American Journal of Epidemiology
ª The Author 2009. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 170, No. 12
DOI: 10.1093/aje/kwp313
Advance Access publication:
November 12, 2009
Practice of Epidemiology
Determinants of Percentage and Area Measures of Mammographic Density
Jennifer Stone*, Ruth M. L. Warren, Elizabeth Pinney, Jane Warwick, and Jack Cuzick
* Correspondence to Dr. Jennifer Stone, Centre for Molecular, Environmental, Genetic, and Analytic Epidemiology, The University of
Melbourne, Level 1, 723 Swanston Street, Carlton, Victoria 3053, Australia (e-mail: [email protected]).
Initially submitted March 2, 2009; accepted for publication September 11, 2009.
Mammographic density is one of the strongest predictors of breast cancer risk. Typically expressed as a percentage of the breast area occupied by radiologically dense tissue on a mammogram, its full value may not be
realized because of its negative association with body mass index. A simpler measure of mammographic density,
independent of other breast cancer risk factors and equally predictive of risk, would be preferable for risk prediction
models. Percentage and area measures of mammographic density were determined for 815 women at high risk for
breast cancer from the baseline assessments in the International Breast Cancer Intervention Study I, a trial of
tamoxifen for breast cancer prevention conducted between 1992 and 2001. Multivariate linear regression was used
to assess associations between risk factors and the mammographic measures. Percent dense area was negatively
associated with age, body mass index, menopausal status, predicted risk, and smoking status (R2 ¼ 24%). Dense
area was negatively associated with only age and body mass index (R2 ¼ 7%), and the latter association was
much weaker than for percent dense area. Nondense area was positively associated with age, body mass index,
and predicted risk (R2 ¼ 36%). Dense area was not associated with the multitude of risk factors that percent dense
area was, making it a simpler biomarker for risk prediction modeling. Both dense area and percent dense area
should be presented whenever possible for comparisons in research.
biological markers; breast neoplasms; mammography; risk; risk factors; women’s health
Abbreviation: SE, standard error.
underestimation of the risk estimates for percent dense area.
In particular, they found that the strength of association
between percent dense area and breast cancer risk increased
after adjustment for body mass index. This suggests that the
most useful risk factor for breast cancer is not simply percent dense area but, rather, percent dense area adjusted for
body mass index (and age).
A simpler measure of density, independent of known
breast cancer risk factors and equally predictive of risk,
would be highly preferable. It is possible that the absolute
area of dense tissue in the breast, instead of the percentage,
is less dependent on common risk factors such as body mass
index. Moreover, if we are to identify the environmental and
genetic determinants of mammographic density in hopes of
improving our understanding of the etiology of breast cancer, then it is essential that we investigate the potential separate pathways of disease. The purpose of this paper was to
Women with extensive mammographic density are more
likely to develop breast cancer than those of the same age
with little or no density (1). Mammographic density is typically expressed as a percentage of the total area of the
breast occupied by opaque/white tissue on a mammogram.
Denoted here as ‘‘percent dense area,’’ its determinants have
been shown to be similar to those associated with breast
cancer risk (2) with one notable exception. Increased body
mass index has been found to increase a women’s risk of
breast cancer (in postmenopausal women) but is negatively
associated with percent dense area (3). The latter is likely
because a woman’s body mass index is positively related to
total breast area, the denominator of percent dense area.
Anthropometric measures such as weight, waist circumference, or body mass index can account for almost a third
of the variation in percent dense area (2), and Boyd et al. (3)
have reported that failure to adjust for this can lead to an
1571
Am J Epidemiol 2009;170:1571–1578
1572 Stone et al.
compare the associations between breast cancer risk factors
and the components of percent dense area, absolute dense
area and absolute nondense area, separately.
MATERIALS AND METHODS
Subjects and methods
Subjects and methods for the International Breast Cancer
Intervention Study I trial have been described previously
and are only briefly mentioned here (4). Baseline mammograms were obtained from 815 women aged 35–70 years at
high (at least twice the population) risk of developing breast
cancer before being randomly assigned to take either tamoxifen or placebo for 5 years. All of the mammograms were
digitized by using a Vidar Advantage Pro digitizer (VIDAR
Systems Corporation, Herndon, Virginia) and were also
measured (J. S.) by use of Cumulus, a computer-assisted
interactive thresholding technique (Imaging Research Program, Sunnybrook Health Sciences Centre, University of
Toronto, Toronto, Canada) that involves an operator estimating the edge of the breast and setting a threshold level to
identify dense tissue. The number of pixels in the digitized
image that lie within the defined areas is then automatically
calculated and recorded. The result is an absolute measure
of the total area of the breast and an absolute measure of
dense area, respectively, which when subtracted gives the
area of nondense tissue and, when expressed as a ratio, gives
a percentage of dense tissue within the breast image.
