Download Studying Mammographic Density: Implications for Understanding

EDITORIALS
Studying Mammographic Density: Implications for
Understanding Breast Cancer
Celia Byrne*
Several epidemiologic studies (1-3) have reported the magnitude of the risk associated with increased mammographic density to be between fourfold and sixfold, exceeding that of most
other predictors of breast cancer. In these studies, this increased
breast cancer risk was not explained by other traditional breast
cancer risk factors. Despite the level of evidence of an independent association with breast cancer and the large number of
women who are having routine mammographic screening, mammographic density is perhaps the most undervalued and underutilized risk factor in studies investigating the causes of breast
cancer. Many studies of breast cancer risk have not included
measures of mammographic density either because of practical
reasons, such as difficulty in obtaining mammographic images
and in assessing features in a systematic manner, or because of
misconceptions stemming from the interpretation of early studies of mammographic features. The fact that this strong predictor
of risk has been overlooked in many studies of breast cancer
seems particularly unfortunate because of the profound implications to both cancer screening and etiology that attention to these
factors may have. Thus, to fully comprehend the causes of breast
cancer, we need to understand the role of breast density in the
disease process and to identify the determinants of mammographic density. To avoid some of the methodologic problems
inherent in the early studies, investigators incorporating mammographic density in their research must pay close attention to
the potential for bias inherent in their study design, in their
assessment and quantification of mammographic features, and in
their interpretation of findings (4-6).
Since first reported by Wolfe (7) in 1976, interest in mammographic features as a predictor of breast cancer has fluctuated.
In Wolfe’s early studies, mammographic features were assessed
primarily from xeromammographic images and classified into
four categories based on the visual interpretation of the relative
amount of dense tissue and the characteristic appearance of the
breast tissue. In response to Wolfe’s initial reports of very high
breast cancer risk that was associated with the high breast density patterns, many others have also evaluated the associations
between mammographic features and breast cancer risk (4,5).
Since these other studies did not all confirm the initially reported
association, many were willing to dismiss any possible association. Many of the nonconfirmatory studies, however, did not
consider the potential bias from the use of lower contrast filmscreen images, not blinding the radiologist reading the mammogram to case status, use of diagnostic images for assessment, and
Journal of the National Cancer Institute, Vol. 89, No. 8, April 16, 1997
inadequate training of the radiologist (6,8-10). The assessment
of ‘‘parenchymal patterns,’’ even by trained radiologists, was
highly variable in many of these studies, although the differences in assessment decreased after specific training in the recognition of mammographic features (9,10). The studies that
were designed to minimize these potential sources for bias consistently reported the association between mammographic features and breast cancer risk (4,5).
Subsequent studies (2,11) evaluating the components of the
parenchymal patterns reported that visual estimation of the percentage of the breast with dense tissue into five or six categories
predicted breast cancer risk. To improve upon the visual assessment, Wolfe et al. (12) first reported marking and measuring,
with a manual planimeter, the area of the breast with dense
appearance and the total breast area on the mammogram. From
these measurements, one could calculate the percentage of the
breast area with any mammographically dense tissue, a measure
now known as ‘‘breast density’’ (12). This measurement technique minimized both the variation due to the subjective measure of mammographic features and the difficulty of visually
assessing relative area from an irregularly shaped image. The
percentage of breast density was thus measured reliably on a
continuous scale. The measurement of breast density has been
adapted in several subsequent studies, using either a computerized planimeter or an interactive software system, where the
radiologist reading the mammogram determines the gray-scale
threshold for a digitized image to identify dense tissue (2,3).
These studies that measured mammographic breast density reported that women whose breast tissue is predominantly dense
(>75% of total breast area) were at a greater than fourfold increased risk of developing breast cancer than women with little
or no breast density (1-3).
In this issue of the Journal, Pankow et al. (13) report their
findings from a segregation analysis of cross-sectional, multigeneration information evaluating whether the transmission of a
major gene influences mammographic density. Incorporating the
assessment of mammographic features into this epidemiologic
and genetic study of breast cancer families enabled these researchers to evaluate predictors of mammographic density at this
time and to determine the role of mammographic density in the
*Correspondence to: Celia Byrne, Ph.D., Channing Laboratory, Harvard
Medical School, 181 Longwood Ave., Boston, MA 02115.
