American Journal of Epidemiology
© The Author 2013. 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. 179, No. 4
DOI: 10.1093/aje/kwt260
Advance Access publication:
October 28, 2013
Original Contribution
Associations of Mammographic Dense and Nondense Areas and Body Mass Index
With Risk of Breast Cancer
Laura Baglietto, Kavitha Krishnan, Jennifer Stone, Carmel Apicella, Melissa C. Southey,
Dallas R. English, John L. Hopper*, and Graham G. Giles
* Correspondence to Dr. John L. Hopper, University of Melbourne, School of Population and Global Health, Level 3, 207 Bouverie Street, Carlton,
Victoria 3053, Australia (e-mail: [email protected]).
Initially submitted March 5, 2013; accepted for publication October 2, 2013.
Mammographic density measurements are associated with risk of breast cancer. Few studies have investigated
the concurrent associations of mammographic dense and nondense areas, body mass index (weight (kg)/height (m)2),
and ages at mammogram and diagnosis with breast cancer risk. We conducted a matched, case-control study nested
within the Melbourne Collaborative Cohort Study (cohort recruitment in 1990–1994 and follow-up until 2007) to estimate the associations between these factors and breast cancer risk under alternative causal models. Mammographic dense area was positively associated with risk, and the strength of this association was only slightly
influenced by the choice of the causal model (relative risk per 1 standard deviation = 1.50, 95% confidence interval:
1.32, 1.70). Mammographic nondense area was inversely associated with risk under the assumption that fat in the
body and fat in the breast cause breast cancer through independent mechanisms (relative risk per 1 standard
deviation = 0.75, 95% confidence interval: 0.65, 0.86), whereas it was not associated with risk under the assumption
that they are both proxies of adiposity. Knowledge about the biological mechanisms regulating the role played by
mammographic nondense area and body fat on breast cancer risk is essential to better estimate their impacts on
individual risk.
breast neoplasms; mammographic density; prospective studies; risk
Abbreviations: AUC, area under the receiver operating characteristic curve; BIC, Bayesian information criterion; BMI, body mass
index; CI, confidence interval; DA, mammographic dense breast area; MCCS, Melbourne Collaborative Cohort Study; NDA,
mammographic nondense breast area; PMD, percent mammographic density.
BMI as a risk factor as a step function, such as is typically
done in studies of PMD as a risk factor for breast cancer, is
not appropriate, and models should allow for the BMI risk
association to depend on age as a continuous variable.
A risk model including PMD as a measure of mammographic density does not allow the estimation of separate associations for mammographic nondense breast area (NDA)
and measures of overall adiposity, such as BMI. In this context, it might be more appropriate to use absolute mammographic dense breast area (DA) as a measure of mammographic
density. Many authors have reported strong positive associations between DA and breast cancer risk (5, 6), whereas few
have investigated the association between NDA and breast
Mammographic density refers to the amount of breast tissue that appears light on a mammogram. In most epidemiologic studies, mammographic density is analyzed as percent
mammographic density (PMD). Women with high PMD
(≥75%) have a 4- to 6-fold increased risk of breast cancer
compared with women of the same age and body mass index
(BMI) (weight (kg)/height (m)2) with primarily fatty breasts
(PMD ≤10%) (1). PMD is negatively associated with age and
BMI, which in turn are positively associated with breast cancer risk in the older age groups typically studied. BMI is negatively associated with risk of breast cancer for young women
(e.g., prior to menopause), but the association is positive at an
older age (e.g., after menopause) (2–4). Therefore, estimating
475
Am J Epidemiol. 2014;179(4):475–483
476 Baglietto et al.
cancer risk with contrasting findings (5, 7–9) that are difficult
to interpret (10, 11). To try to clarify the relationship between
mammographic density and BMI and breast cancer risk, we
conducted a case-control study nested within the Melbourne
Collaborative Cohort Study (MCCS) and estimated the associations of close-to-baseline measurements of DA, NDA,
PMD, and BMI with breast cancer risk, allowing these associations to depend on age at diagnosis.
age at baseline and reference age. From these mammograms,
we selected for the study the mammogram closest to baseline;
measurements were taken by using the craniocaudal image of
the contralateral breast with respect to the laterality of the
tumor in the matching case. After we excluded women belonging to incomplete matching sets (i.e., no case or no controls),
590 cases and 1,695 controls (2,162 women) remained for
analysis. Mammograms were taken between 1991 and 2007.
