The association of mammographic density with ductal carcinoma in situ of the breast: the Multiethnic Cohort

Introduction It is well established that women with high mammographic density are at greater risk for breast cancer than are women with low breast density. However, little research has been done on mammographic density and ductal carcinoma in situ (DCIS) of the breast, which is thought to be a precursor lesion to some invasive breast cancers. Method We conducted a nested case-control study within the Multiethnic Cohort, and compared the mammographic densities of 482 patients with invasive breast cancer and 119 with breast DCIS cases versus those of 667 cancer-free control subjects. A reader blinded to disease status performed computer-assisted density assessment. For women with more than one mammogram, mean density values were computed. Polytomous logistic regression models were used to compute adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for two measurements of mammographic density: percentage density and dense area. Results Mammographic density was associated with invasive breast cancer and breast DCIS. For the highest category of percentage breast density (≥50%) as compared with the lowest (<10%), the OR was 3.58 (95% CI 2.26–5.66) for invasive breast cancer and 2.86 (1.38–5.94) for breast DCIS. Similarly, for the highest category of dense area (≥45 cm2) as compared with the lowest (<15 cm2), the OR was 2.92 (95% CI 2.01–4.25) for invasive breast cancer and 2.59 (1.39–4.82) for breast DCIS. Trend tests were significant for invasive breast cancer (P for trend < 0.0001) and breast DCIS (P for trend < 0.001) for both percentage density and dense area. Conclusion The similar strength of association for mammographic density with breast DCIS and invasive breast cancer supports the hypothesis that both diseases may have a common etiology.


Introduction
As a result of increasing early detection efforts, breast carcinoma in situ (CIS) constitutes more than 20% of newly diagnosed breast cancer cases in the USA [1]. Although breast CIS shares a number of risk factors with invasive breast cancer, and ductal carcinoma in situ (DCIS) of the breast is considered to be a precursor to some invasive breast cancers [2], it is not clear to what extent breast CIS and invasive breast cancer have the same etiology. Several case-control studies have confirmed that mammographic density is associated with risk for breast cancer [3]; women with high breast density have a three-fold to six-fold greater risk for developing breast cancer than women with low breast density [3,4]. Most studies of mammographic density included only invasive breast cancer cases [5] or combined invasive and CIS cases [6][7][8][9][10]. One study investigating breast CIS observed that DCIS was more likely to occur in mammographically dense areas [11], and another study reported an increase in breast hyperplasia or atypia/CIS in women with greater than 50% breast density [12]. This analysis examines the association between mammographic density and risk for breast DCIS in comparison with invasive breast cancer and breast cancer-free control subjects.

Study population
The data for this analysis were collected using a nested casecontrol study within the Hawaii component of the Multiethnic Cohort, which was established between 1993 and 1996 [13]. As described in detail elsewhere [4], all female members diagnosed with primary breast cancer between cohort entry and December 2000 were identified as potential cases (n = 1,587). A similar number of randomly selected control subjects (n = 1,584) who were not known to have breast cancer were frequency matched to the distribution of ethnicity and 5year age groups of the cases. Cases and controls with a previous diagnosis of breast cancer, a history of breast augmentation or reduction, and no mammogram were excluded. Approximately 13% of breast cancer cases and 4% of controls were ineligible primarily because of death or pre-existent breast cancer. Of the 1,396 cases eligible to participate, 52.6% responded to the mailings and gave full consent. Of the 1,500 eligible controls, 48.7% responded to the mailings and gave full consent. After removing women who did not have suitable mammograms, the final sample consisted of 607 breast cancer cases and 667 control subjects. The original cohort and the nested case-control study were approved by the Committee on Human Studies at the University of Hawaii.
All participants provided informed consent to participate in both studies.

Data collection
Details of the study procedures were reported previously [4]. In brief, information on demographics, medical history, reproductive behavior, hormone replacement therapy (HRT) use, and body mass index (BMI) were collected with an extensive questionnaire at entry into the cohort during the period from 1993 to 1996 [13]. As part of the nested case-control study, a one-page Breast Health Questionnaire (BHQ) was completed to elicit information on menopausal status, previous breast surgery, mammography, and HRT medications [4].