See ‘‘Notes’’ following ‘‘References.’’
EDITORIALS
531
incidence of breast cancer among the participants in the future.
The results of the segregation analyses presented by Pankow et
al. demonstrated that the data collected in these families were
consistent with a dominant, recessive, or codominant major
gene. The authors based their choice of a mendelian dominant
major gene on the slightly smaller Akaike’s Information Criterion (AIC) that is associated with this model (AIC of 10 889.88
versus 10 892.45 or 10 891.88 for the recessive and codominant
models, respectively). With such minimal differences in these
models, however, extreme caution must be observed in identifying one mode of inheritance over another. The possible associations with the mendelian models of genetic inheritance and
the rejection of the non-mendelian models in the study by Pankow et al. occurred only when other measured breast cancer risk
factors were considered simultaneously. Of greater interest is the
finding that, even if the hypothesis of a dominant mendelian
model of inheritance is correct, this model would explain less
than 30% of the variance in breast density among participants in
this study. While any nondifferential misclassification of breast
density may be partially obscuring a stronger association, additional genetic and/or environmental factors, not measured in this
study, are also likely to play a role in determining breast density.
Data are limited regarding familial patterns in mammographic features (14,15). A previous study from a referral hospital (14) indicated that the mammographic features from
mother–daughter and sister–sister pairs were more similar than
those from unrelated pairs of women of comparable ages. Comparing the breast density in 65 mother–daughter pairs and 275
sister–sister pairs, Pankow et al. (13) reported statistically significant, although fairly modest, correlations in sister–sister
pairs and lower nonsignificant correlations in mother–daughter
pairs. Acknowledging the limitation that these correlations were
adjusted only for two covariates simultaneously, the authors suggested that the smaller number of mother–daughter pairs and the
absence of information from fathers may explain the difference
in the magnitude of the correlations. In this regard, the authors
failed to consider a possible limitation of their study design.
Since the mammograms were all obtained within a few calendar
years, the mammograms of mothers and daughters reflect the
mammographic features of women of different generations.
Breast density changes throughout a woman’s life. Adjustment
for age (or age2) in this study design would not be adequate.
Ideally, one would like to know whether the breast density assessed from a woman’s mammogram at a given age correlates
with the breast density from her daughter’s mammogram when
her daughter is the same age.
Having mammographic information from only one point in
time may have constrained the ability of Pankow et al. to assess
the potential for genetic influence on breast density. While many
known factors are associated with a woman’s breast density,
Pankow et al. found that only about 30% of the variance in
breast density can be explained by the traditional breast cancer
risk factors. Despite knowing that a woman’s breast density
changes during her lifetime, to date the published studies of
breast density and breast cancer risk have all assessed breast
density from one mammographic examination. Whether changes
in breast density modify a woman’s breast cancer risk is unknown. While mammograms in premenopausal women have
been found to predict postmenopausal breast cancer risk and
532
EDITORIALS
breast density assessed on mammograms taken 10 or more years
previously predicts breast cancer risk, if the true risk is only in
the women whose breast density does not change, then the magnitude of the effect for those with dense breast patterns that do
not change may be even greater than that previously reported
(3). Our need to better understand the predictors of breast density and to evaluate the impact of change in breast density on
breast cancer risk becomes more relevant when change in breast
density is considered an intermediate end point that can be used
to evaluate potential breast cancer prevention strategies, as has
been proposed (16).
In designing future studies that will assess breast density, all
attempts should be made to minimize the potential misclassification of breast density. Rather than rely on an individual to
visually estimate 20 categories of breast density, as was done in
the study by Pankow et al. (13), methods to mark and measure
the area of mammographic density that are easily performed
should be considered. Individuals assessing breast density
should demonstrate that they are identifying the features of
breast density that are associated with breast cancer risk, either
by comparison with others who have reported this measured risk
or in a pilot assessment study. Many well-qualified individuals
may be consistent in their assessment, but they may not be
measuring the appropriate aspect of breast density. Previous
studies (9,10) have reported that the gradient in breast cancer
risk generally increased after training was given in assessment
techniques. Although it is not possible to assess the impact of
such misclassification in the study by Pankow et al., these two
study aspects—training the reader and measuring the breast density—should be considered in future studies.