The films were randomized by subject into reading sets of
approximately 100 and measured over the course of 3
weeks. The mediolateral oblique views of the right breasts
were used, and the reader was blinded to all other identifying information. A 10% random sample of mammograms
was repeated in each set and between every fifth set to test
the reliability of the measurement.
Statistical analysis
The intraclass correlation coefficient was used to assess
the reader’s intra- (within a reading set) and inter- (between
reading sets) reliability for each of the computer-assisted
measures. Univariate and multivariate linear least-squares
regression was used to examine associations between the
mammographic measures and known breast cancer risk factors measured at baseline. The predicted risk of each breast
cancer variable was generated via the Tyrer-Cuzick model
(5) and has been shown to be the most consistently accurate
model for prediction of breast cancer in high-risk women
(6).
Covariate-adjusted residuals of all of the regression
models were inspected for normality, and the baseline measures of dense area and nondense area required square-root
transformation. In the text, the regression coefficients of
dense area and nondense area were divided by their means
(transformed), providing a percentage increase/decrease per
risk factor unit on a linear scale.
Fractional polynomial smooth plots were generated in
STATA, version 10.1, software (StataCorp LP, College Station, Texas) that calculates the prediction for the mammo-
graphic density measure from estimation of a fractional
polynomial of body mass index and plots the resulting curve
along with the confidence interval of the mean.
RESULTS
Baseline mammograms from 815 women from the International Breast Cancer Intervention Study I were obtained;
however, 11 mammograms were excluded because of poor
quality. Anthropometric information was missing for an additional 5 subjects, leaving 799 for the main analyses. The
reliability of the density measurement within reading sets,
given by the intraclass correlation coefficient, was excellent
with estimates of 0.96, 0.99, and 0.94 for dense area, nondense area, and percent dense area, respectively. Between
reading sets, the intraclass correlation coefficients were
0.98, 0.99, and 0.97 for dense area, nondense area, and
percent dense area, respectively. Baseline characteristics
were reported previously by Cuzick et al. (4) and are briefly
reviewed in Table 1. The table also gives the means and
standard deviations of each of the mammographic measures
at baseline.
Figure 1 shows a scatterplot of absolute dense area by
percent dense area at baseline. Percent dense area was positively correlated with dense area (r ¼ 0.75), but the variation in dense area increased with increased percent dense
area. Conversely, percent dense area was negatively associated with nondense area (r ¼ 0.80), and the variation in
nondense area decreased with increased percent dense area
(Figure 2). Dense and nondense areas were also negatively
correlated (graph not shown; r ¼ 0.36), and the association between the computer-assisted and the visual assessments of percent dense area used in a previous paper (4) was
high (graph not shown; r ¼ 0.85).
Table 2 shows the linear regression coefficients (and their
standard errors) for the different breast cancer risk factors
for 3 of the mammographic measures. The regression estimates for percent dense area were very similar in magnitude
and direction to those produced by the visual assessment
reported previously (data not shown). Like the previous report (4), a multivariate model found that age, body mass
index, menopausal status, predicted breast cancer risk, and
smoking were all negatively associated with percent dense
area. Combined, these risk factors accounted for 24% of the
variation in percent dense area, with 15% of this explained
by body mass index alone.
Separate examination of the components of percent dense
area, dense and nondense area, revealed that, univariately,
postmenopausal subjects were found to have less squareroot dense area than premenopausal subjects (b ¼ 0.80
(standard error (SE), 0.13; Ptrend < 0.0001), and previous
hormone therapy users had less square-root dense area than
women who had never used hormone therapy (b ¼ 0.66
(SE, 0.18); Ptrend ¼ 0.002). Neither of these covariates remained significantly associated with square-root dense area
once adjusted for age in the multivariate model. In fact, only
age and body mass index were significantly associated with
square-root dense area in a multivariate model. Regression
estimates of the association between square-root dense area
Am J Epidemiol 2009;170:1571–1578
Determinants of Mammographic Density
Risk Factor
Mean (SD)
Age, years
No.
