Comparison of mammographic densities and their determinants in women from Japan and Hawaii

Authors


Abstract

Breast cancer incidence increases considerably in women who migrate from Japan to the United States. Based on the hypothesis that mammographic density in healthy mammograms reflects differences in breast cancer risk, we compared mammographic density in 3 groups of women at different levels of risk: Caucasian and Japanese women in Hawaii and Japanese women in Japan. In a cross-sectional design, pre- and postmenopausal women without a history of breast cancer and with a mammogram free of suspicious lesions were recruited in mammography clinics and completed a self-administered questionnaire. Cranio-caudal mammograms were scanned into a computer and the densities measured using a computer-assisted method. Statistical analyses included ANOVA and multiple linear regression. Breast size among women of Japanese ancestry was similar in Hawaii and Japan but 50% smaller than that among Caucasian women. Dense areas were smallest among women in Japan, intermediate among Japanese women in Hawaii and largest among Caucasian women. Percent densities were greater in Japanese women than Caucasian women because of the larger breast sizes in Caucasians. However, percent densities were significantly higher among Japanese women in Hawaii than in Japan. These results indicate that the size of the total breast differs primarily by ethnicity and the size of the dense areas differs mainly by place of residence. Therefore, when comparing ethnic groups with distinct physical proportions, the absolute size of the dense areas appears to be a better measure of breast cancer risk than the relative density. © 2002 Wiley-Liss, Inc.

Breast cancer incidence among women in Japan is much lower than among women of Japanese ancestry living in the United States. Age-adjusted (to the world standard population) incidence rates were 30.8 cases/100,000 in Japan1 compared to 78.3 and 97.1 cases/100,000 among women of Japanese and Caucasian ancestry in Hawaii, respectively (unpublished report for 1990–1992 from the Hawaii Tumor Registry). Reasons for the discrepant breast cancer rates are not well understood, but the fact that risk in migrants increases over 2 or 3 generations2, 3 suggests that environmental factors, including diet,4 may be determinants of risk in addition to the well-established reproductive factors, adult adiposity, weight change and height.5 Mammographic density patterns, which refer to the distribution of fat, connective and epithelial tissue in the healthy female breast, have been related to breast cancer risk.6 A high percentage of dense parenchyma on mammographic images appears to confer a 4-fold risk of developing breast cancer.7 Immigration from Japan to Hawaii occurred primarily during 1868–1924,8 when more than 200,000 Japanese came to Hawaii. We hypothesized that Japanese women in Hawaii are more likely to have a dense parenchymal pattern than women in Japan. The primary objective of our cross-sectional study was to compare mammographic density characteristics as an indicator of breast cancer risk among women at different levels of risk: women in Japan, women of Japanese ancestry in Hawaii and Caucasian women in Hawaii. In addition, we investigated the determinants of mammographic density in these 3 groups.

MATERIAL AND METHODS

Study Population

Recruitment sites were mammography clinics in Honolulu, Hawaii, and Gifu, Japan. Eligibility criteria included participation in mammography screening, a normal mammogram [BI-RADS9 categories 1–3], no serious medical conditions, no history of breast surgery and no previous history of cancer. Women with palpable lumps or any other medical indication for a diagnostic mammogram were excluded from the study. Women taking oral contraceptive pills, hormone replacement therapy or any other hormone preparations were eligible to participate; but we recorded the details of their medications. All women completed a self-administered questionnaire that inquired about demographic, anthropometric, medical and reproductive characteristics. The Institutional Review Board at the Gifu University School of Medicine and the Committee on Human Studies at the University of Hawaii approved the study protocols. All participants gave written informed consent.

The details of the recruitment in Hawaii have been described elsewhere.10 Due to legislatively mandated insurance coverage of mammography screening and high insurance coverage, the majority of women residing in Hawaii have access to screening services. Between 70% and 80% of women 40 years and older reported participation in mammography screening.11 Women were recruited at 5 mammography clinics. Flyers describing the study were mailed with the appointment reminder or handed to the women at the time of their appointment. Because the clinics did not distribute flyers to all eligible women, neither the total number of potential subjects nor their ethnic distribution is known. Interested women sent their address to the mammography clinic or to the Cancer Research Center of Hawaii. The informed consent form, mammogram release form and questionnaire were then mailed to potential study participants. All mammography clinics were accredited by the Food and Drug Administration, used dedicated mammography units and high-speed film and participated in regular training and quality control.

