1258

Breast Cancer Risk Associated with Proliferative Breast Disease and Atypical Hyperplasia William D. Dupont, Ph.D.,* Fritz F. Pad, M.D., Ph.D.,t William H. Hartmann, M.D.,$ Louise A. Brinton, Ph.D.,§ Alan C. Winfield, M.D., ( 1 John A . Worrell, M.D., (I Peggy A. Schuyler, X.N.," and Walton D. Plummer, B.S.*

Background. Women with proliferative breast dis- ease (PD) have been observed to have an increased risk of breast cancer. The authors evaluated the effect of PD on breast cancer risk in a case-control study among partici- pants of the Breast Cancer Detection Demonstration Proj- ect (BCDDP).

Methods. More than 280,000 women were screened in the BCDDP at 29 centers. Study subjects were selected from BCDDP participants who underwent biopsy that re- vealed benign breast tissue. There were five BCDDP centers for which histologic slides were available on more than 85% of the benign biopsy specimens. Case pa- tients for this study were the 95 women from these five centers who had breast cancer develop during follow-up. Two matched control patients who did not have breast

From the Departments of 'Preventive Medicine, tPathology, and ))Radiology, Vanderbilt University School of Medicine, Nash- ville, Tennessee; the +Department of Pathology, Memorial Medical Center, Long Beach, California; and the §Environmental Epidemiol- ogy Branch, National Cancer Institute, Bethesda, Maryland.

Supported by National Cancer Institute Grants R01 CA46492 and R01 CA50468.

The authors thank Dr. David L. Page for reviewing his histologic classification scheme with them at the start of this project, Dr. Philip R. Taylor for his help in making data from the Breast Cancer Detec- tion Demonstration Project (BCDDP) available, Thomas L. Riley for preparing the follow-up data file, Marcia Freudenthal for help in creating the histologic data file, and Janelle Steele for assistance in preparing the manuscript. The authors also thank Drs. Margaret Abernathy, Jerry B. Buchanan, Thomas Carlile, H. P. Constantine, Robert L. Egan, Bernard Fisher, Fred I. Gilbert, Morton J. Goodman, Lewis W. Guiss, JoAnn D. Haberman, L. J. Humphrey, Harold J. Isard, Richard Lester, A. Hamblin Letton, John E. Martin, Marvin V. McClow, John Milbrath, Myron Moskowitz, Henry Pendergrass, Alfred M. Popma, Arthur J . Present, Ned Rodes, Benjamin F. Rush Jr., Robert Schweitzer, Phillip Strax, Barbara Threatt, Leslie W. Whitney, and Donald C. Young for their assistance as directors of the individ- ual BCDDP projects.

Address for reprints: William D. Dupont, Ph.D., Vanderbilt Uni- versity School of Medicine, Department of Preventive Medicine, A-1124 Medical Center North, Nashville, TN 37232-2637.

Accepted for publication August 20, 1992.

cancer develop were selected for each case. The biopsy slides were reviewed by two pathologists who were blinded with regard to cancer outcome.

Results. Women with atypical hyperplasia (AH) had 4.3 times the breast cancer risk of women without PD (95% confidence interval [CI], 1.7-11). In women with PD lacking AH, the relative risk was 1.3 (95% CI, 0.77-2.2). A family history of breast cancer (FH] increased breast cancer risk 2.4 times (95% CI, 1.4-4.3). The joint occur- rence of FH and AH had a strong synergistic effect on breast cancer risk.

Conclusions. AH is a reliable marker of increased breast cancer risk among women undergoing breast biopsy. Cancer 1993; 71:1258-65.

Key words: breast cancer, breast disease, proliferative disease, atypical hyperplasia, family history, cysts, calci- fications.