%
50.48 (6.16)
Body mass index, kg/m2
a
b
Age at menarche, years
26.73 (4.82)
12.87 (1.73)
Age at first birth, years
Nulliparous
106
13.18
>30
143
17.79
26–30
315
39.18
21–25
185
23.01
55
6.84
20
Absolute Dense Area, cm2
Table 1. Characteristics of International Breast Cancer Intervention
Study I Subjects at Baseline (N ¼ 804), 1992–2001
1573
100
50
0
0
20
Menopausal status
40
60
80
100
Percent Dense Area, %
Premenopause
391
48.63
Perimenopause
42
5.22
Postmenopause
376
46.14
Figure 1. Scatterplot of absolute dense area by percent dense area
at baseline, International Breast Cancer Intervention Study I, 1992–
2001.
Predicted risk of breast cancer
Low (<2-fold)
323
40.80
Moderate (2–3-fold)
395
49.13
81
10.07
Never
523
65.05
Previous
109
13.56
Current
172
21.39
Never
419
52.11
Previous
232
28.86
Current
153
19.03
High (>3-fold)
Hormone therapy use
Smoking status
Mammographic measures
Dense area, cm2
Nondense area, cm
25.02 (15.88)
2
Percent dense area, %
47.62 (26.35)
37.16 (21.46)
Square-root dense area
4.69 (1.74)
Square-root nondense area
6.62 (1.94)
for age in the multivariate model. Square-root nondense area
increased by approximately 0.8% with each yearly increase
in age (b ¼ 0.057 (SE, 0.009); Ptrend < 0.0001) and 3.5%
with each unit increase in body mass index (b ¼ 0.24 (SE,
0.01); Ptrend < 0.0001). The risk factors in the multivariate
model accounted for over a third (40%) of the variation in
square-root nondense area with body mass index explaining
36% alone.
Figures 3–5 are fractional polynomial smooth plots with
95% confidence intervals of all 3 mammographic measures
against body mass index. Figure 3 shows a weak negative
association between dense area and body mass index and an
increase in the width of the confidence intervals beyond
a body mass index of 30–35. Figure 4 depicts the strong
negative association between percent dense area and body
mass index that appears to flatten out before turning upward
for body mass index values greater than 40. Like dense area,
Abbreviation: SD, standard deviation.
n ¼ 799 because of missing values.
b
n ¼ 801 because of missing values.
150
and age indicated an approximate decrease in square-root
dense area of 1.5% with each yearly increase in age
(b ¼ 0.070 (SE, 0.010)). Similarly, square-root dense area
decreased by approximately 0.6% with each unit increase in
body mass index (b ¼ 0.029 (SE, 0.012)). Together, age
and body mass index accounted for only 7% of the variation
in square-root dense area.
Regression estimates between the risk factors and squareroot nondense area were opposite in direction to those of
square-root dense area and both measures of percent dense
area (computer-assisted and visual assessments). Postmenopausal subjects were found to have more square-root nondense area (b ¼ 0.64 (SE, 0.14)) in the univariate analysis,
but there was no evidence of this association once adjusted
Am J Epidemiol 2009;170:1571–1578
Nondense Area, cm2
a
100
50
0
0
20
40
60
80
100
Percent Dense Area, %
Figure 2. Scatterplot of absolute nondense area by percent dense
area at baseline, International Breast Cancer Intervention Study I,
1992–2001.
Dense Areaa
Percent Dense Area
Variable
Univariate
b (SE)
2
P Value
Body mass index, kg/m
1.71 (0.15) <0.0001
Weight, kg
0.52 (0.05) <0.0001
Height, cm
Age at entry, years
Age at menarche, years
0.31 (0.12)
>30
P Value
Univariate
b (SE)
1.64 (0.14) <0.0001 0.031 (0.013)
P Value
0.01
0.10
0.016 (0.010)
0.11
0.74 (0.15) <0.0001 0.069 (0.010)
0.026 (0.036)
0.3
Nondense Areaa
Multivariate
0.0075 (0.0046)
0.2e
Age at first birth, years
Nulliparious
b (SE)
0.01
0.90 (0.12) <0.0001
0.49 (0.44)
Multivariate
b
<0.0001
0.5
c
b (SE)
P Value
0.029 (0.012)
0.02
b (SE)
P Value
0.24 (0.01)
<0.0001
Referent
Referent
0.021 (0.011)
0.070 (0.010) <0.0001
0.50 (0.32)
4.14 (2.61)
0.37 (0.21)
0.29 (0.23)
5.00 (2.41)
0.20 (0.20)
0.61 (0.22)
20
6.30 (2.75)
Premenopausal
Referent
Perimenopausal
1.17 (3.38)
0.12 (0.22)
Postmenopausal
10.72 (1.51)
Referent
Referent
4.80 (1.80)
0.5e
Predicted risk of
breast cancer
Referent
Referent
1.24 (1.60)
0.65 (1.42)
0.19 (0.13)
6.00 (2.48)
0.23 (0.22)
1.41 (2.66)
Never
Am J Epidemiol 2009;170:1571–1578
Previous
7.27 (2.25)
Never
Referent
Referent
Referent
0.044 (0.11)
Referent
0.29 (0.20)
0.15e
Referent
0.13e
Referent
Referent
Referent
Current
5.05 (2.02)
4.91 (1.79)
0.31 (0.16)
0.34 (0.18)
Previous
2.15 (1.75)
2.24 (1.56)
0.019 (0.14)
0.22 (0.16)
Abbreviation: SE, standard error.