MATERIALS AND METHODS
Measurement of mammographic density
Melbourne Collaborative Cohort Study
Mammograms were digitized by the Australian Mammographic Density Research Facility at the University of
Melbourne with an Array 2905 high-density film digitizer
(Array Corporation Europe, Roden, the Netherlands). The digitized images were masked, and total breast area and DA were
measured by using a semiautomated computer-assisted thresholding technique called Cumulus (Imaging Research Program, Sunnybrook Health Sciences Centre, University of
Toronto, Toronto, Canada). NDA was calculated by subtracting DA from the total breast area. PMD was calculated by dividing DA by the total breast area. Images were measured by 2
independent readers (J.S. and K.K.) blind to the disease status
of participants. Mammograms were read in sets of approximately 100 images. A random sample of 10% of the images
was repeated within each set to assess within-reader reliability.
The sample of mammograms repeated in the first set was also
repeated in every fifth set to assess within-reader reliability between sets.
The MCCS is a prospective cohort study of 41,514 people
(including 24,469 women) aged 27–76 years at baseline (12).
Recruitment occurred between 1990 and 1994. At baseline
attendance, participants completed questionnaires relative
to demographic characteristics, lifestyle, and diet; standard
procedures were used to measure height and weight (3, 13),
from which BMI was calculated.
Obtaining mammograms through BreastScreen Victoria
BreastScreen Victoria is part of Australia’s national breast
cancer screening program, established in 1992 to offer free
mammographic screening to women over 40 years of age.
The target group for the screening program is women aged
50–69 years, who are invited for screening every 2 years. At
screening, mammograms are taken, and information is collected on a range of variables. Women attending BreastScreen
Victoria give consent for their results to be used for research
and evaluation. In 2009, we conducted a record linkage between the MCCS and BreastScreen Victoria. Of the 24,469
women in the MCCS, 20,444 (84%) had attended BreastScreen
Victoria at least once and were eligible for this study.
Nested case-control study
Cases were women with a first diagnosis of ductal carcinoma in situ or invasive adenocarcinoma of the breast (International Classification of Diseases for Oncology codes
C50.0–C50.9) that occurred during follow-up between
the baseline interview and December 31, 2007. Cases were
ascertained by record linkage to the population-complete
Victorian Cancer Registry and to the Australian Cancer Database. There were 800 breast cancer cases diagnosed during
follow-up, including 120 ductal carcinoma in situ cases and
680 invasive breast cancer cases.
We designed a nested case-control study by using incidence density sampling with age as the time scale. For each
breast cancer case, we randomly selected 4 women as controls from among those who had not been diagnosed with
breast cancer at the age of diagnosis of the case (reference
age). The controls were also matched on year of birth, year
of MCCS baseline attendance, and country of origin (classified as Australia/New Zealand/United Kingdom/others, Italy,
or Greece). The Cancer Council Victoria’s human research
ethics committee approved the study.
Sixty-nine percent of the women initially selected in the
case-control sample had at least 1 mammogram between
Statistical analysis
To estimate within-reader and between-reader reliabilities,
we applied linear mixed-effects models and calculated intraclass correlation coefficients. DA, NDA, and PMD were
transformed to the powers of 0.2, 0.5, and 0.3, respectively,
to achieve normalization. The within-reader reliabilities were
0.96 for DA, 1.00 for NDA, and 0.97 for PMD for both readers. The between-reader reliabilities were 0.87 for DA, 0.99
for NDA, and 0.90 for PMD. Given the high between-reader
reliability, we used the averages from the 2 readers for each of
the measurements of DA, NDA, and PMD for the analyses.
We worked under the assumption of 2 alternative models
representing the relationship between DA, NDA, BMI, and
breast cancer risk (Figure 1). In the first causal diagram (Figure 1A), DA, NDA, and BMI are all causes of breast cancer;
the diagram also indicates that the relationship between DA
and NDA and breast cancer risk goes through a function F of
these 2 variables, which we assume to be either a linear combination of the 2 or the proportion of DA over total breast area
(the PMD). The correlation structure between DA, NDA, and
BMI is represented by the latent (unmeasured) variables X, Y,
and Z. In the second causal diagram (Figure 1B), DA and the
latent variable Y are both causes of breast cancer, whereas
NDA and BMI are not. For example, if we think of Y as the
generic variable “fat,” Figure 1A illustrates the assumption
that fat causes breast cancer via its effects on both nondense
area (fat in the breast) and BMI (overall body fat). Conversely, Figure 1B makes the assumption that fat is a risk factor for breast cancer irrespective of its distribution in the
Am J Epidemiol. 2014;179(4):475–483
Mammographic Density and Risk of Breast Cancer 477
A)
DA
X
Z
F
NDA
BC
Y
BMI
B)
DA
X
NDA
Z
BC
Y
BMI
Figure 1. Causal diagrams for the relationship between body mass
index (BMI) (weight (kg)/height (m)2), mammographic dense area
(DA), mammographic nondense area (NDA) and breast cancer risk
(BC). Z, X, and Y are latent variables responsible for the correlation
between BMI, DA, and NDA. A) DA, NDA, and BMI are risk factors
for breast cancer. The relative risk for breast cancer can be expressed
as follows: ln(relative risk (RR)BC) ∼ α × BMI + β × F with F = DA + γ ×
NDA or F = percent mammographic density = DA / (DA + NDA). B)
DA and adiposity are risks factors for breast cancer. NDA and BMI
are proxies of adiposity. Relative risk can be expressed as ln(RRBC) ∼
δ × Y + ε × DA.