A history of biopsy that reveals benign breast tissue is a well established risk factor for breast cancer.',' In a ret- rospective cohort study, Dupont and Page3 found that women with atypical hyperplasia (AH) had a four to fivefold increase in breast cancer risk, that women with proliferative disease without atypia (PDWA) had a twofold increase in breast cancer risk, and that women without proliferative disease (PD) had no elevation in breast cancer risk compared with women of similar age from the general population. In this original study, two pathologists, David L. Page and Lowell W. Rogers, used Page's criteria4 for benign breast disease to evaluate his- tologic slides and establish the diagnosis of I'D, PDWA, and AH. In the current study, we sought to determine whether these results could be confirmed in other women with other pathologists applying Page's criteria. To do this we performed a nested case-control study of women who had undergone a breast biopsy that re- vealed benign breast disease while participating in the Breast Cancer Detection Demonstration Project (BCDDP).

Proliferative Disease and Breast Cancer/Dupont et al. 1259

Methods

Study Design

The methods of the BCDDP have been described in detail e l ~ e w h e r e . ~ , ~ This project, which was sponsored jointly by the American Cancer Society and the Na- tional Cancer Institute (NCI), recruited more than 280,000 women to undergo annual screening for breast cancer during a 5-year period. The study subjects ranged in age from 35 to 74 years. They received annual physical examinations of their breasts and mammo- grams at one of 29 centers across the country and were asked to complete a questionnaire at each examination or when a breast operation was performed. Informed consent for participation in this project was obtained from all patients. The initial phase of this study was from 1973 to 1978. An additional 5 years of follow-up data were collected by telephone for a subset of these patients, which included 19,734 study subjects who had undergone breast biopsy as a result of findings ob- served during the screening examinations. The study cohort consisted of these biopsied women except for those excluded as described here:

1. 2319 women with invasive breast cancer diagnosed

2. 18 women with self-reported breast cancer; 3. 26 women with an unknown date of cancer diag-

4. 194 women whose age was unknown; and 5. 2016 women whose breast biopsy specimens origi-

nally were diagnosed by their BCDDP pathologists as acute inflammation or abscess, chronic inflam- mation, chronic abscess, granulomatous inflamma- tion, fat necrosis, galactocele, fibroadenoma with- out evaluable parenchyma, or congenital or devel- opmental anomaly.

at or before their first BCDDP breast biopsy;

nosis;

After these exclusions, 15,161 women remained in our study cohort. The entry biopsy for these subjects was their first breast biopsy in the BCDDP (1467 women had multiple biopsies during the screening phase of this study). Each subject’s follow-up started with her entry biopsy and ended when the first of the following occurred: she was diagnosed with invasive breast cancer, had her second mastectomy, she died, was lost to follow-up, or answered the last question- naire during the additional 5 years of follow-up. Five hundred seven cohort members had invasive breast cancer develop during follow-up. These patients were the case subjects of our nested case-control study. Their cancers were confirmed by their BCDDP pathologist for 97.5% of cases. The remainder were confirmed by a

review of hospital records. Information on nonanato- mic risk factors was obtained from patient interviews at or before the end of the screening phase of the BCDDP.

For each case subject, two matched control subjects were selected from the remainder of the cohort. Each control subject was selected from the same BCDDP center as their corresponding case patient and matched for age and year of biopsy. In addition, the length of follow-up for control subjects had to be at least as long as that of their matched case patient. That is, each con- trol patient had to be at risk of breast cancer at the time when her matched case patient developed the disease. The maximum difference in age and year of biopsy be- tween case patients and their matched controls was 6.2 and 0.9 years, respectively. These differences were less than 1.0 and 0.3 years, respectively, for 90% of control patients.

Most of the histologic slides from the BCDDP had been previously collected by Dr. William H. Hartmann when he was chairman of the Pathology Quality Con- trol Program of the BCDDP. After our nested case-con- trol study subjects were identified, an extensive effort was made to locate all slides from these patients’ entry biopsies that were not previously in Hartmann’s reposi- tory.

Histologic Methods

The entry biopsy specimens of study subjects wereinde- pendently read by Drs. Hartmann and Parl using the criteria defined by Page.4 These pathologists have been colleagues of Dr. Page for many years, and were famil- iar with his classification scheme for benign breast dis- e a ~ e . ~ , ~ Before reading the study slides, these three pa- thologists met to review Page’s criteria using nonstudy histologic specimens. Thereafter, no discussion of Page‘s criteria occurred between Page and this study’s pathologists. After recording their independent diag- noses, Hartmann and Parl met to resolve discrepancies.