Women in Japan were recruited at Gihoku General Hospital, located in the southern part of Gifu Prefecture. The hospital provides care to a primarily middle class population residing in its surroundings and has been conducting mass screening campaigns for breast cancer since the early 1980s. The prefecture mails letters to women 40 years and older every 2 years and invites them to receive a mammogram at the outpatient clinic. Therefore, Gihoku Hospital, unlike other institutions in Japan, provides a large number of mammograms to asymptomatic women. In response to invitation letters from the prefectural government, approximately 20% of women residing in the catchment area of the hospital receive a screening mammogram. During the year 2000, the hospital performed 3,043 mammograms and diagnosed 8 cases of breast cancer, a detection rate of 2.6/1,000 mammograms. At the screening, a research nurse explained the study to women and obtained informed consent from interested women. Among women attending the Gifu screening clinic, 73.8% agreed to participate. Mammographic images at Gihoku Hospital were taken on a model ESP 200 Acoma (Tokyo, Japan). Films were mailed to Hawaii and digitized on the same scanner as the images from the Honolulu clinics.

Mammographic Density Assessment

After assigning a unique identifying number to each subject, we scanned the cranio-caudal mammograms into a computer at the Cancer Research Center using a Cobrascan CX612T X-ray digitizer (Radiographic Digital Imaging, Compton, CA) with a resolution of 150 dpi. With the help of a software program for image manipulation, all personal identifying information was deleted. For density determinations, the left mammograms were chosen. We used the method described by Byng et al.12 in Toronto and modified by Ursin et al.13 in Los Angeles. Basically, it classifies the density of breast tissue by counting pixels within the regions identified as representative of radiographically dense tissue. The software uses 256 different gray levels, with 255 being the darkest value and 0 the lightest. In the digitized mammographic image displayed on the computer screen, the reader selects a threshold value (gray scale on the computer screen) that best distinguishes the breast from the dark background. An edge detection computer program (algorithm) is used to outline the breast and to measure its total area (effectively in terms of computer screen pixels). Subsequently, the reader identifies a second threshold value, the gray value that best identifies the edges of the mammographically dense areas within the breast parenchyma. The number of pixels in these areas is then counted by the computer, and the fraction of pixels in the breast parenchyma that are located in the dense areas (percent mammographic density) is calculated. Pixel values were divided by 3,487.5 (1 cm = 59.055 pixels at 150 dpi scanning resolution) to transform them into square centimeters. All density assessments were performed by 1 of the authors (G.M.). Mammograms were read in batches, with a similar number of mammograms from each study location in every session.

The problem of observer variation in the classification of mammographic patterns has been addressed. Comparing 100 mammograms, Boyd et al.14 reported that 20 of 28 pairs of readers agreed on >80% of readings. Byng et al.12 obtained a correlation coefficient of 0.91 between the computer-aided method and the amount of mammographic densities determined using a subjective 6- category classification and high between-observer (r > 0.9) and within-observer (r > 0.9) correlations. Researchers at the University of Southern California compared their method to the amount of mammographic densities determined using Wolfe's outlining method and found a correlation of >0.85.13 We performed mammographic density assessment for 13.5% of all mammograms in duplicate. Type 3 intraclass correlation coefficients (ICCs), according to Shrout and Fleiss,15 were 0.962 [95% confidence interval (CI) 0.939–0.976] for the size of the dense areas and 0.997 (95% CI 0.996–0.998) for the total breast area.

Data Management and Statistical Analysis

Information from the questionnaires and the mammographic measurements was combined into one database. Body mass index (BMI) was calculated as weight (kg) divided by height (m2). Several variables had to be converted into a common format. If appropriate, remedial measures such as log transformation or creation of a categorical variable were taken to normalize variables. The SAS statistical package for PC (SAS Institute, Cary, NC), version 8.2, was used for all data management and analyses. First, we evaluated differences between population groups by ANOVA. Then, we computed simple regression models of mammographic characteristics with demographic, anthropometric and reproductive variables to explore determinants of mammographic appearance. Using multiple linear regression with the stepwise selection process,16, 17 we developed 2 separate models, to describe the determinants of the size of the dense areas and of percent density. To adjust for confounders when comparing the means of mammographic densities, the least-squares means, or population marginal means, of mammographic measures were estimated by ethnicity and place of residence in the PROC GLM procedure.18 These predicted values were the expected means by place of residence and ethnicity after controlling for confounders.

