A Pilot Study of Mammographic Density Patterns Among Japanese Women

Mammographic density patterns, which refer to the distribution of fat, connective, and epithelial tissued in the healthy female breast, have been shown to be related to breast cancer risk. We used a quantitative method to assess mammographic densities in 41 mammograms from women in Japan without a diagnosis of breast cancer. Information about reproductive behavior and family history for breast cancer was available from a questionnaire. The statistical analysis applied Spearman correlation coefficients and multiple linear regression. The breast size as measured on the cranio-caudal view of the mammogram was approximately 12% larger, the size of the dense areas was 20% smaller, and the mean percent mammographic densities were 30% greater among premenopausal than among postmenopausal women. We found a strong relation between age at menarche and mammographic densities in premenopausal women and significant associations for age, family history of breast cancer, and age at menopause with mammographic densities in postmenopausal women. These preliminary data will be used to plan a future study that will compare mammographic density patterns and the relative importance of dietary, reproductive, and anthropometric factors between women in Japan and in the United States. J Epidemiol, 1999 ; 9 : 73-77

Breast cancer incidence among women in Japan is much lower than among women living in the United States 1).During 1988 to 1992, breast cancer incidence rates (Table 1) were two to three fold higher among women of Japanese ancestry in the United States than among women in Japan.Whereas changes in the age at menarche and reproductive behavior 2,3) partially explain the differences, the discrepant rates are not completely understood.The fact that breast cancer risk in Asian women who migrated to the United States approaches the risk among U.S. Whites after two or three generations 4) suggests that environmental factors and in particular diet may be important determinants of breast cancer risk.
Mammographic density patterns, which refer to the distribution of fat, connective, and epithelial tissue in the healthy female breast, have been shown to be related to breast cancer risk.A high percentage of dense parenchyma on mammographic images appears to confer a fourfold risk to develop breast cancer 5).Three different methods have been used to assess mammographic densities: Wolfe 6) first described four parenchymal patterns of increasing mammographic densities [N(radiologically lucent), Pl, P2, and DY(diffuse or nodular densities)], which was followed by a method 7) of measuring the extent of mammographic densities with a planimeter .More recently Byng 8)   The objectives of this pilot study were to apply the quantitative method to a sample of mammograms from Japanese women and to explore the relation between reproductive factors and mammographic densities in this population.

MATERIALS AND METHODS
We obtained 41 mammograms of women living in the Okayama area.Mammography was performed at Kawasaki Medical School in Kurashiki during 1997 using a Toshiba MGU-10C mammography unit and Kodak Min-R X-ray films.The background conditions of the subjects' breasts were similar to those of women who participate in screening programs.Although routine mammography screening was not conducted at Kawasaki hospital during the time of this study, physicians at Kawasaki Medical School have made it a rule to perform mammograms on women who visit the clinic to receive breast screening and who complain of symptoms concerning breast diseases and/or are postmenopausal.Women who, as a result of a mammogram or an echogram, were suspected to have breast diseases, such as cancer, benign tumors, and fibrocystic disease (mastopathy), were excluded from the study.Therefore, none of the participants had a diagnosis of breast cancer or a suspicious lesions at the time of the mammogram.Information about reproductive behavior and family history for breast cancer was abstracted from a questionnaire that the women completed at the time of the mammogram.In order to maintain complete confidentiality, we did not record names and other personal identifiers for this study.
Mammograms for all subjects were mailed from Japan to Hawaii.As a backup, a copy of the original mammogram was kept in Japan.After scanning the left cranio-caudal mammographic films into a PC using a Cobrascan CX312 X-ray digitizer from Radiographic Digital Imaging, Compton, CA, we performed computerized mammographic density assessment.We used a similar method as described by Byng et al. 8) in Toronto, which was modified by researchers at the Department of Preventive Medicine, University of Southern California at Los Angeles 9).The system uses 256 different grey levels, with 256 being the darkest value and 0 the lightest.In the digitized mammographic image displayed on the computer screen, the reader first draws the outline of the breast (using an outlining tool) and then searches for the best threshold gray level value X where all pixels with values above X are considered to represent mammographic densities.The pixel count corresponding to the area colored within the outline of the breast is determined by the computer as is the total area within the outline of the breast.The proportion of the breast with densities is calculated as the ratio of the colored area to the total area of the breast.The density assessment was performed by two readers (G.M. and L.M.), who were blinded as to the study participants' characteristics.The individual results for the two readers were averaged.The correlation coefficient between the measurements of the two readers was 0.97.
We calculated means and proportions by menopausal status for variables of interest and applied Student's t-tests and X2-tests 16) to test for differences.We replaced a small number of missing values for breast feeding (3 women), age at first full-term birth (6 women), age at menarche (2 women), and parity (7 women) with the mean of the study population.
To determine associations between mammographic densities and the reproductive variables, we computed Spearman s correlation coefficients and applied multiple linear regression models 16) using the stepwise selection method.Because the distribution of the dense area variable was not normal, we used the natural logarithm of the dense area in the regression analysis.For the same reason, we created a categorical variable for age at menarche.The SAS* statistical package for PC (SAS Institute, Cary, NC) was used for all data management and analysis.

