohol-fixed smears using the Fisher exact test or the Wilcoxon rank sum test (Tables 1 and 2). These features included the presence of myoepithelial cells in epithelial groups, background myoepithelial cells, cell cohesion, monolayers and cell swirling/streaming in epithelial groups, micronucleoli, and a greater degree of architectural complexity of cell groups in the NPBD/PBD group. The PBDA/DCIS group demonstrated greater cellularity, nuclear overlap, pleomorphism, chromatin clumping, hyperchromasia, and necrosis; a greater number of single cells, macronucleoli, and cytoplasmic vacuoles; and a larger nuclear size. However, a few 36 CANCER (CANCER CYTOPATHOLOGY) February 25, 2000 / Volume 90 / Number 1 TABLE 1 Comparison of Cytologic Features in NPBD/PBD and PBDA/DCIS on Air-Dried Smears Cytologic feature Myoepithelial cells in epithelial groupsa Background myoepithelial cells Cellular cohesion Nuclear size Nuclear overlap Nuclear pleomorphism Single cells Cell swirling/streaminga Necrosis Macronucleoli Monolayers Micronucleoli Complexity of epithelial groups Cellularity Cribriform pattern Papillae Cytoplasmic vacuoles Size of epithelial groups Dual cell population Slitlike lumenal spaces Nucleoli absent NPBD/PBD % present or median score 79/84 (94%) 82/86 (95%) 2.5 1.0 2.0 2.0 1.0 61/84 (73%) 3/86 (4%) 10/84 (12%) 44/84 (52%) 68/84 (81%) 1.5 2.0 13/84 (15%) 8/84 (10%) 15/84 (18%) 1.75 27/84 (32%) 18/84 (21%) 8/84 (10%) PBDA/DCIS % present or median score 2/12 3/12 1.0 2.0 3.0 3.25 2.5 2/12 5/12 7/12 1/12 5/12 1.5 2.5 4/12 0/12 3/12 1.75 4/12 2/12 0/12 (17%) (25%) (17%) (42%) (58%) (8%) (42%) (33%) (0%) (25%) (33%) (17%) (0%) TABLE 2 Comparison of Cytologic Features in NPBD/PBD and PBDA/DCIS on Alcohol-Fixed Smears P valueb c 0.00001 0.00001c 0.0001c 0.0001c 0.0001c 0.0001c 0.0001c 0.00031c 0.00054c 0.00080c 0.0046 0.0068 0.037 0.022 0.216 0.590 0.692 0.777 0.999 0.999 0.999 NPDB: non-proliferative breast disease; PBD: proliferative breast disease without atypia; PBDA: proliferative breast disease with atypia; DCIS: ductal carcinoma in situ. a Not statistically significant in alcohol-fixed smears. b Based on the Fisher exact test or the Wilcoxon rank sum test. c Statistically significant compared with the sequential rejective bonferroni procedure. of the features were only significant in either air-dried or alcohol-fixed smears (Tables 1 and 2). These differences remained significant after comparison with the sequential rejective bonferroni procedure, with the exception of cellularity, complexity of epithelial cell groups, and the presence of monolayers and micronucleoli on air-dried smears and the presence of background myoepithelial cells on alcohol-fixed smears (Tables 1 and 2). Table 3 presents the comparison of cytologic features between NPBD and PBD using air-dried smears. Only 2 of the 21 evaluated cytologic features (nuclear pleomorphism and cell swirling) differed significantly based on the Fisher exact test or the Wilcoxon rank sum test. Moreover, none of these features appeared to be significantly different when assessed with the more conservative sequential bonferroni procedure. Using alcohol-fixed smears (Table 4), 5 of the 23 evaluated cytologic features (complexity of epithelial groups, epithelial group size, the presence of a dual epithelial cell population, cell swirling, and the presence of slitlike lumens) differed significantly between Cytologic feature Macronucleoli Monolayers Cellular cohesion Chromation clumping Hyperchromasia Nuclear size Nuclear overlap Nuclear pleomorphism Single cells Necrosis Cytoplasmic Vacuolesa Micronucleoli Cellularity Background myoepithelial cells Myoepithelial cells in epithelial groups Dual cell population Slitlike lumenal spaces Cribriform pattern