Square-root transformed.
b
Adjusted for body mass index, age at entry, menopausal status at entry, predicted risk of breast cancer, and smoking.
c
Adjusted for body mass index and age at entry.
d
Adjusted for body mass index, age at entry, and predicted risk of breast cancer. Only significant variables were retained in the multivariate models.
e
Ptrend.
a
0.60 (0.20)
0.3e
0.17 (0.17)
0.66 (0.18)
0.02e
0.008e
0.066 (0.14)
0.49 (0.24)
0.18 (0.15)
0.04e
Smoking status
0.1e
0.002e
Referent
2.74 (1.88)
0.64 (0.14)
Referent
0.004e
Current
Referent
0.3e
Moderate (2–3-fold)
High (>3-fold)
<0.0001e
0.035 (0.31)
0.80 (0.13)
0.03e
Low (<2-fold)
Use of hormone
therapy
0.79 (0.25)
<0.0001e
0.25 (0.28)
0.22 (3.07)
0.057 (0.009) <0.0001
0.005
0.01e
26–30
0.02e
<0.0001
Referent
0.015 (0.29)
<0.0001e
0.24 (0.01)
P Value
0.05
0.052 (0.011) <0.0001
0.11 (0.04)
21–25
Menopausal
status at entry
b (SE)
0.078 (0.004) <0.0001
0.4e
1.76 (3.56)
Multivariated
Univariate
1574 Stone et al.
Table 2. Linear Regression Estimates of All 3 Mammographic Measures at Entry to the Study for Various Risk Factors, International Breast Cancer Intervention Study I, 1992–2001
Determinants of Mammographic Density
Figure 3. Fractional polynomial smooth plot with 95% confidence
intervals (CIs) of absolute dense area against body mass index at
baseline, International Breast Cancer Intervention Study I, 1992–
2001.
the variation in percent dense area increases beyond a body
mass index of 30. Finally, Figure 5 shows the strong positive
association between nondense area and body mass index;
however, the increase in the variation of nondense area starts
at body mass index values greater than 35–40.
DISCUSSION
In summary, percent dense area was negatively associated
with age, body mass index, menopausal status, predicted
risk, and smoking status. Nondense area was positively associated with age, body mass index, and predicted risk.
Dense area was negatively associated with only age and
body mass index, but to a lesser degree than for percent
dense area. Although the determinants of mammographic
density may not be novel, investigating how they differ for
different measurements (e.g., absolute vs. percent) highlights the importance of understanding the potentially different pathways that influence both percent dense area and
absolute dense area.
Although body mass index was found to be significantly
associated with dense area, it explained only a fraction
Figure 4. Fractional polynomial smooth plot with 95% confidence
intervals (CIs) of percent dense area against body mass index at
baseline, International Breast Cancer Intervention Study I, 1992–
2001.
Am J Epidemiol 2009;170:1571–1578
1575
Figure 5. Fractional polynomial smooth plot with 95% confidence
intervals (CIs) of nondense area against body mass index at baseline,
International Breast Cancer Intervention Study I, 1992–2001.