body, whereas NDA and BMI are both proxies for fat. Directed acyclic graph theory was used to identify the variables
to include in the risk model (14). The goodness of fit of the 4
models was compared by using the Bayesian information criterion (BIC) and the area under the receiver operating characteristic curve (AUC). BIC and AUC values are from the
model including all the variables as pseudocontinuous.
Relative risks of breast cancer and their 95% confidence
intervals were estimated by fitting conditional logistic regression. Interactions between BMI and reference age were included to allow the associations of BMI and risk to depend
on age at diagnosis. Also, the potential dependence of the associations of DA and NDA with age were evaluated by introducing into the model their interaction with reference age. All
models were adjusted for age at mammography. The potential
confounding effect of the following variables was also examined for all women and for those with mammograms taken
within 5 years since MCCS baseline: BMI at age 18–21
years (missing for 1% of cases and 1% of controls); age at
menarche; parity, lactation, and menopausal status at baseline
(missing for 2% of cases and 2% of controls); oral contraceptive and hormone replacement therapy use (missing for 0.8%
of cases and 0.7% of controls); alcohol consumption and total
Am J Epidemiol. 2014;179(4):475–483
energy intake from diet (no missing data). Because of the
large number of incomplete sets caused by missing values
in the covariates, the adjusted analysis was conducted by applying the unconditional logistic regression model adjusting
for the matching variables (553 cases and 1,594 controls were
without missing values and, of these, 468 cases and 1,352 controls had mammography within 5 years since baseline). Additional potential confounding for family history of breast
cancer was also examined (data for this variable were obtained from BreastScreen Victoria and were missing for 9%
of cases and 11% of controls).
DA and NDA were categorized into quintiles on the basis
of their distributions in the controls, and their dose-response
associations were evaluated by using pseudocontinuous variables defined as within-quintile medians. Variables were standardized to the mean and standard deviation of the controls.
Tests for linear trend were performed on the pseudocontinuous
variables by using the Wald test, and tests for departure from
linearity were performed by using the likelihood ratio test comparing the model with quintiles versus the model with pseudocontinuous variables.
Sensitivity analyses were conducted under the following
scenarios: 1) exclusion of ever users of hormone replacement
therapy; 2) exclusion of cases diagnosed within 2 years of
mammography and their matching controls; and 3) exclusion
of ductal carcinoma in situ cases and their matching controls.
All analyses were conducted by using Stata, version 12.1,
software (StataCorp LP, College Station, Texas).
RESULTS
The characteristics of the study sample are reported in
Table 1. The mean age at diagnosis of the cases was 63 years
(standard deviation, 8.0; range, 48–83 years) with 82% of the
cases diagnosed after age 55 years. Cases were diagnosed, on
average, 8 years after MCCS baseline interview (range, 3
months–16 years), and mammography was performed, on average, 2.8 years (standard deviation, 2.6 years; range, 0–14
years) after MCCS baseline with no difference between
cases and controls. Compared with controls, on average,
cases had higher DA (23 cm2 vs. 17 cm2, P < 0.0001) and
lower NDA (113 cm2 vs. 117 cm2, P = 0.14).
DA and NDA were negatively correlated (Spearman’s rank
correlation = −0.32). BMI was positively correlated with
NDA (Spearman’s rank correlation = 0.62) and negatively
correlated with DA (Spearman’s rank correlation = −0.28).
Table 2 reports the risk estimates for DA, NDA, PMD, and
BMI according to the causal diagrams in Figure 1. Models
1 and 2 refer to Figure 1A with F = DA + γ × NDA and
F = DA / (DA + NDA) = PMD, respectively; models 3 and
4 refer to Figure 1B with BMI or NDA as a proxy for Y.
There was no difference in the goodness of fit between the 2
models in Figure 1A (for model 1, BIC = 1,508 and AUC =
0.68; for model 2, BIC = 1,506 and AUC = 0.68), whereas
models 3 and 4 gave a slightly worse fit (for model 3,
BIC = 1,517 and AUC = 0.67; for model 4, BIC = 1,533 and
AUC = 0.65).