RESULTS
Among the 41 women, the mean age was 51 years with a range of 29 to 76 years.Eighteen women reported premenopausal and 23 women reported postmenopausal status.
Although the difference was not statistically significant, the mean age at menarche (Table 2), 13.3 years for premenopausal vs.14 years for postmenopausal women demonstrated the decreasing secular trend among younger women all over the world.Parity, mean age at first full-term birth, and lactation history were similar for the two groups.One third of women reported breast feeding at least one child.Only four women reported a family history of breast cancer and three women reported a previous breast biopsy.While the breast size as measured on the cranio-caudal view of the mammogram (Table 2) was approximately 12% larger in postmenopausal than in premenopausal women, the size of the dense areas was 20% smaller in postmenopausal than in premenopausal women.The mean percent mammographic densities were one third less in postmenopausal women as compared to premenopausal women.Among premenopausal women (Table 3), we detected a strong relation between age at menarche and percent mammographic densities and a weaker association between a family history of breast cancer and percent mammographic densities.For postmenopausal women, however, age at menopause and age had the strongest association with percent mammographic densities.In addition to a family history of breast cancer, age at menopause and age were slightly related to the size of the dense areas.
In a multiple linear regression model including all women, 26% of the variation of the logarithm of the dense area was explained by age (partial r2= 16%, p = 0.003) and a family history of breast cancer (partial r2 = 10%, p = 0.04).While age was negatively related to the size of the dense areas, a family history of breast cancer predicted larger dense areas.With percent mammographic densities as dependent variable, the explained variation increased to 38%.Age (partial r2 = 25%, p = 0.001) and menarche at age 13 years or older (partial r2 = 13%, p = 0.009) were both related inversely to percent mammographic densities.

DISCUSSION
In this pilot study, we found a significant difference in mammographic densities among pre-and postmenopausal women, a strong correlation between age at menarche and mammographic densities among premenopausal women, and significant associations for age, family history of breast cancer, and age of menopause with mammographic densities among postmenopausal women.These findings are consistent with the fact that age at menarche and age at menopause, as well as family history of breast cancer are established risk factors for breast cancer 17) that have also been confirmed in Japan 18. 19).Age at menopause was not significant in our regression model, probably because of a strong relation with age (r8 = 0.56, p = 0.007), a finding that is difficult to explain.The decrease in mammographic densities observed with increasing age has been described in most mammographic density studies and is thought to be the consequence of the fatty involution of the female breast related to declining endogenous estrogen levels 20).Parity and age at first full-term pregnancy that were related to mammographic densities in previous reports 11, 12,14) did not show significant correlations with mammographic density in this study, but given the small number of women and the limited variation in these variables, the results are not unexpected.
This small pilot study had several limitations.Because the subjects were not randomly selected from the population, but had a medical indication for a mammogram, it is very likely that the women in this study were at higher than average risk for breast cancer.In comparison with a mammography study from Tokushima, Japan 21) using Wolfe's classification scheme, the women in our study appear to have denser mammographic patterns.If we consider the N1 and P1 pattern in the Wolfe scheme equivalent to a density of 25% or less, fewer women aged 45 to 59 years had a low density pattern in our study than in the Tokushima report.This could be a result of our small sample size, the selection of more high risk women, and differences in menopausal status and age.On the other hand, the proportion of women with more than 50% mammographic densities was considerably lower in our study (19.5%) than in a Canadian report5) (35.9%) that used a quantitative classification method for mammographic density assessment.
Unfortunately anthropometric information, such as weight and body mass index, which are important determinants of mammographic densities 22), was not available for this study.The strength of this project is that we were able to apply the quantitative mammographic assessment method to a sample of Japanese mammograms for the first time and demonstrate associations with risk factors for breast cancer in this population.To investigate the hypothesis that Japanese women in Hawaii are more likely to have a dense parenchymal pattern than women in Japan, we are planning a comparative study that will collect anthropometric and dietary information among a larger sample of Japanese women and compare them with subjects in an ongoing study in Hawaii.

Table 1 .
Breast Cancer Incidence Rates for Japanese Women in Different Geographic include Locations (Aee-adiusted to the World Standard Population)1).

Table 2 .
Selected Characteristics of the Study Population in Okayama by Menopausal Status.
* Depending on variable type , Student's t-test or X2-test was used.*Spearman Correlation Coefficients