Complexity of epithelial groups Size of epithelial groups Papillae Cell swirling/streaming Nucleoli absent NPBD/PBD % present or median score PBDA/DCIS % present or median score P-valueb 1/84 (1%) 58/83 (70%) 3.0 1.0 1.0 1.0 2.0 1.5 1.0 0/84 (0%) 6/83 (7%) 74/83 (89%) 1.5 77/84 (92%) 6/12 1/12 1.5 2.0 2.0 1.75 2.5 3.0 2.25 5/12 6/12 5/12 2.75 7/12 0.00001c 0.00007c 0.0001c 0.0001c 0.0001c 0.0001c 0.0001c 0.0001c 0.0001c 0.0001c 0.00036c 0.0005c 0.0029c 0.0064 78/83 (94%) 29/83 (35%) 16/83 (19%) 1/83 (1%) 1.5 1.5 9/83 (11%) 42/83 (51%) 8/83 (10%) 9/12 7/12 0/12 1/12 1.25 1.5 0/12 5/12 1/12 (50%) (8%) (42%) (50%) (42%) (58%) (75%) (58%) (0%) (8%) (0%) (42%) (8%) 0.061 0.201 0.208 0.238 0.244 0.5739 0.597 0.759 0.999 NPBD: non-proliferative breast disease; PBD: proliferative breast disease without atypia; PBDA: proliferative breast disease with atypia; DCIS: ductal carcinoma in situ. a Not statistically significant in air-dried smears. b Based on the Fisher exact test or the Wilcoxon rank sum test. c Statistically significant compared with the sequential rejective bonferroni procedure. NPBD and PBD based on the Fisher exact test or the Wilcoxon rank sum test. However, only complexity and epithelial group size remained significant after adjustment for experimentwise error rate. DISCUSSION Several reports in the literature document the diagnostic accuracy in identifying DCIS by cytology and demonstrate great success at recognizing DCIS as carcinoma (invasiveness not specified) with rates approaching 100%.6 –9,11 In contrast, the reported diagnostic accuracy in identifying benign proliferative breast disease cytologically is more varied. Some authors have reported mediocre success (with overall cytohistologic correlation in 40 – 64% of cases with biopsy-proven ductal hyperplasia without atypia),8,10 whereas others have reported somewhat greater success (with 75% and 87% cytohistologic correlation);7,11,15 but the diagnostic accuracy never equals that for carcinoma. In all reports, the number of cases Cytology of Proliferative Breast Lesions/Frost et al. TABLE 3 Comparison of Cytologic Features in NPBD and PBD on Air-Dried Smears Cytologic feature Nuclear pleomorphism Cell swirling/streaming Size of epithelial groups Dual cell population Cellularity Background myoepithelial cells Complexity of epithelial groups Monolayers Cytoplasmic vacuoles Papillae Necrosis Myoepithelial cells in epithelial groups Nuclear overlap Single cells Cellular cohesion Nucleoli absent Nuclear size Macronucleoli Cribriform pattern Slitlike lumenal spaces Micronucleoli TABLE 4 Comparison of Cytologic Features in NPBD and PBD on AlcoholFixed Smears NPBD % present or median score PBD % present or median score P valuea Cytologic feature 1.5 33/52 (63%) 1.5 13/52 (25%) 2.0 52/53 (98%) 1.5 24/52 (46%) 7/52 (14%) 3/52 (6%) 3/53 (6%) 2.0 28/32 (88%) 2.0 14/32 (44%) 2.0 30/33 (91%) 2.0 20/32 (63%) 8/32 (25%) 5/32 (16%) 0/33 (0%) 0.016 0.023 0.054 0.092 0.097 0.155 0.160 0.180 0.242 0.249 0.282 50/52 (96%) 2.0 1.0 2.5 3/52 (6%) 1.0 7/52 (13%) 8/52 (15%) 11/52 (21%) 42/52 (81%) 29/32 (91%) 2.0 1.0 2.5 3/32 (9%) 1.0 3/32 (9%) 5/32 (16%) 7/32 (22%) 26/32 (81%) 0.364 0.414 0.509 0.544 0.670 0.678 0.735 0.999 0.999 0.999 Complexity of epithelial groups Size of epithelial groups Dual cell population Cell swirling/streaming Slitlike lumenal spaces Nuclear size Cellular cohesion Nuclear pleomorphism Cellularity Chromation clumping Papillae Nuclear overlap Hyperchromasia Single cells Myoepithelial cells in epithelial groups Cytoplasmic vacuoles Monolayers Background myoepithelial cells Cribriform pattern Necrosis Nucleoli absent Micronucleoli Macronucleoli NPBD: non-proliferative breast disease; PBD: proliferative breast disease without