A proposed epidemiologic study of the cause of breast cancer
that did not intend to assess menopausal status, use of postmenopausal hormones, parity, or age at first birth would be considered
incomplete even though each of these factors is associated with
only a moderate change in breast cancer risk. However, increased breast density, which is associated with a much greater
increase in risk, is typically overlooked. Pankow et al. (13) are
to be credited for the inclusion of mammographic assessment in
the baseline measures in their epidemiologic and genetic followup study of breast cancer families. Their analysis is consistent
with a genetic component being one of the factors influencing
mammographic density. Perhaps the most notable result of their
analysis, however, is the implication that a large number of the
factors that may influence mammographic density are unknown
at this time. Since more than 46% of breast cancers in some
studies are attributable to having any measurable breast density
(3), a better understanding of what is measured by mammographic density, what factors predict mammographic density,
and what happens to breast cancer risk if mammographic density
changes are three areas particularly in need of further research.
So as not to repeat past problems, studies of mammographic features must make an extensive effort to conduct a standardized
assessment of measured breast density, which would include
assessment of both the validity and the reliability of the measures used.
References
(1) Saftlas AF, Hoover RN, Brinton LA, Szklo M, Olson DR, Salane M, et al.
Mammographic densities and risk of breast cancer. Cancer 1991;67:2833-8.
Journal of the National Cancer Institute, Vol. 89, No. 8, April 16, 1997
(2) Boyd NF, Byng JW, Jong RA, Fishell EK, Little LE, Miller AB, et al.
Quantitative classification of mammographic densities and breast cancer
risk: results from the Canadian National Breast Screening Study. J Natl
Cancer Inst 1995;87:670-5.
(3) Byrne C, Schairer C, Wolfe J, Parekh N, Salane M, Brinton LA, et al.
Mammographic features and breast cancer risk: effects with time, age, and
menopause status. J Natl Cancer Inst 1995;87:1622-9.
(4) Saftlas AF, Szklo M. Mammographic parenchymal patterns and breast
cancer risk. Epidemiol Rev 1987;9:146-74.
(5) Oza AM, Boyd NF. Mammographic parenchymal patterns: a marker of
breast cancer risk. Epidemiol Rev 1993;15:196-208.
(6) Boyd NF, O’Sullivan B, Fishell E, Simor I, Cooke G. Mammographic
patterns and breast cancer risk: methodologic standards and contradictory
results. J Natl Cancer Inst 1984;72:1253-9.
(7) Wolfe JN. Breast parenchymal patterns and their changes with age. Radiology 1976;121(3 Pt. 1):545-52.
(8) Whitehead J, Carlile T, Kopecky KJ, Thompson DJ, Gilbert FI Jr, Present
AJ, et al. Wolfe mammographic parenchymal patterns. A study of the
masking hypothesis of Egan and Mosteller. Cancer 1985;56:1280-6.
(9) Grove JS, Goodman MJ, Gilbert FI Jr, Russell H. Wolfe’s mammographic
classification and breast cancer risk: the effect of misclassification on apparent risk ratios. Br J Radiol 1985;58:15-9.
(10) Carlile T, Thompson DJ, Kopecky KJ, Gilbert FI, Krook PM, Present AJ,
et al. Reproducibility and consistency in classification of breast parenchymal patterns. AJR Am J Roentgenol 1983;140:1-7.
(11) Brisson J, Merletti F, Sandowsky NL, Twaddle JA, Morrison AS, Cole P.
Mammographic features of the breast and breast cancer risk. Am J Epidemiol 1982;115:428-37.
(12) Wolfe JN, Saftlas AF, Salane M. Mammographic parenchymal patterns and
quantitative evaluation of mammographic densities: a case-control study.
AJR Am J Roentgenol 1987;148:1087-92.