Twenty-eight (3.2%) patients whose entry biopsy specimen was found on review to contain carcinoma in situ were excluded from the study. This exclusion was done after the slide review was performed because the differential diagnosis of AH and low-grade microscopic carcinoma in situ is subtle, and the original diagnoses by BCDDP pathologists were made before the develop- ment of Page’s histologic criteria.

Cysts larger than 3 mm in diameter were diagnosed according to the criteria of Page and Anderson.8

Statistical Methods

Data were analyzed using conditional and uncondi- tional logistic regression.’ Conditional analyses were

1260 CANCER Feebruary 15, 1993, Volume 71, No. 4

performed for matched data using the Parameter Esti- mation through Conditional Probability Analysis (PE- CAN") program (SAS Institute, Cary, NC). Uncondi- tional analyses were done with the SAS statistical pack- age." These two approaches gave highly consistent results, with the unconditional models giving narrower confidence intervals (CI). For this reason, the results presented here were derived using unconditional mod- els. Model selection was guided by examining model deviance statistic^'^^'^ and the size of parameter esti- mates for potential confounding variables. All uncondi- tional regression models were adjusted for BCDDP center, patient age at biopsy, year of biopsy, and parity by including appropriate covariates as independent variables. Other potential confounders that were con- sidered were age at birth of first child, age at menarche and menopause, race, education, use of female hor- mones, and body mass index. These variables had a trivial effect on model deviance and were associated with parameter estimates of small magnitude. They were excluded from our analyses for these reasons.

Interaction terms were included in the model when multiple risk factors were assessed simultaneously. The potential effects of multivariate outliers on our parame-

ter estimates were evaluated using the delta beta rnethod.l4 Adjusted relative risks were estimated by odds ratios obtained from the logistic regression mod- els. CI for these estimates were derived from the asymptotic covariance matrix of the parameter esti- mates using the SAS/IML program (SAS Institute, Cary, NC).15 Interobserver variability was assessed us- ing kappa statistics." All P values were calculated with respect to two-sided alternative hypotheses. The P val- ues in Tables 1 and 2 were derived under the null hy- pothesis that the corresponding relative risks equal one.

Results

Histologic Consistency

Table 3 shows the original independent diagnoses of the study pathologists. Independent agreement was ob- tained for 63% of study subjects, with most discrepan- cies near the border between AH and PDWA or be- tween PDWA and no I'D. The interobserver kappa sta- tistic was 0.39, suggesting a fair level of agreement beyond that which would be expected by chance alone. All discrepant diagnoses were resolved without difi-

Table 1. Relative Risks* of Breast Cancer Associated With Proliferative Disease Without Atypia, Atypical Hyperplasia, A First-Degree Family History of Breast Cancer, and Menopausal Status of Biopsy

Risk factors cases controls risk interval P value

No PD 35 106 PDWA 46 111 1.3 0.77-2.2 0.32 AH 14 10 4.3 1.7-11 0.002

No FH 65 189 3 $ FH 30 38 2.4 1.4-4.3 0.002

No PD without FH 20 87 PDWA without FH 36 93 1.7 0.92-3.2 0.09

AH without FH 9 9 4.2 1.4-12 0.09 No PD with FH 15 19 3.6 1.5-8.6 0.004 PDWA with FH 10 18 2.6 1.0-6.4 0.05

0.006 AH with FH 5 1 22

No. of No. of Relative 95% confidence

2.4-203

Postmenopausal 57 149 1$ Premenopausal 38 78 1.2 0.67-2.2 0.54

No PD, postmenopause 20 68 PDWA, postmenopause 29 73 1.4 0.71 -2.7 0.33 AH, postmenopause 8 8 3.3 1.1-10 0.04

PDWA, premenopause 17 38 1.6 0.69-3.7 0.27 AH, premenopause 6 2 12 2.0-68 0.006

No PD, premenopause 15 38 1.3 0.56-3.2 0.52

PD: proliferative disease = PDWA or AH; PDWA: proliferative disease without atypia; AH: atypical hyperplasia; FH: first-degree family history of breast cancer. * Adjusted for patient age at biopsy, year of biopsy, and study center. $ Denominator of subsequent relative risks.