(<2%) of its variation in this study. Conversely, 15% of
the variation in percent dense area (computer assisted or
visual) and over a third of the variation in nondense area
were explained by body mass index. Therefore, the negative
correlation between percent dense area and body mass index
appears to be largely driven by the area of nondense tissue in
the breast. This implies that dense area could be a simpler
predictor of breast cancer risk, as it would minimize the need
to adjust for body mass index. From the results in Table 2,
it appears that dense area is independent of other breast
cancer risk factors as well. Assuming that dense area is also
a strong predictor of risk (see below), this finding has many
practical implications. It could help to simplify breast cancer risk prediction models that, in turn, could be used to
optimize breast-screening intervals. It could also help to
focus the search for not only environmental causes of mammographic density but also genetic causes. All 3 of the
mammographic measures have been shown to have a strong
heritable component (7–9). Previous work with twins
showed that the genetic correlation between dense and nondense areas was significant but inverse, indicating that common genetic influences exist but likely act in opposite
directions. The paper by Douglas et al. (8) suggested the
exact opposite but showed significant positive genetic correlations between several measures of body size and nondense (but not dense) area. Another paper by Ursin et al.
(10) suggests that a portion of the heritable variation in
percent and absolute dense area may be due to variation in
modifiable factors such as body mass index, and efforts to
replicate this finding are underway.
There have been previous reports examining the association between breast cancer risk factors and absolute dense
and nondense areas (11–15). All found a negative or no
association between dense area and weight or body mass
index with the exception of one study of Singaporean
Chinese women (14), where body mass index was positively
associated with dense area and one other study that found
total body fat mass to be positively associated with dense
area in premenopausal women but negatively associated in
postmenopausal women (12). Those that reported associations with nondense area showed strong positive
1576 Stone et al.
Table 3. Comparison of Odds Ratios of the Association of Breast Cancer With Dense Area and Percent Dense Area, International Breast Cancer
Intervention Study I, 1992–2001
Dense Area
First Author, Year
(Reference No.)
Byrne, 1995 (19)b
c
Kato, 1995 (20)
Maskarinec, 2000 (21)d
Ursin, 2003 (24)f
No. of
Cases
No. of
Controls
1,880
2,152
52
195
91
Subgroup
Odds
Ratio
95%
Confidence
Interval
Percent Dense Area
Comparisona
Odds
Ratio
95%
Confidence
Interval
Comparisona
3.35
2.6, 4.3
Quintiles
4.35
3.1, 6.1
>75 vs. 0
Premenopausal
incident cases
4.6
1.6, 12.9
Tertiles
3.6
1.4, 9.1
Tertiles
178
Postmenopausal
incident cases
3.0
1.5, 6.1
Tertiles
2.3
1.1, 4.8
Tertiles
80
221
Incident cases
diagnosed <3 years
5.8
2.3,14.3
Tertiles
3.9
1.7, 8.7
Tertiles
63
152
Incident cases
diagnosed 3 years
2.0
0.9, 4.2
Tertiles
1.9
0.8, 4.1
Tertiles
647
647
All cases/controlse
1.8
1.2, 2.6
Quintiles
1.8
1.1, 3.0
>50 vs. <10
Caucasian/Native
Hawaiian
2.3
1.4, 4.0
Quintiles
1.8
0.9, 3.6
>50 vs. <10
622
443
3.80
1.43, 10.08
Sextiles
5.23
1.70, 16.13 >75 vs. 0
Nagata, 2005 (22)g
71
370
Premenopausal
2.78
0.77, 10.1
Quintiles
1.36
0.31, 6.06
75
289
Postmenopausal
4.02
1.80, 8.94
Quintiles
4.19
1.33, 13.2
>75 vs. 0
Vachon, 2007 (25)h
372
708
Craniocaudal view;
cancer side
2.45
1.59, 3.78
Quartiles
3.06
1.88, 4.98
Quartiles
371
709
Mediolateral oblique
view; cancer side
2.43
1.58, 3.73
Quartiles
2.77
1.74, 4.42
Quartiles
368
710
Craniocaudal view;
control side
2.63
1.64, 4.23
Quartiles
3.66
2.25, 5.94
Quartiles
370
713
Mediolateral oblique
view; control side
2.33
1.50, 3.60
Quartiles
2.47
1.56, 3.91
Quartiles
111
3,100
2.69
1.40, 5.16
Quartiles
3.49
1.69, 7.18
Quartiles
Torres-Mejia, 2005 (23)i
>75 vs. 0
a
Odds ratio estimates compare highest with lowest categories. For percent dense area, values are percentages.
Control subjects were matched to screening cases on study center, race, year of birth, date of entry, and number of screening visits and to
follow-up case subjects on study center, race, year of birth, and follow-up group. Odds ratio estimates were adjusted for weight, age at first birth,
first-degree relatives with history of breast cancer, years of education, alcohol use, number of prior breast biopsies reported as benign, and
reproductive years.
c
Cases and controls were matched on age, menopausal status, date of enrollment, and number and dates of subsequent blood donations.