(13) Pankow JS, Vachon CM, Kuni CC, King RA, Arnett DK, Grabrick DM, et
al. Genetic analysis of mammographic breast density in adult women:
evidence of a gene effect. J Natl Cancer Inst 1997;89:549-56.
(14) Wolfe JN, Albert S, Belle S, Salane M. Familial influences on breast
parenchymal patterns. Cancer 1980;46:2433-7.
(15) Kaprio J, Alanko A, Kivisaari L, Standertskjold-Nordenstam CG. Mammographic patterns in twin pairs discordant for breast cancer. Br J Radiol
1987;60:459-62.
(16) Cuzick J, Berridge D, Whitehead J. Mammographic dysplasia as entry
criterion for breast cancer prevention trials [letter]. Lancet 1991;337:1225.
Notes
Supported in part by a Massachusetts Department of Public Health Breast
Cancer Research Grant (SC-DPH-3408519C017) and by Public Health Service
grant CA464055 from the National Cancer Institute, National Institutes of
Health, Department of Health and Human Services.
I thank Drs. A. F. Saftlas, A. M. Goldstein, S. E. Hankinson, and G. A. Colditz
for their discussions of these issues.
Cell Transformation, Invasion, and Angiogenesis:
a Regulatory Role for Ornithine Decarboxylase
and Polyamines?
Merja Auvinen*
In this issue of the Journal, Kubota et al. (1) report that
overexpression of ornithine decarboxylase (ODC) triggers mitogen-activated protein (MAP) kinase activity, which in turn
implies a proportional increase in invasiveness. They demonstrate that transfection of mouse 10T1/2 fibroblasts with rat
ODC complementary DNA (cDNA) elicited morphologic transformation; the ODC transformants formed colonies in monolayer tissue culture and proliferated in semisolid soft agar.
Kubota et al. suggest that the ODC-induced cell transformation
mechanism can be explained, at least in part, by activation of the
MAP kinase pathway, since the ODC transformants that they
studied displayed increased protein kinase activity against myelin basic protein used as an MAP kinase substrate in in-gel
assays. In addition, they postulate that deregulated ODC activity
may influence invasion of a cancer cell. This was evidenced by
migration of ODC transformants through a reconstituted basement membrane-coated filter in Boyden chambers in vitro. Furthermore, Kubota et al. show increased secretion of matrix metalloproteinase (MMP)-2, a 72-kd progelatinase considered to be
one of the key players in extracellular matrix degradation. The
results obtained by these investigators provide strong evidence
that ODC activity and polyamines may affect a number of processes that have an impact on tumor development, including
abnormal cell proliferation and invasion.
Polyamines—putrescine, spermidine, and spermine—are
Journal of the National Cancer Institute, Vol. 89, No. 8, April 16, 1997
small cationic organic molecules that are indispensable for cell
proliferation and differentiation. All eukaryotic cells contain one
or more of the polyamines. Their concentrations vary during the
cell cycle, and one of the first events in proliferating cells is
induction of polyamine synthesis, preceding both nucleic acid
and protein syntheses. Beyond that, however, it is hard to elucidate the specific biologic functions of polyamines.
The key enzyme in the polyamine biosynthetic pathway is
ODC. In its active form, ODC is a dimer of two identical subunits of 51 kd. It is one of the most strictly regulated enzymes
known. ODC has a turnover rate ranging from a few minutes to
1 hour. In contrast, the average life of mammalian enzymes is
counted in days. This rapid turnover is essential for fast and
dramatic changes in ODC levels in response to many kinds of
growth stimuli that affect the rate of enzyme synthesis, making
its induction pattern more ‘‘pulse-like.’’ The activity of ODC is
regulated by many putative mechanisms. These mechanisms include transcriptional control of gene expression, suppression of
translation, post-translational modifications, and proteolytic destruction.
*Correspondence to: Merja Auvinen, Ph.D., Cellular Signaling Group, Division of Biochemistry, P.O. Box 56, Viikinkaari 5, FIN-00014 University of
Helsinki, Helsinki, Finland.
See ‘‘Notes’’ following ‘‘References.’’
EDITORIALS
533