Proliferative Disease and Breast Cancer/Dupont et al. 3 261

Table 2. Relative Risks* of Breast Cancer Associated With Calcification, Cysts, and Other Risk Factorst

No. of No. of Relative 95% confidence Risk factors cases controls risk interval P value

No CAL 59 151 1$

No PD without CAL 29 84 1$

CAL 36 76 1.2 0.75-2.1 0.40

PD without CAL 30 67 1.3 0.71-2.4 0.39 No PD with CAL 6 22 0.80 0.29-2.2 0.66 PD with CAL 30 54 1.7 0.89-3.2 0.11

No cysts cysts

74 197 21 30 1.9 0.99-3.5 0.05

No I'D without cysts 31 94 1$ PDWA without cysts 33 96 1.1 0.61-1.9 0.80

No PD and cysts 4 12 1.0 0.29-3.4 1.00 PDWA and cysts 13 15 2.7 1.1-6.5 0.02

AH without cysts 10 7 4.2 1.5-12 0.08

AH with cysts 4 3 4.3 0.89-21 0.07

No cysts or FH 51 164

No cysts with FH 23 33 2.4 1.3-4.6 0.006 Cysts without FH 14 25 1.9 0.89-3.9 0.10

Cysts with FH 7 5 4.2 1.2-14 0.02 C A L calcification; PD: proliferative disease = PDWA or AH; P D W A proliferative disease without atypia; AH: atypical hyperplasia; FH: first-degree family history of breast cancer. ' Adjusted for patient age at biopsy, year of biopsy, and study center. i Denominator of relative risk.

culty through discussion at a double-viewing micro- scope. In only a few cases was a major discrepancy observed. They were found on review to be attributable to an oversight by one or the other of the pathologists. The results given in Tables 1 and 2 of this report are based on the consensus diagnoses.

Table 3. Percent Concordance* of Independent Diagnoses by Study Pathologists?

Dr. Hartmann's diagnosis Dr. Parl's diagnosis No PD PDWA AH Total

No PD 38.9 10.1 2.0 51.0 PDWA 14.1* 19.4 6.3 39.8 AH 1.0 3.2 5.0 9.2 Total 54.0 32.7 13.3 100

Consensus Diagnosest

Total 43.8 48.8 7.4 100

I'D: proliferative disease; PDWA: proliferative disease without atypia; AH: atypical hyperplasia.

E.g., 14.1 90 of study biopsy specimens were independently diagnosed as PDWA by Dr. Pad, and no PD by Dr. Hartrnann. Dr. Par1 diagnosed 39.8% of study biopsy specimens as PDWA. Dr. Hartmann diagnosed 5440 as having no PD. t The consensus diagnoses were reached after a joint review of all discrepant diagnoses at a double-viewing microscope.

Elimination of Centers wi th Poor Slide Availability

Entry biopsy slides were unavailable for review in 43% of study subjects. The percentages of case and control patients whose slides were missing were 49% and 39%, respectively. The difference in slide retrieval rates be- tween cases and controls was highly significant and suggests that a serious availability bias may affect our overall results. Such a bias is plausible because physi- cians of patients with cancer may wish to obtain pre- vious benign biopsy slides for review; once obtained, these slides may not have been returned. If slides with AH were less likely to be returned than those with less worrisome lesions, a slide availability bias would ensue. The crude overall relative risks of breast cancer for pa- tients with AH and PDWA were 2.1 and 1.2, respec- tively. When these risks were calculated by BCDDP center, we found that centers with low percentages of missing slides tended to have higher estimates of breast cancer risk, a finding that is consistent with an availabil- ity bias. To cope with this problem we restricted our analyses to the five centers with at least 85% of the entry biopsy slides available. These centers were lo- cated in Portland, Oregon; Des Moines, Iowa; Hono- lulu, Hawaii; Nashville, Tennessee; and Wilmington, Delaware. For these centers combined, slides were avail-

1262 CANCER February 25, 1993, Volume 71, No. 4

able for 90% of the study subjects (81% of case subjects and 95% of control subjects). All results given in the remainder of this report are derived from these patients. For these 5 centers, 1.8% of case and control subjects were excluded because of a diagnosis of carcinoma in situ on their entry biopsy. A total of 95 cases and 227 controls were used in the following analyses.