Premenopausal subjects were also matched by day and phase of the menstrual cycle at enrollment. Odds ratio estimates were adjusted for body
mass index, parity, and time since menopause.
d
Odds ratio estimates were adjusted for age at menarche, menopausal status, parity, age at first livebirth, family history of breast cancer,
hormone use at time of mammogram, and previous breast problems.
e
Caucasian, Native Hawaiian, Chinese, Filipino, and Japanese.
f
Odds ratio estimates were adjusted for age, body mass index, age at menarche, breast cancer family history, number of full-term pregnancies,
menopausal status and hormone replacement therapy use, and age at first full-term pregnancy.
g
Odds ratio estimates were adjusted for age, body mass index, age at menarche, age at first birth, number of full births, use of hormone
replacement therapy, history of breastfeeding, and family history of breast cancer among first-degree relatives.
h
Odds ratio estimates were adjusted for age, family history, menopausal status at mammogram, hormone replacement therapy at mammogram,
body mass index at mammogram, age at first birth, and number of livebirths.
i
Odds ratio estimates were adjusted for age, age at leaving full-time education, social class, job status, parity, height, body mass index at
Guernsey III (1977–1985), and body mass index change from Guernsey III to Guernsey IV (1986–1989).
b
associations with weight and/or body mass index. A recent
longitudinal study by Reeves et al. (16) showed that changes
in body mass index were not associated with changes in
dense area but may be negatively associated with percent
dense area. The authors concluded that absolute dense area
is likely the better surrogate for breast cancer, but they also
highlight the importance of nondense area in the etiology of
the disease as well.
The direction of the associations between the mammographic density measures and the other investigated risk
factors was mostly consistent with that of the literature. A
marginal negative association between percent dense area
and the level of risk predicted by known risk factors was
seen, but we believe this is likely to be a chance finding in
view of little association seen in previous reports and a positive association with atypical hyperplasia (17). Hormone
therapy use was negatively associated with absolute and
percent dense areas but only in the univariate analysis,
and the association did not persist after adjustment for
age. We excluded the current hormone therapy users and
Am J Epidemiol 2009;170:1571–1578
Determinants of Mammographic Density
repeated the regression analysis that produced the estimates
in Table 2 and found no differences in the results. The
women in this study were at higher risk for breast cancer
and had higher than average density. We recognize that the
associations of density with the risk factors may be different
in different populations.
There is increasing evidence suggesting that the key parameter for breast cancer risk may be related more to the
absolute amount of dense area than to the percentage of
breast density (18). Several studies have investigated the
association between dense area and breast cancer risk
(19–25) and are summarized in Table 3. Overall, 5 found
that the risk estimates were similar in magnitude to that of
percent dense area, and the other 2 found that the risk of
breast cancer associated with increased dense area was
stronger than that of percent dense area. Maskarinec et al.
(26) compared differences in percent and absolute dense
areas with breast cancer risk incidence in populations at
different breast cancer risk and found that age-adjusted
dense area may reflect breast cancer incidence better than
percent densities. None of the above studies provided risk
estimates for nondense area.
The measurement of dense area was found to be as reliable as that of percent dense area, which is consistent with
results from other studies (10, 26), but currently, measurement of dense area can be obtained only from digital images
or digitized copies of traditional films created by using
a high-resolution scanner (measurement of percent dense
area is not restricted to computer-assisted methods as it
can also be assessed visually). There are several sources
of error when measuring mammographic density, whatever
the method used. Different film types, film projections,
mammography machines, and radiographer technique can
all potentially alter the appearance of dense tissue in a mammogram. In particular, dense area is potentially more susceptible to variation due to the levels of compression during
mammography, particularly when comparing images from
the same woman over time. The level of compression may
depend on radiographer technique or patient sensitivity due
to menstrual phase or hormone use. In addition, there are
several technical issues that must be taken into account
when calculating dense area, such as traditional versus digital film, image resolution, and image size. These issues are
of particular importance when comparing dense area measurements across studies.
Differences in body mass index also complicate the
measurement of mammographic density, since it is a 3dimensional object measured from a 2-dimensional image.