RESULTS

We observed significant differences in anthropometric and reproductive characteristics between the 3 groups of women (Table I). However, women of Japanese ancestry were very similar in mean height and weight independent of place of residence. All women in Hawaii had received more education and reported earlier menarche, fewer live births and later age at first live birth than women in Japan. Previous hysterectomies and ovariectomies, as well as a family history of breast cancer, were much more commonly reported in Hawaii than in Japan. Use of hormone replacement therapy was considerably higher among postmenopausal women in Hawaii than in Japan.

Table I. Characteristics of the Study Population
CharacteristicsJapanese in GifuJapanese in HawaiiCaucasians in Hawaiip value1
  • 1

    For continuous variables, ANOVA was used; for categorical variables, the χ2 test and likelihood ratio were used.

Number210146167
Age (years)50.1 (9.1)52.5 (9.7)53.7 (9.9)0.0011
BMI (kg/m2)23.2 (3.0)23.1 (3.9)25.0 (4.9)<0.0001
Height (cm)154.9 (5.6)156.4 (5.4)164.3 (6.8)<0.0001
Weight (kg)55.6 (8.0)56.5 (10.1)67.4 (13.5)<0.0001
Years of education11.9 (2.0)16.2 (2.8)16.2 (3.2)<0.0001
Family history of breast cancer (%)3.312.312.60.0014
Age at menarche (years)13.3 (1.5)12.2 (1.5)12.5 (1.6)<0.0001
Age at first live birth (years)24.9 (2.9)27.6 (5.1)26.1 (5.2)<0.0001
Number of children2.3 (0.8)1.9 (1.4)1.6 (1.6)<0.0001
History of contraceptive pill use (%)6.763.079.0<0.0001
History of ovariectomy (%)8.124.021.6<0.0001
History of hysterectomy (%)10.019.918.00.017
Number of premenopausal women10770750.51
Number of postmenopausal women1037692 
Age at menopause (years)48.7 (4.4)46.8 (5.9)47.5 (5.7)0.06
History of estrogen use (%)9.475.478.8<0.0001

Breast size among women of Japanese ancestry was similar in Hawaii and Japan (Table II) but 50% smaller than among Caucasian women. This relation changed little after adjustment for anthropometric and reproductive factors. Dense areas were smallest among women in Japan, intermediate among women of Japanese ancestry in Hawaii and largest among Caucasian women. The difference became statistically significant after adjustment for variables related to density. The adjusted means for the 2 groups in Hawaii were very similar in magnitude. Unadjusted percent densities were greater among women of Japanese ancestry than among Caucasian women because of the larger breast sizes among Caucasians. However, percent densities were significantly higher among Japanese women in Hawaii than among women in Japan. When percent densities were classified in categories, 23.3% of the Japanese women in Hawaii were in the lowest group (25% density or less), whereas 35.7% of the women in Japan were in that group. At the other extreme, 18.5% of Japanese women in Hawaii were in the highest group (50% density or more) compared to 11.9% in Japan. Adjustment for age, menopausal status, BMI, number of children and history of ovariectomy resulted in slightly higher percent densities in Caucasian women and lower percent densities among women in Japan.

Table II. Mammographic Measures by Ethnicity and Place of Residence
CharacteristicJapanese in GifuJapanese in HawaiiCaucasians in Hawaiip value
  • 1

    Logarithmic (area) and square root (dense) transformation was used for significance testing because of non-normal distribution.

  • 2

    Adjusted for age, menopausal status, number of children, BMI and history of ovariectomy.

Breast size (cm2)1    
 Mean55.6956.4187.77<0.0001
 (SD)(20.86)(24.79)(43.96) 
 Adjusted mean254.6653.2671.43<0.0001
Dense area (cm2)1    
 Mean16.6919.1620.840.105
 (SD)(9.09)(11.18)(15.76) 
 Adjusted mean214.7917.6518.640.003
Percent density    
 Mean31.336.428.30.0001
 (SD)(14.4)(15.9)(20.3) 
 Adjusted mean229.435.731.3<0.0001

Simple regressions with a number of breast cancer risk factors (Tables III, IV) showed several significant associations. Whereas Japanese ancestry was related to higher percent densities, residence in Japan, age, menopause, age at first live birth 30 years or younger, having 3 or more children, history of hysterectomy or ovariectomy and a higher BMI were inversely correlated with percent densities. In a multiple linear regression model, the total explained variance was close to 42% and included all of the above variables except age at first live birth and history of hysterectomy. The total variance explained was only 11% for the size of the dense areas (Table IV). Residence in Japan, age and BMI were inversely related to the size of the dense areas. Age at menarche, family history of breast cancer, having at least 1 child, age at menopause and hormone replacement use were not related to any mammographic measure.