atypia; PBDA: proliferative breast disease with atypia; DCIS: ductal carcinoma in situ. a Based on the Fisher exact test or the Wilcoxon rank sum test. of PBD has been small, ranging from 9 to 50. Aspirates of PBD are most frequently misclassified as PBDA or suspicious for carcinoma but can also be confused with NPBD.7,8,10,11,15 It is recognized that proliferative breast disease is a morphologic continuum with overlapping cytologic and histologic features between lesions within the continuum. Indeed, significant interobserver variability in the histologic and cytologic classification of proliferative breast disease has been documented. 14,20,21 Such morphologic overlap has hampered our ability to identify PBD accurately and separate it from NPBD and PBDA. In a prior evaluation of cytologic features useful in separating PBD from NPBD, we identified only one cytologic feature that was significantly useful, namely, the presence of cell swirling within epithelial groups in air-dried aspirates of PBD.15 The results of our diagnostic accuracy in this study corroborates those of others. Cytohistologic correlation was achieved in 58% of NPBD, 47% of PBD, and 82% of PBDA/DCIS cases. Only low and intermediate grade DCIS were misclassified cytologically. Other factors that may be hampering our ability to accurately identify PBD cytologically is the heteroge- 37 NPBD % present or median score PBD % present or median score P valuea 1.5 1.5 12/51 (24%) 21/51 (41%) 6/51 (12%) 1.0 3.0 1.0 1.5 1.0 4/51 (8%) 1.5 1.0 1.0 1.5 2.0 17/32 (53%) 21/32 (66%) 10/32 (31%) 1.0 3.0 1.5 2.0 1.0 5/32 (16%) 2.0 1.0 1.0 0.0027b 0.003b 0.009 0.042 0.044 0.065 0.074 0.092 0.096 0.230 0.297 0.297 0.302 0.335 49/51 (96%) 2/51 (4%) 34/51 (67%) 47/52 (90%) 1/51 (2%) 0/52 (0%) 5/51 (10%) 45/51 (88%) 1/52 (2%) 29/32 (91%) 4/32 (13%) 24/32 (75%) 30/32 (94%) 0/32 (0%) 0/32 (0%) 3/32 (9%) 29/32 (91%) 0/32 (0%) 0.369 0.369 0.470 0.704 0.99 0.99 0.99 0.99 0.99 NPDB: non-proliferative breast disease; PBD: proliferative breast disease without atypia; PBDA: proliferative breast disease with atypia; DCIS: ductal carcinoma in situ. a Based on the Fisher exact test or Wilcoxon rank-sum test. b Statistically significant compared with the sequential rejective bonferroni procedure. neity of changes within a given breast lesion. Foci of DCIS are often accompanied by PBD and PBDA, and proliferative breast disease may be present in a background of nonproliferative breast tissue; consequently, it is difficult to identify with certainty the origin of the aspirated material. Cytohistologic correlation has also been hampered in previous studies by the difficulty in determining the exact area within an excisional biopsy from which an aspirate was derived. The presence of needle tracts can be helpful, but they are not always present. In an effort to eliminate much of these sampling difficulties, we scraped specified foci of excisional breast biopsies and correlated them with the histologic sections prepared from the exact areas that were scraped. Smears of PBD and NPBD were similar in many respects. Both typically exhibited cohesive cell groups with mild nuclear overlap and few single cells, no or mild nuclear pleomorphism, a nuclear size less than three times the size of an RBC, and monolayers. Chromatin clumping and hyperchromasia were absent. Mi- 38 CANCER (CANCER CYTOPATHOLOGY) February 25, 2000 / Volume 90 / Number 1 FIGURE 1. Nonproliferative breast disease is shown. (A) Ductal epithelial cells are arranged in a monolayer with good cellular polarity and interspersed myoepithelial cells (Diff-Quik stain, original magnification ⫻1000). (B) A monolayer of ductal epithelium with honeycomb arrangement, regular nuclei, and fine, evenly distributed chromatin is shown (H & E, original magnification ⫻1000). cronucleoli were often present. Less consistently, smears of NPBD and PBD demonstrated a low to moderate degree of cellularity. However, in standard univariate analysis, we were able to identify six cytologic features that differed in NPBD and PBD. PBD showed more complexity of epithelial groups, slitlike lumenal spaces, a mixture of apocrine metaplastic cells, nuclear pleomorphism, very large epithelial groups/sheets, and cell swirling/streaming (as in our previous analysis15) (Figs. 1 and 2). The latter feature was the only one that was significant in both air-dried and alcohol-fixed smears (Tables 3 and 4). However, when the more conservative sequential rejective bonferroni procedure was used, only the degree of complexity and size of epithelial groups in the alcoholfixed preparations remained significant in distinguishing NPBD from PBD. The different results observed for alcohol-fixed and air-dried smears in the comparison of NPBD and FIGURE 2. Proliferative breast disease without atypia is shown. (A) Cellular smear showing large sheets of epithelial cells with complex architecture (Diff-Quik stain, original magnification ⫻400). (B) This epithelial group illustrates a swirling pattern. The myoepithelial cells are evident (Diff-Quik stain, original magnification ⫻1000). PBD may have been a result, at least in part, of sampling. Even though the study design allowed precise histologic correlation and greatly reduced sampling error, it may not have eliminated it. NPDB and PBD were often mixed in the histologic sections prepared from the areas sampled. The amount of each epithelial type and the sizes of the epithelial sheets resulting from each smear of the same area may not have been entirely consistent between the smear prepared for alcohol fixation and the air-dried smear. Our data suggest, however, that the architecture of epithelial groups and sheets is better assessed in alcohol-fixed smears. The spreading of cells that results from airdrying may result in flattening of the epithelial sheets, making appreciation of architectural detail more difficult. There were far more cytologic features that distinguished NPBD/PBD from DCIS/PBDA than distinguished NPBD from PBD. Seventeen of the 23 evaluated features showed significance in distinguishing Cytology of Proliferative Breast Lesions/Frost et al. FIGURE 3. Ductal carcinoma in situ, low grade, is shown. A cellular smear shows two sheets of epithelial cells exhibiting mild nuclear enlargement and pleomorphism as well as relatively abundant cytoplasm. The epithelial sheets lack architectural complexity and exhibit some loss of cohesion at their peripheries. Myoepithelial cells are absent in the sheet on the right (Diff-Quik stain, original magnification ⫻400). the NPBD/PBD group from DCIS/PBDA in air-dried and/or alcohol-fixed smears (Tables 1 and 2). Eleven features were significant in both preparations, including the presence of myoepithelial cells in the background, cellularity, cohesion, single cells, monolayers, necrosis, nuclear enlargement, pleomorphism and overlap, and nucleoli (micro vs. macro). Since hyperchromasia and chromatin clumping were evaluated in alcohol-fixed smears only, their statistical significance was only demonstrated in this preparation. Smears of DCIS usually exhibited loss of cellular cohesion with many single cells, large nuclear size (greater than 3 times the size of an RBC) with moderate to marked overlap, and moderate to marked pleomorphism. The chromatin was clumped and hyperchromatic on alcohol-fixed smears. There was often a background of necrosis. Less consistently, the smears of DCIS were highly cellular. There was an absence of myoepithelial cells within cell groups and