Mammograms
The mammographic films were retrieved from clinics located throughout the State of Hawaii using the authorization forms signed by the study participants. The original cohort study had no records on mammography use except for one item in the baseline questionnaire. At that time, 90% of Caucasian and Japanese women and 75% of Native Hawaiian women reported previous mammography [14]. Only craniocaudal views were digitized using a Kodak LS 85 Film Digitizer with a pixel size of 260 µm. If available, mammograms for every second or third year were scanned with the goal being to cover as wide a time period as possible for each woman. For cases, only mammograms taken before treatment for breast cancer were selected. However, the image of the contralateral breast taken at the time of diagnosis was used for five cases. The scanned images for both breasts were assessed for densities using Cumulus108 software [15] by one reader (GM), who was blinded to case status and time sequence of the mammograms. After the reader determined a threshold for the edge of the breast and for the edge of the dense tissue [15], the computer calculated the total number of pixels in the digitized image that constituted the total area and the dense area and computed the ratio between the two values as percentage density. Because readings for the right and left breast were very similar (correlation coefficient >0.90), we averaged the values for both to obtain one measure of total breast area, dense area and percentage density.
A random sample of 410 mammograms was read in duplicate to assess the reliability of the mammographic readings. The intraclass correlation coefficients [16] were 0.96 (95% confidence interval (CI) 0.95-0.97) for the size of dense area and 0.996 (95% CI 0.995-0.997) for the total breast area, resulting in an intraclass correlation coefficient for percentage density of 0.974 (95% CI 0.968-0.978).

Statistical analysis
To test for differences in covariate values across breast cancer cases and control subjects, we performed t tests for continuous variables and χ 2 tests for categorical variables. We used unconditional polytomous logistic regression modeling with the SAS software [17] to compute odds ratios (ORs) and 95% CIs for the risks for DCIS and invasive cancer associated with breast density [18]. All P values reported are two sided. Breast cancer cases were divided into CIS and invasive based on information provided by the state-wide Hawaii Tumor Registry, a member of the National Cancer Institute's Surveillance, Epidemiology and End Results program. Of the 125 breast CIS cases, 119 were classified as having DCIS.
We chose two measures of mammographic breast density as our exposure variables: size of the dense area and percentage density. Percentage density was grouped into four commonly used categories: <10%, 10-24.9%, 25-49.9%, and ≥50%.
The size of the dense areas was classified as follows: <15 cm 2 , 15-29.9 cm 2 , 30.0-44.9 cm 2 , and ≥45 cm 2 . Study participants were grouped into four categories: Japanese, Caucasians, native Hawaiians and others (mostly Filipinos). To maximize the number of participants per group, women of mixed ancestry -regardless of admixture -were assigned to one ethnic category according to the following priority ranking: native Hawaiian, Japanese, Caucasian, and, finally, other [13]. We created an HRT variable using the responses from the questionnaire at cohort entry and from the BHQ at enrollment into the breast density study. A comparison of the HRT information from the two questionnaires exhibited good agreement for overlapping years when both questionnaires reported HRT use. If a woman indicated that she had used HRT at any time but the write-in field in the BHQ was empty, we assigned the type of HRT from the cohort questionnaire completed at baseline. For the women with missing HRT type information (5.4%), we imputed the type based on hysterectomy status: estrogen only for women with a hysterectomy and combined therapy otherwise.
All models were adjusted for the following covariates that are known to be associated with breast cancer and mammographic density: mean age of all mammograms (continuous), ethnicity, BMI (<22.5, 22.5 to <25, 25 to <30, or ≥30 kg/m 2 ), parity (0-1, 2-3, or ≥4), age at menarche (<13, 13-14, or ≥15 years), age at first live birth (<21, 21-30, >30 years, or no children), menopausal status (pre-or postmenopausal), family history of breast cancer (breast cancer in a first-degree relative or no history), and HRT use (never, estrogen only, or estrogen + progestin). Tests for trend were performed by fitting a variable representing ordinal categories (described above) of percentage density or dense area.
We were also interested in comparing how well percentage breast density and size of dense area predicted invasive breast cancer and breast DCIS. We performed unconditional binary logistic regression and examined the area under the receiver operating characteristic (ROC) curve, which is a method used in sensitivity-specificity analyses. The ROC curve assesses the ability of the model to distinguish between two groups (for example, diseased and disease free). If the model is able to separate the two groups perfectly, then the area under the ROC curve is equal to 1; if the model performs no better than chance, then the area will be 0.5.