Figures 3–5 graphically showed the associations between
the mammographic measures and body mass index, which
also highlighted the large variation in the measures in
women with very high body mass index, particularly in
dense area. This is likely due to small numbers (only 55
subjects had a body mass index greater than 35); however,
it could also be related to the fact that women with higher
body mass index generally have larger breasts. This could
mean that either the measurement of dense area is less
accurate because of measurement error in thicker breasts,
or it could be that dense area is actually more variable in
larger breasted women. If the first is true, the measurement
Am J Epidemiol 2009;170:1571–1578
1577
error for the computer-assisted measurement of percent
dense area should be more or less the same as for dense
area, because it requires the calculation of the total area in
addition to absolute dense area. If the second is true, then
dense area is a better measure of mammographic density
than percent dense area, as it captures something that percent dense area does not. A recent paper by Stuedal et al.
(27) found that the association between mammographic
density and breast cancer may be weaker in women with
larger breasts. One possible explanation given by the authors is that breast fat could be protective due to the activity
of adipocytes in fat tissue. Another explanation could be
measurement error, as described above. A volumetric measurement of mammographic density could provide valuable insight into the relation between mammographic
density and body mass index, as it takes into account the
thickness of the breast. Several methods to measure volumetric mammographic density have been developed (28–
30); however, there is currently only one study to date that
has examined its association with breast cancer risk. In
a recent paper by Ding et al. (31), volumetric percent density as measured by a fully automated computer program,
standard mammogram form, was not found to be a better
predictor of breast cancer risk than percent dense area
measured via the same computer-assisted method used in
this study. The authors did not report on the association
between dense area and breast cancer risk nor did they
adjust for body mass index.
In summary, a simple, reliable, and reproducible measure
of mammographic density, independent of known breast
cancer risk factors and equally predictive of risk, would
be very useful for research purposes and in clinical practice.
We found that absolute dense area was not associated with
the multitude of risk factors that percent dense area was,
thus minimizing the need to adjust for these risk factors,
particularly body mass index, making it a potentially simpler biomarker for breast cancer research. It also highlights
the importance of understanding the potentially different
pathways that influence both percent dense area and absolute dense area. We would recommend that both percent
dense area and dense area be presented whenever possible
for comparisons in research.
ACKNOWLEDGMENTS
Author affiliations: Centre for Molecular, Environmental,
Genetic, and Analytic Epidemiology, The University of
Melbourne, Melbourne, Victoria, Australia (Jennifer Stone);
Department of Radiology, Addenbrooke’s Hospital,
Cambridge, United Kingdom (Ruth M. L. Warren); and
Cancer Research United Kingdom Department of Epidemiology, Mathematics, and Statistics, Wolfson Institute of
Preventive Medicine, Barts and The London, Queen Mary
School of Medicine, University of London, London, United
Kingdom (Elizabeth Pinney, Jane Warwick, Jack Cuzick).
This work was supported by Cancer Research United
Kingdom program grant C569/A10404.
Conflict of interest: none declared.
1578 Stone et al.
REFERENCES
1. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis.
Cancer Epidemiol Biomarkers Prev. 2006;15(6):1159–1169.
2. Boyd NF, Lockwood GA, Byng JW, et al. Mammographic
densities and breast cancer risk. Cancer Epidemiol Biomarkers
Prev. 1998;7(12):1133–1144.
3. Boyd NF, Martin LJ, Sun L, et al. Body size, mammographic
density, and breast cancer risk. Cancer Epidemiol Biomarkers
Prev. 2006;15(11):2086–2092.
4. Cuzick J, Warwick J, Pinney E, et al. Tamoxifen and breast
density in women at increased risk of breast cancer. J Natl
Cancer Inst. 2004;96(8):621–628.
5. Tyrer J, Duffy SW, Cuzick J. A breast cancer prediction model
incorporating familial and personal risk factors. Stat Med. 2004;
23(7):1111–1130.
6. Amir E, Evans DG, Shenton A, et al. Evaluation of breast
cancer risk assessment packages in the family history evaluation and screening programme. J Med Genet. 2003;40(11):
807–814.
7. Boyd NF, Dite GS, Stone J, et al. Heritability of mammographic density, a risk factor for breast cancer. N Engl J Med.
2002;347(12):886–894.
8. Douglas JA, Roy-Gagnon MH, Zhou C, et al. Mammographic
breast density—evidence for genetic correlations with established breast cancer risk factors. Cancer Epidemiol Biomarkers
Prev. 2008;17(12):3509–3516.
9. Stone J, Dite GS, Gunasekara A, et al. The heritability of
mammographically dense and nondense breast tissue. Cancer
Epidemiol Biomarkers Prev. 2006;15(4):612–617.
10. Ursin G, Lillie EO, Lee E, et al. The relative importance of
genetics and environment on mammographic density. Cancer
Epidemiol Biomarkers Prev. 2009;18(1):102–112.