in the smear background, and an absence of cell streaming and swirling (Fig. 3). Cytoplasmic vacuoles and macronucleoli were sometimes present. Only one case of PBDA was included in this study. It demonstrated a degree of nuclear overlap intermediate between NPBD/PBD and DCIS, and the epithelial cells were arranged in small groups. Again, differences between alcohol-fixed and airdried smears in the comparison of NPBD/PBD and DCIS/PBDA may have resulted from differences in sampling between the two smears. However, the three features that differed in significance between air-dried smears and alcohol-fixed smears (i.e., cell swirling/ 39 streaming, myoepithelial cells in epithelial groups, and cytoplasmic vacuoles) were more often identified in the alcohol-fixed smears than in air-dried smears, suggesting the superiority of alcohol-fixed smears in evaluating these features. The presence of necrosis in some air-dried smears, but not alcohol-fixed smears, prepared from NPBD/PBD likely results from the presence of necrosis-like material present in ectatic ducts. This material is more likely to remain adherent to air-dried than alcohol-fixed slides. In order to minimize the sampling error, our study design was based on direct scrapes of excisional breast biopsies. There are few studies that assess the comparability of cytologic morphology in scrape preparations and FNA smears. In a recent comparison of morphology in scrape preparations and FNAs, it was demonstrated that scrape preparations yielded smears with smaller clusters and more single cells and were uniformly more cellular than FNA smears, but were otherwise similar.22 These few differences must be considered when applying the results of this study to the interpretation of FNAs of the breast, but they do not affect our comparison of NPD with PBD and PBDA/DCIS, because only scrape preparations were included for evaluation. It is important for the cytopathologist to distinguish PBDA/DCIS from NPBD/PBD by FNA, because the distinction carries management implications. Rendering a cytologic diagnosis of PBDA/DCIS prompts further workup, often a needle or open biopsy, to define better the type and extent of the lesion. On the other hand, patients with a cytologic diagnosis of benign NPBD or PBD may not require additional workup. The clinical significance of distinguishing NPBD from PBD is less dramatic; although women with PBD have an approximately twofold greater risk of developing invasive breast carcinoma than women in the general population, further treatment is currently not indicated.17,23 However, the ability to identify women with PBD accurately, particularly when these women also have a family history of breast carcinoma, by FNA may be potentially useful in identifying candidates for chemoprevention regimens, such as tamoxifen, which has recently been demonstrated to be effective in reducing the incidence of breast carcinoma in high risk women by 49%.24 In conclusion, we have demonstrated that many cytologic features allow a separation of DCIS from NPBD/PBD. However, there are relatively few features that reliably allow the distinction of PBD from NPBD. Consequently, the identification of PBD cytologically is limited. 40 CANCER (CANCER CYTOPATHOLOGY) February 25, 2000 / Volume 90 / Number 1 REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. Koss LG. The palpable breast nodule: a cost-effectiveness analysis of alternate diagnostic procedures. The role of needle aspiration biopsy. Cancer 1993;72:1499 –502. Donegan WL. Evaluation of a palpable breast mass. N Engl J Med 1992;327:937– 41. DeMay RM. The breast. In: DeMay, editor. The art and science of cytopathology. Aspiration cytology, Volume II. Chicago: ASCP Press, 1996:847–937. Silverman JF. Breast. In: Bibbo M, editor. Comprehensive cytopathology. 