Table III. Determinants of Percent Density
VariableSimple regressionStepwise multiple regression
Parameter estimateStandard errorPr > |t|Parameter estimateStandard errorPr > F
  • 1

    Logarithm of variable was used for modeling. ns, not significant.

Japanese ethnicity5.151.590.00134.411.530.0041
Residence in Japan−0.741.530.6278−6.351.45<0.0001
Age (years)−0.660.07<0.0001−0.310.090.0005
Postmenopausal status−12.401.40<0.0001−5.561.710.0012
Age at menarche1−5.846.180.3457 ns 
Age at first live birth ≤ 30 years−6.121.620.0002 ns 
At least 1 child−3.071.970.1198 ns 
Three or more children−7.661.59<0.0001−4.671.280.0003
Family history of breast cancer−0.732.650.7842 ns 
History of ovariectomy−6.821.980.0006−2.821.620.0820
History of hysterectomy−7.072.060.0007 ns 
BMI (kg/m2)−2.210.16<0.0001−1.970.15<0.0001
Age at menopause (years)−0.230.180.1899 ns 
Use of hormones0.101.880.9593 ns 
Variance explained na  0.4182 
Table IV. Determinants of Dense Area1
VariableSimple regressionStepwise multiple regression
Parameter estimateStandard errorPr > |t|Parameter estimateStandard errorPr > F
  • 1

    Logarithm of variable or square root (dense area) was used for modeling. ns, not significant.

Japanese ethnicity−9.487.700.2185 ns 
Residence in Japan−15.537.300.0338−20.818.050.0100
Age (years)−1.990.36<0.0001−1.790.37<0.0001
Postmenopausal status−36.387.01<0.0001 ns 
Age at menarche1 (years)−32.0829.580.2786 ns 
Age at first live birth ≤ 30 years−27.867.760.000413.768.450.1042
At least 1 child−17.699.410.0606 ns 
Three or more children−23.637.700.0023 ns 
Family history of breast cancer−9.9312.680.4339 ns 
History of ovariectomy−21.949.560.022114.449.510.1294
History of hysterectomy−22.849.930.0219 ns 
BMI (kg/m2)−3.480.88<0.0001−3.190.860.0002
Age at menopause (years)−1.050.940.2648 ns 
Use of hormones12.529.930.2083 ns 
Variance explainedna 0.1057 

Our results indicate that the size of the total breast differed primarily by ethnicity and the size of the dense areas differed mainly by place of residence. Because percent densities depend on the size of the breast and the dense areas, they were significantly different by place of residence and by ethnicity. Comparison of women in Japan with descendants of women who migrated to a high-risk environment several generations ago showed nearly unchanged anthropometric characteristics but larger dense areas and higher percent densities, reflecting a higher breast cancer risk. Mean breast area was approximately the same in our sample of women from Gifu as in the women living in Hawaii, though >90% of these women were born in Hawaii and a large proportion are third- or fourth-generation migrants. Comparing women of Japanese and Caucasian ancestry was strongly confounded by anthropometric characteristics and, in particular, by the larger breast sizes in Caucasian women.

DISCUSSION

This study compared mammographic densities in women of the same ethnicity who reside in countries with different levels of breast cancer risk. We found significantly lower percent densities among women of Japanese ancestry living in Japan than among third- and fourth-generation Japanese migrants in Hawaii. Given the fact that breast sizes were approximately the same by place of residence, the differences in the absolute size of the dense areas and in the percent densities were of similar magnitude: 61,000 vs. 51,000 pixels and 35.7% vs. 29.4% for Japanese women in Hawaii vs. Japanese women in Japan, respectively. Caucasian women in Hawaii had larger dense areas than all Japanese women, but, due to the much larger breast areas, percent densities were smaller than in Japanese women.