Breast Cancer Risk in Women with Benign Breast Disease

Breast cancer risks associated with PD, a first degree family history of breast cancer (FH), and menopausal status at entry biopsy are given in Table 1. The first line in each group of risk factors in this table gives the de- nominator of the subsequent relative risks in the group. For example, the second line of the first group in Table 1 states that women who had PDWA had 1.3 times the risk of women without I'D. The second line of the third group in Table 1 states that women with PDWA but no FH had 1.7 times the risk of women with neither I'D nor FH. Note that the magnitudes of these relative risks depend on the absolute risks in the numerator and de- nominator groups. Thus, the fact that the latter of these two risks is greater than the former suggests that the absence of FH has a greater protective effect on patients without PD than on patients with PDWA. It also should be noted that the precision of the relative risk estimates associated with the joint occurrence of two rare risk factors is low. This is because these estimates are neces- sarily based on a small number of patients. The preci- sion of the risk estimates in Tables 1 and 2 may be judged by the widths of the associated CI.

Women with AH had a fourfold elevation in breast cancer risk compared to women without PD. FH also was an independent risk factor for breast cancer in these study subjects. Together, these risk factors had a pronounced synerptic effect on breast cancer risk. Women with both AH and FH had a fivefold increase in breast cancer risk compared with women with AH but not FH (respective relative risks of 22 and 4.2 compared with women with neither PD nor FH). However, the CI of the risk for both AH and FH is extremely wide, and only one control patient had both AH and FH. Meno- pausal status had little overall effect on the age-ad- justed risk of breast cancer. Breast cancer risk ratios for women with AH compared with those for women without I'D were higher in premenopausal women than in postmenopausal women.

An average of 2.9 histologic slides per biopsy were available for review. Significantly more slides were available for patients with AH or PDWA than for those without PD (an average of 6.6, 3.0, and 2.0 slides per

biopsy were available in patients with AH, PDWA, and no I'D, respectively).

Table 2 shows the effects of calcifications and cysts on breast cancer risks. (In this report, calcification refers to epithelial calcification observed on the histologic slide. Mammographic calcification is not discussed.) Calcification was not associated with a significant in- crease in breast cancer risk. However, calcification was more likely to be found in biopsy specimens with prolif- erative disease than in those without (correlation coeffi- cient, 0.28; P < 0.0005). This suggests that calcification may be a risk factor for PD. The presence of cysts signifi- cantly increased breast cancer risk, although this in- crease was largely restricted to patients with PDWA. There was no evidence that the presence of cysts af- fected breast cancer risk in women without PD or in those with AH. A fourfold increase in breast cancer risk was observed in women with both cysts and FH.

Discussion

Other Studies That Have Used Page's Criteria

Figure 1 summarizes the effects of AH and PDWA that have been reported in the literature. The results of this study for PD are consistent with those reported by Du- pont and Page.3 Our relative risks of 1.3 and 4.3 for PDWA and AH are slightly lower than the correspond- ing risks of 1.9 and 5 .3 reported previ~usly.~ However, there is broad overlap between the CI of the two stud- ies. Both studies showed strong and similar synergistic effects on the breast cancer risk of PD and FH. The greater statistical power of the earlier study makes its tenfold elevation in risk for women with both AH and FH relative to women with neither PD nor FH more credible than the corresponding 20-fold increase in breast cancer risk reported here.