2nd edition, Philadelphia: W. B. Saunders, 1997:731– 80. Wang HH, Ducatman BS, Eick D. Comparative features of ductal carcinoma in situ and infiltrating ductal carcinoma of the beast on fine needle aspiration biopsy. Am J Clin Pathol 1989;92:736 – 40. Bibbo M, Scheiber M, Cajulis R, Keebler CM, Wied GL, Dowlatshahi K. Stereotaxic fine needle aspiration cytology of clinically occult malignant and premalignant breast lesions. Acta Cytol 1988;32:193–201. Masood S, Frykberg ER, McLellan GL, Dee S, Bullard JB. Cytologic differentiation between proliferative and nonproliferative breast disease in mammographically guided fineneedle aspirates. Diagn Cytopathol 1991;7:581–90. Sneige N, Staerkel GA. Fine-needle aspiration cytology of ductal hyperplasia with and without atypia and ductal carcinoma in situ. Hum Pathol 1994;25:485–92. Abendroth CS, Wang HH, Ducatman BS. Comparative features of carcinoma in situ and atypical ductal hyperplasia of the breast on fine-needle aspiration biopsy specimens. Am J Clin Pathol 1991;96:654 –9. Thomas PA, Cangiarella J, Raab SS, Waisman J. Fine needle aspiration biopsy of proliferative breast disease. Mod Pathol 1995;8:130 – 6. Gangopadhyay M, Nijhawan R, Joshi K, Gupta S. Cytology of “significant” breast ductal proliferations. Acta Cytol 1997;41: 1112–20. Dziura BR, Bonfiglio TA. Needle cytology of the breast: a qualitative study of the cells of benign and malignant ductal neoplasia. Acta Cytol 1979;23:332– 40. Dawson AE, Mulford DK, Sheils LA. The cytopathology of 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. proliferative breast disease: comparison with features of ductal carcinoma in situ. Am J Clin Pathol 1995;103:438 – 42. Sidawy MK, Stoler MH, Frable WJ, Frost AR, Masood S, Miller TR, et al. Interobserver variability in the classification of proliferative breast lesions by fine needle aspiration: results of the Papanicolaou Society of Cytopathology Study. Diagn Cytopathol 1998;18:150 – 65. Frost AR, Aksu A, Kurstin T, Sidawy MK. Can nonproliferative breast disease and proliferative breast disease without atypia be distinguished by fine-needle aspiration cytology? Cancer ( Cancer Cytopathol) 1997;81:22– 8. Page DL, Anderson TJ, Rogers LW. Carcinoma in situ (CIS). In: Page DL, Anderson TJ, editors. Diagnostic histopathology of the breast. New York: Churchill Livingstone, 1987:157–92. Page DL, Anderson TJ, Rogers LW. Epithelial hyperplasia. In: Page DL, Anderson TJ, editors. Diagnostic histopathology of the breast. New York: Churchill Livingstone, 1987:120 – 45. Silverstein MJ, Poller DN, Weisman JR, Colburn WJ, Barth A, Gierson ED, et al. Prognostic classification of breast ductal carcinoma-in-situ. Lancet 1995;345:1154 –7. Zhang J, Quan H, Ng J, Stepanavage ME. Some statistical methods for multiple endpoints in clinical trials. Control Clin Trials 1997;18:204 –21. Rosai J. Borderline epithelial lesions of the breast. Am J Surg Pathol 1991;15:209 –21. Schnitt SJ, Connolly JL, Tavassoli FA, Fechner RE, Kempson RL, Gelman R, et al. Interobserver reproducibility in the diagnosis of proliferative breast lesions using standardized criteria. Am J Surg Pathol 1992;16:1133– 43. Blumenfeld W, Hashmi N, Sagerman P. Comparison of aspirate, touch and scrape preparations simultaneously obtained from surgically excised specimens. Acta Cytol 1998; 42:1414 – 8. Dupont WD, Page DL. Risk factors for breast cancer in women with proliferative breast disease. N Engl J Med 1985; 312:146 –51. Fisher B, Costantino JP, Wickerham DL, Redmond CK, Kavanah M, Cronin WM, et al. Tamoxifen for prevention of breast cancer: report of the national surgical adjuvant breast and bowel project P-1 study. J Natl Cancer Inst 1998;90: 1371– 88.