Few previous studies have investigated mammographic densities among Japanese women. A case-control study showed that women with breast cancer were nearly 7 times as likely to be classified as one of the high-risk Wolfe categories than controls.19 Normal premenopausal Japanese women were found to have significantly more favorable mammographic patterns (Wolfe grades20) than British women.21 Results from a mammography screening study using Wolfe's classification scheme suggest a very low prevalence of the DY (mammary dysplasia) patterns.22 A small cross-sectional study in Japan found that the proportion of women with >50% mammographic densities was considerably lower in the Japanese study (19.5%) than in a Canadian report7 (35.9%).23 In comparison to our studies among women of different ethnic groups in Hawaii,10, 24 this study found very similar determinants of mammographic densities.

The differences in mammographic densities between the 2 populations may be a result of different hormonal patterns. As demonstrated repeatedly,25, 26, 27 Asian women appear to have lower levels of circulating estrogens than Caucasian women. In one study, the adjusted mean values of estrone and estradiol were 162% (p < 0.0001) and 152% (p < 0.0001) higher, respectively, in U.S. women than in Singapore Chinese women.28 An association of mammographic densities with estrogens and progestogens levels is supported by several observations: hormone replacement therapy has increased breast densities in several studies;29, 30, 31 suppression of ovarian function through a gonadotropin-releasing hormone agonist (GnRHA) reduced mammographic densities after 1-year follow-up;32 after stopping treatment with the GnRHA, the reduced percentages of mammographic densities experienced during the 24 months of treatment returned to baseline;33 and tamoxifen treatment improved mammographic densities in a number of interventions.34, 35, 36 IGF-I was proposed as another hormone that may influence mammographic densities37 and act as a mediator of nutritional effects on breast cancer risk.38

We recognize that the sampling of study participants in Japan has weaknesses because mammography participation in Japan is lower than in Hawaii. The women who attended mammography screening in Gifu may not have been representative of the general population with respect to breast cancer risk. The fact that they decided to receive a mammogram may be related to a true or to a perceived elevated breast cancer risk. However, we were careful to exclude women with suspicious lesions in their mammograms. To a smaller degree, the same selection bias was probably also present in Hawaii despite self-reported mammography participation rates of 70% and higher.11 Unfortunately, an ideal research design of comparing a truly random sample of women from low- and high-risk countries is difficult. In low-incidence countries, population-based mammography screening programs do not exist and would not be recommended because the low positive predictive value in such a population could not achieve a favorable risk/benefit ratio, which is required to justify a screening program.39

Although national breast cancer incidence rates are available, it is difficult to compare the current breast cancer risk in the 3 populations. First, subjects were relatively young; and as they age, breast cancer risk, especially in Japan, increases significantly.1, 40 Age-adjusted breast cancer incidence rates rose from 13.5/100,000 women in 1975 to 24.3/100,000 women in 198540 and to 100,000 women in 1995, with increases in all women 40 years and older. Second, despite the mass screening for breast cancer in Gifu, it is likely that the women who actually attended the screening sessions are at higher risk than the population at large. Therefore, our samples in Hawaii and Gifu are unlikely to represent a 3-fold range of breast cancer incidence as suggested by breast cancer incidence rates cited above (30.8, 78.3 and 97.1 cases/100,000 in Japan and among Japanese and Caucasian women in Hawaii, respectively). According to large mammographic density studies,7, 41 a 3-fold difference in breast cancer risk would require substantially larger differences in percent densities than we observed in this study.

Our major findings, little change in breast size and BMI as a result of migration but considerable difference in mammographic density, suggest that the size of the dense areas is a better indicator of breast cancer risk across ethnic groups than percent densities. The majority of breast cancer research on mammographic densities has been performed among Caucasian women in North America.6 Within that group, percent densities serve as a useful indicator of breast cancer risk, with a 4- to 6-fold higher risk of developing the disease for women with percent densities of 75% and greater compared to the lowest density category. However, in ethnic groups such as Japanese with distinctly different physical proportions, relative density does not appear to be a good measure for comparison. The size of the dense areas, representing in some way the extent of breast tissue at risk, may offer a better method of classifying women from different ethnic groups into low- and high-risk categories.

Acknowledgements

We greatly appreciate the time and effort contributed by the participating women. Thanks to our graduate research assistant Ms. Y. Takata for assistance in data management and analysis. The research in Japan was supported by a grant from the National Cancer Institute (Planning a Hawaii/Japan Mammographic Density Study, R03 CA 81620). Data collection in Hawaii was funded by the U.S. Army Medical Research and Material Command (DAMD17-96-1-6284).

Ancillary