Dupont and Page3 observed an interaction between the effects of PD and menopausal status on breast cancer risk: premenopausal women without PD had the same breast cancer risk as the general population, whereas postmenopausal women without PD appeared to be at reduced breast cancer risk; the relative risk of breast cancer associated with PD was substantially greater among postmenopausal subjects than among premenopausal women. We observed a similar protec- tive effect of nonproliferative lesions in postmeno- pausal women, but did not observe the same interaction between PD and menopausal status. We also agreed with the results of this earlier study concerning the ab- sence of breast cancer risk associated with calcification in women without I'D. However, we did not observe a significant elevation in risk associated with calcification among women with PD. Women with both PD and

Proliferative Disease and Breast CancerlDupont e t al . 1263

Atypical Hyperplasia

Proliferative Disease without Atypia

Dupont & Page3

PaIIi et a1.18

London et al.17

I No Proliferative Disease @ Dupont et al.(This Paper)

Reference Population

I I I I I I I I I I I I I I I

0 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 14

Relative Risk of Breast Cancer Figure 1 Relative risks of breast cancer associated with AH and PDWA that have been reported in the literature. Relative risks are calculated with respect to women who underwent biopsy but did not have PD. The horizontal black lines denote 95% CI for these relative risks.

calcification had an estimated 69% elevation in risk compared with those without either of these risk factors (P = 0.1 1). The corresponding risk elevation from Du- pont and Page was 266% (P < 0.0001). The combined effects of cysts and FH on breast cancer risk in the two studies was similar, with cysts having a somewhat greater effect on cancer risk in our study than in that of Dupont and Page.3

Dupont and Page3 observed a prevalence of AH, PDWA, and no PD of 4%, 27% and 69%, respectively. Their reported prevalence of AH and PDWA is some- what lower than that reported here. However, the women studied by Dupont and Page3 were biopsied in the pre-mammographic era, and AH does not form a palpable lesion. Rubin et al.I7 reported a prevalence of AH of 10% in mammographically directed biopsies, a finding that is more consistent with our current results in a mammographically screened population.

London et a1." conducted a nested case-control study of women from the Nurses' Health Study. They selected study subjects who had undergone breast biopsies that revealed benign breast tissue, and they evaluated the slides from these benign lesions using Page's criteria. They found breast cancer risks for

women with PDWA and AH relative to women without PD of 1.6 (95% CI, 1.0-2.5) and 3.7 (95% CI, 2.1-6.8), respectively. Thus, their findings are consistent with ours and those of Dupont and Page.3 They also found that the risk associated with AH was greater in pre- menopausal women than in postmenopausal women. This finding agrees with the results of our current study but not with those of Dupont and Page.3 Menopausal status had little effect on breast cancer risk in women with PDWA in our study or that of London et al."

Palli et al.I9 performed a case-control study nested in a cohort of approximately 6000 women who under- went benign breast biopsies in the Florence district of Italy. They found that breast cancer risks in women with PDWA and AH relative to those without I'D were 1.3 (95% CI, 0.5-3.5) and 13 (95% CI, 4.1-42), respec- tively. Their relative risk for AH is higher than that reported by other authors, although their CI for this risk is wide. Their risk estimate for PDWA agrees with ours.

The studies described in Figure 1 were performed by investigators with varying degrees of familiarity with Page's histologic classification scheme. In the study by Dupont and Page,3 the histologic diagnoses were performed by Page in collaboration with a pathol-

1264 CANCER February 25, 2993, Volume 71, No. 4

ogist colleague. In this report, they were performed by a current and former member of Page’s department who had extensive discussions with Page before starting the diagnosis process. In the study by London et al.,” two Harvard pathologists reviewed a training set of slides with Page before conducting their review. In the study by Palli et al.,19 the study pathologist made her diag- noses using Page’s published criteria but without re- viewing Page’s set of training slides. Thus, although the studies’ pathologists varied considerably in their experi- ence with Page’s classification scheme, their diagnoses led to risk estimates that are reassuringly consistent. This suggests that this system can be used by patholo- gists with extensive experience in breast pathology to make clinically useful diagnoses of benign breast dis- ease.