11. Dite GS, Gurrin LC, Byrnes GB, et al. Predictors of mammographic density: insights gained from a novel regression
analysis of a twin study. Cancer Epidemiol Biomarkers Prev.
2008;17(12):3474–3481.
12. Caire-Juvera G, Arendell LA, Maskarinec G, et al. Associations between mammographic density and body composition
in Hispanic and non-Hispanic white women by menopause
status. Menopause. 2008;15(2):319–325.
13. Haars G, van Noord PA, van Gils CH, et al. Measurements of
breast density: no ratio for a ratio. Cancer Epidemiol Biomarkers
Prev. 2005;14(11 pt 1):2634–2640.
14. Heng D, Gao F, Jong R, et al. Risk factors for breast cancer
associated with mammographic features in Singaporean Chinese women. Cancer Epidemiol Biomarkers Prev. 2004;13(11
pt 1):1751–1758.
15. Maskarinec G, Meng L, Ursin G. Ethnic differences in mammographic densities. Int J Epidemiol. 2001;30(5):959–965.
16. Reeves KW, Stone RA, Modugno F, et al. Longitudinal association of anthropometry with mammographic breast density
in the Study of Women’s Health Across the Nation. Int J
Cancer. 2009;124(5):1169–1177.
17. Cuzick J, Berridge D, Whitehead J. Mammographic dysplasia
as entry criterion for breast cancer prevention trials [letter].
Lancet. 1991;337(8751):1225.
18. Ursin G, Hovanessian-Larsen L, Parisky YR, et al. Greatly
increased occurrence of breast cancers in areas of mammographically dense tissue. Breast Cancer Res. 2005;7(5):R605–
R608.
19. Byrne C, Schairer C, Wolfe J, et al. Mammographic features and breast cancer risk: effects with time, age, and
menopause status. J Natl Cancer Inst. 1995;87(21):1622–
1629.
20. Kato I, Beinart C, Bleich A, et al. A nested case-control study
of mammographic patterns, breast volume, and breast cancer
(New York City, NY, United States). Cancer Causes Control.
1995;6(5):431–438.
21. Maskarinec G, Meng L. A case-control study of mammographic densities in Hawaii. Breast Cancer Res Treat. 2000;
63(2):153–161.
22. Nagata C, Matsubara T, Fujita H, et al. Mammographic density
and the risk of breast cancer in Japanese women. Br J Cancer.
2005;92(12):2102–2106.
23. Torres-Mejı́a G, De Stavola B, Allen DS, et al. Mammographic features and subsequent risk of breast cancer: a comparison of qualitative and quantitative evaluations in the
Guernsey prospective studies. Cancer Epidemiol Biomarkers
Prev. 2005;14(5):1052–1059.
24. Ursin G, Ma H, Wu A, et al. Mammographic density and breast
cancer in three ethnic groups. Cancer Epidemiol Biomarkers
Prev. 2003;12(4):332–338.
25. Vachon CM, Brandt KR, Ghosh K, et al. Mammographic
breast density as a general marker of breast cancer risk. Cancer Epidemiol Biomarkers Prev. 2007;16(1):43–49.
26. Maskarinec G, Pagano I, Chen Z, et al. Ethnic and geographic
differences in mammographic density and their association
with breast cancer incidence. Breast Cancer Res Treat. 2007;
104(1):47–56.
27. Stuedal A, Ma H, Bernstein L, et al. Does breast size modify
the association between mammographic density and breast
cancer risk? Cancer Epidemiol Biomarkers Prev. 2008;17(3):
621–627.
28. Astley SM, Brady M, Rose C, et al., eds. Digital Mammography: 8th International Workshop, IWDM 2006, Manchester,
United Kingdom, June 18–21, 2006, Proceedings. New York,
NY: Springer-Verlag; 2006.
29. Highnam R, Pan X, Warren R, et al. Breast composition
measurements using retrospective standard mammogram form
(SMF). Phys Med Biol. 2006;51(11):2695–2713.
30. Klifa C, Carballido-Gamio J, Wilmes L, et al. Quantification of breast tissue index from MR data using fuzzy clustering. Conf Proc IEEE Eng Med Biol Soc. 2004;3:1667–
1670.
31. Ding J, Warren R, Warsi I, et al. Evaluating the effectiveness of
using standard mammogram form to predict breast cancer risk:
case-control study. Cancer Epidemiol Biomarkers Prev. 2008;
17(5):1074–1081.
Am J Epidemiol 2009;170:1571–1578