Results
Of all breast cancer cases, 119 (19.8%) were breast DCIS (Table 1). Japanese women had a greater proportion of breast DCIS than invasive breast cancer, and native Hawaiian women had almost twice the proportion of invasive cases compared with DCIS cases (P = 0.19). The mean age at diagnosis was similar for breast DCIS and invasive cases (P = 0.50). However, the breast DCIS cases had a greater proportion of postmenopausal women (89.9% versus 86.3% and 77.7% (for DCIS, invasive cases, and controls, respectively)) and had fewer children (2.2 versus 2.4 and 2.6 (for DCIS, invasive cases and controls, respectively)). Women with invasive breast cancer or breast DCIS had more mammograms than did controls (P < 0.0001). However, mean unadjusted age at first mammogram and last mammogram were not different across breast DCIS cases, invasive breast cancer cases, and controls (P = 0.32 and P = 0.66, respectively; results not shown). Although invasive breast cancer cases had a greater age-adjusted mean dense breast area than did breast DCIS cases (36.7 cm 2 and 34.9 cm 2 , respectively), the breast DCIS cases had higher percentage breast densities (38.2% and 36.5%, respectively) than did the invasive breast cancer cases because of their smaller mean total breast area.
Both percentage density and the size of the dense area were associated with breast DCIS and invasive breast cancer in our study (

Discussion
Our analyses revealed that mammographic breast density, as measured by percentage density and the size of dense area, is associated with breast DCIS as well as invasive breast cancer. The association was slightly weaker for breast DCIS than for invasive breast cancer. Comparisons of areas under the ROC curve indicated that breast density and dense area performed equally well in distinguishing breast DCIS and invasive breast cancer cases from controls. The ORs for breast DCIS in the present study were not as large as those reported in a study of a cohort of Canadian women [12], which estimated relative risks between 7 and 12 for detecting atypia/CIS in biopsy specimens from women with more than 50% density. However, the cell sizes within the strata of density were small in that Other studies of breast DCIS support the idea that there is a relation between in situ lesions and mammographic density. An investigation of 28 mammograms confirmed that breast DCIS occurs to a greater extent in areas of the breast that exhibit high mammographic density [11]. The relative risk for a second breast cancer was found to be three times higher among women with primary DCIS who had mammographic densities of 75% and greater compared with women with less than 25% density [20].
Although it is known that some breast DCIS lesions will develop into invasive breast cancers [21], and that DCIS is often found near invasive breast cancer lesions [22], it remains problematic to distinguish DCIS lesions that may progress to invasive breast cancer from those that may not. Given the growing number of breast CIS cases with increasing mammography screening [1], this question has important implications when decisions must be made regarding the aggressiveness of treatment [23]. A comparison of risk factors for CIS (ductal and/or lobular) and invasive breast cancer found mixed results. Established invasive breast cancer risk factors shown to be consistently associated with breast CIS include family history of breast cancer [24][25][26][27][28][29][30], low BMI among premenopausal women [24,29,30], and nulliparity [24,28,30,31]. However, other invasive breast cancer risk factors such as early age at menarche, late age at menopause, increased endogenous estrogen levels, and alcohol intake were associated with breast CIS in some studies [26,27,29,32], but not in others [24][25][26][27]30,33].
Our study and other reports mentioned above indicate that mammographic density is associated with breast CIS and with invasive breast cancer. However, because some studies describe differential effects for some risk factors, more research needs to be done. Studies with larger samples of breast CIS must be performed to assess whether factors such as postmenopausal obesity, growth factors (insulin-like growth factor-I), exogenous hormones, and cell proliferation biomarkers can help in elucidating the association between breast density, breast CIS, and progression to invasive breast cancer. New molecular techniques may have the ability to identify factors that are responsible for the progression from DCIS to invasive cancer [2].
A limitation of the present study is the lack of data on the frequency of mammography use. Although this population has high screening rates [14], we cannot rule out bias toward the null in estimates of DCIS risk because of the possibility of undetected breast DCIS among controls. We also had relatively low participation rates, although a comparison of participating and eligible women revealed that they had very similar characteristics [4]. Our assessments of HRT and BMI are limited because we had to rely on self-report, and assumed that their values remained constant between the time they were reported and when the study was completed, but an examination of BMI from a follow-up questionnaire five years after cohort entry showed that the mean BMI changes by only 0.50 kg/m 2 during that time. Therefore, differences in BMI are unlikely to change the results materially. The use of BMI in categories is unlikely to have confounded our results because analyses with continuous values of BMI gave nearly identical results. We had limited power to estimate the risk for DCIS; there were nearly four times as many invasive breast cancer cases as breast DCIS cases. Nevertheless, the cohort design, the multiple mammograms, and the frequent mammography use in this population must be considered strengths of this project. The high rate of mammography participation decreases the probability that a large number of cancers were missed in this population.

Conclusion
In the present study similar patterns of association for mammographic density with DCIS and invasive breast cancers add to the growing body of evidence that certain breast CIS and invasive breast cancers may share etiologic factors. At the same time, it appears likely that factors that have not yet been identified may influence the effects of breast density and other known breast cancer risk factors in the progression of breast CIS to invasive breast cancer.