Comparison with Studies That Used Other Classification Criteria

Perhaps the greatest cause of confusion in Page’s classi- fication scheme for benign breast disease concerns his definition of AH. Page uses this term to denote rare, worrisome lesions that have some but not all of the features of carcinoma in sit^.^,^ However, the term AH has been used more loosely by many pathologists for many years. Black et a1.” devised an atypia score in which 13% of benign biopsy specimens obtained in the pre-mammographic era were classified as having some degree of atypia.’l This is approximately three times the prevalence of Page’s AH in a comparable series of be- nign biopsy specimen^.^ This higher prevalence is attrib- utable to the far weaker criteria needed to meet Black’s AH definition. Kodlin et a1.” found that women whose biopsy specimens had a Black atypia score of 3 had the same breast cancer risk as women with lower scores, whereas women with an atypia score of 4 had 2.6 times the risk of women with scores of 1 or 2. This finding is not inconsistent with those of investigators who have found greater cancer risks associated with Page’s AH because this lesion is more morphologically similar to cancer than lesions defined by Black atypia scores of 3 or 4.

The term ”atypical hyperplasia” also was used by the original pathologists from the BCDDP. (These diag- noses were made more than 7 years before Page pub- lished his rite ria.^) The precise definition used by these pathologists is uncertain. However, they classified 33 of the 322 biopsies given in Table 1 as containing AH, whereas we made this diagnosis in only 24 of these subjects. The BCDDP pathologists agreed with us in only 12 of these 24 cases. Thus, it appears that the BCDDP criteria were somewhat looser and consider- ably different from those used by Page to define AH.

Carter et aL6 found that women with AH diagnosed by the original BCDDP pathologists had a threefold in- crease in breast cancer risk. This finding of a somewhat lower risk associated with a lesion that differs substan- tially from Page’s AH is not inconsistent with those of authors who have used Page’s definition of this lesion.

Tavassoli and Norris” defined a lesion called ”atyp- ical intraductal hyperplasia.” The criteria for this lesion are similar to those of Page’s AH, ”. . . but emphasize specific cellular, architectural, and quantitative aspects of the proliferation.” Despite these minor histologic dif- ferences, they found that 9.80’0 of 82 women with atypi- cal intraductal hyperplasia had invasive breast cancer develop during an average follow-up of 12.4 years. The corresponding incidence of Page’s AH in our Nashville cohort3 is 9.6%, suggesting that the criteria of Tavassoli and Norris identify women whose cancer risk is similar to that of women who meet Page’s criteria for AH.

RosaiZ3 published a survey in which five leading pathologists were asked to independently classify 17 breast lesions according to the following diagnoses: normal, hyperplasia, AH, atypical lobular hyperplasia, and carcinoma in situ. They were asked to use the diag- nostic criteria that they use in their daily practice. There was wide variation in their diagnoses. Uniform agree- ment was not obtained for any lesion, and two or three of the five pathologists diagnosed AH for 70% of the lesions presented. This study vividly demonstrates that there is widespread variation in the criteria used to diagnose AH. Thus, it is important when considering the clinical meaning of this diagnosis to determine whether Page‘s criteria have been used.

This study does not support the findings of Hut- chinson et al.24 or Dupont and Page3 concerning calcifi- cation. Dupont and Page3 found that calcification in- creased breast cancer risk in women with PD. Hutchin- son et al.24 found that calcification increased breast cancer risk in women with fibrocystic disease, with a fivefold elevation in risk among those with both calcifi- cation and either epithelial hyperplasia or papillomato- sis. However, our study provides some support to the findings of Haagensen et al.,25 who found that breast cancer risk was elevated by cysts. Although cysts in the absence of PD did not increase breast cancer risk in our study, we found that cysts elevated breast cancer risk in women with PDWA or FH.

Conclusions

This study and those of London et al.,” Palli et al.,19 and Dupont and Page3 show that different levels of breast cancer risk among women with benign breast disease can be reliably distinguished on the basis of histologic criteria. Page’s histologic definitions4 can be

Proliferative Disease and Breast Cancer/Dupont et al. 1265

successfully learned by pathologists who are thor- oughly familiar with the wide and complex variations in histologic morphology found in the human breast. This study and two others3*18 have found AH to be as- sociated with a four- to fivefold increase in breast cancer risk, with a higher risk reported in another study." PDWA is associated with a 30-90% elevation in breast cancer risk in this and other ~tudies."'~*'~ Women for whom proliferative breast disease is diag- nosed require careful follow-up, with annual mammo- graphic screening being mandatory for women with AH. Women who undergo breast biopsy that reveals nonproliferative lesions can be reassured by the knowl- edge that their breast cancer risk is no greater than that of women of their age from the general p~pulat ion.~

References

1. Ernster VL. The epidemiology of benign breast disease. Epide- miol Rev 1981; 3:184-202.

2. Kelsey JL, Gammon MD. Epidemiology of breast cancer. Epide- miol Rev 1990; 12:228-40.

3. Dupont WD, Page DL. Risk factors for breast cancer in women with proliferative breast disease. N Engl J Med 1985; 312:146- 51.

4. Page DL, Anderson TJ, Rogers LW. Epithelial hyperplasia. In: Page DL, Anderson TJ. Diagnostic histopathology of the breast. New York: Churchill Livingstone, 1987:120-56.

5. Baker LH. Breast cancer detection demonstration project: five- year summary report. C A 1982; 32:194-225.

6. Carter CL, Corle DK, Micozzi MS, Schatzkin A, Taylor PR. A prospective study of the development of breast cancer in 16,692 women with benign breast disease. An2 ] Epidemiol 1988;

7. Page DL, Dupont WD, Rogers LW, Rados MS. Atypical hyper- plastic lesions of the female breast: a long-term follow-up study. Cancer 1985; 55:2698-708.

8. Page DL, Anderson TJ, Rogers LW. Diagnostic histopathology of the breast. New York Churchill Livingstone, 1987:43-50.

128:467-89.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

Breslow NE, Day NE. Statistical methods in cancer research. vol. 1. Lyon: International Agency for Research on Cancer,

Breslow NE, Day NE. Statistical methods in cancer research. vol 2. Lyon: International Agency for Research on Cancer, 1987:206. SAS Institute Inc. SAS/STAT user's guide, version 6. vol. 2. 4th ed. Cary, NC: SAS Institute Inc., 1990. McCullagh P, Nelder JA. Generalized Linear Models. New York: Chapman and Hall, 1989:21-47. GLlM Working Party. The generalized linear interactive model- ing system. Oxford Numerical Algorithms Group, 1987. Pregibon D. Logistic regression diagnostics. Ann Stat 1981;

SAS Institute Inc. SAS/IML software: usage and reference, ver- sion 6. Cary, NC: SAS Institute Inc., 1990. Fleiss JL. Statistical methods for rates and proportions. New York: John Wiley & Sons, 1981; 212-236. Rubin E, Visscher DW, Alexander RW, Urist MM, Maddox WA. Proliferative disease and atypia in biopsies performed for non- palpable lesions detected mammographically. Cancer 1988;

London SJ, Connolly JL, Schnitt SJ, Colditz GA. A prospective study of benign breast disease and risk of breast cancer. JAMA

Palli D, Rosselli del Turco M, Simoncini R, Bianchi S. Benign breast disease and breast cancer: a case-control study in a cohort in Italy. In t J Cancer 1991; 47:703-6. Black MM, Barclay TH, Cutler SJ, Hankey BF, Asire AJ. Associa- tion of atypical characteristics of benign breast lesions with sub- sequent risk of breast cancer. Cancer 1972; 29:338-43. Kodlin D. Winger EE, Morgenstern NL, Chen U. Chronic mas- topathy and breast disease. Cancer 1977; 39:2603-7. Tavassoli FA, Norris HJ. A comparison of the results of long- term follow-up for atypical intraductal hyperplasia and intra- ductal hyperplasia of the breast. Cancer 1990; 65:518-29. Rosai J . Borderline epithelial lesions of the breast. Am J Surg

Hutchinson WB, Thomas DB, Hamlin WB, Roth GJ, Peterson AV, Williams B. Risk of breast cancer in women with benign breast disease. J Null Cancer lrrst 1980; 6513-20. Haagensen CD, Bodian C, Haagensen DE Jr. Breast cancer risk and detection. Philadelphia: WB Saunders, 1981:70-5.

1980:191-279.

9:705-24.

61:2077-82.

1992; 267941-4.

Path01 1991; 15:209-21,