DA was positively associated with breast cancer risk. In
model 1, women in the fifth quintile of DA had a risk of breast
cancer that was 2.73 (95% confidence interval (CI): 1.95,
478 Baglietto et al.
Table 1. Characteristics of Study Participants in a Nested Case-Control Study Within the Melbourne Collaborative
Cohort Study (Recruitment 1990–1994; Follow-up to End of 2007)
Cases (n = 590)
Characteristic
No.
%
487
83
Italy
58
Greece
45
<12
12
Controls (n = 1,695)
Mean (SD)
No.
%
1,400
83
10
164
10
8
131
8
112
19
297
18
117
20
332
20
13
147
25
418
25
≥14
212
36
648
38
Nulliparous
92
16
224
13
Parous and never
lactated
37
6
126
7
450
76
1,317
78
Premenopausal
194
33
556
33
Postmenopausal
395
67
1,136
67
Never
397
67
1,214
72
Ever
188
32
474
28
Never
240
41
654
39
Ever
349
59
1,035
61
No
422
72
1,336
79
Yes
113
19
171
10
Mean (SD)
P Valuea
Country of birth
Australia/New
Zealand/United
Kingdom/others
0.99
Age at menarche, years
0.76
Parity and lactation
Parous and lactated
0.25
Menopausal status
0.97
Hormone replacement
therapy use
0.06
Oral contraceptive use
0.39
Family history of breast
cancer
<0.0001
Table continues
3.83) times higher than that for women in the first quintile,
and the relative risk for 1 standard deviation of DA was
1.50 (95% CI: 1.32, 1.70). There were only small differences
in the estimate of the association between DA and breast cancer risk across the models, suggesting that DA is a risk factor
independent of both NDA and BMI.
NDA was negatively associated with breast cancer risk in
model 1 (relative risk for 1 standard deviation of NDA = 0.75,
95% CI: 0.65, 0.86). In model 4, in which NDA was included
as a proxy for Y, its association was no longer significant.
In model 2, after adjustment for age and BMI, PMD was
associated with breast cancer risk; women with more than
50% density had an increased risk of breast cancer compared
with women with less than 5% density (relative risk = 4.43,
95% CI: 2.64, 7.41). There was no evidence of heterogeneity
by age for the association between DA, NDA, or PMD and
breast cancer risk in any of the fitted models (results not
shown). In all fitted models, the association with BMI increased
with age at diagnosis (P < 0.05 for all tests for interaction).
The ages at diagnosis when BMI was predicted to change
from being protective to being a risk factor were 42 years
in model 1, 46 years in model 2, and 51 years in model 3.
The estimates from the unconditional, fully adjusted model
conducted by using data from women without missing values
in the covariates were very similar to those from the conditional unadjusted model. For model 1, the odds ratios per 1
standard deviation were 1.49 (95% CI: 1.30, 1.69) for DA
and 0.79 (95% CI: 0.68, 0.91) for NDA; in model 2, the
odds ratio for women with more than 50% DA was 3.76
(95% CI: 2.22, 6.38); in model 3, the odds ratio per 1 standard
deviation of DA was 1.52 (95% CI: 1.34, 1.73); in model 4,
the odds ratios per 1 standard deviation were 1.46 (95% CI:
1.28, 1.66) for DA and 0.97 (95% CI: 0.86, 1.10) for NDA.
Similar results were obtained after further adjustment for
Am J Epidemiol. 2014;179(4):475–483
Mammographic Density and Risk of Breast Cancer 479
Table 1. Continued
Cases (n = 590)
Characteristic
No.
%
223
38
Controls (n = 1,695)
Mean (SD)
No.
%
Mean (SD)
P Valuea
Alcohol consumption
Lifetime abstainer
Former drinker
Low intake
618
36
23
4
57
3
272
46
818
48
Medium intake
56
9
160
9
High intake
16
3
42
2
0.90
Age at baseline, years
56 (8)
55 (8)
0.62
Age at mammogram,
years
58 (7)
58 (7)
0.59
5 (4)
5 (4)
0.05
8.6 (3.0)
8.7 (3.4)
0.53
All women
27.1 (5.0)
26.5 (4.8)
<0.01
Premenopausal women
26.1 (5.2)
26.0 (4.9)
0.93
Postmenopausal women
27.6 (4.9)
26.7 (4.8)
<0.001
21.3 (2.8)
21.4 (2.9)
0.43
Total
136 (61)
135 (58)
0.70
Nondense
113 (62)
117 (60)
0.14
23 (22)
17 (20)
<0.0001
Time between age at
mammogram and
reference age, years
Total energy intake,
MJ/day
BMIb
BMI at age 18–21 years
2
Breast area, cm
Dense
Mammographic densityc
19 (17)
15 (16)
<0.0001
Abbreviations: BMI, body mass index; SD, standard deviation.
a
P values refer to Pearson’s χ2 tests for categorical variables and to 2-sided Student’s t tests for continuous
variables.
b
Weight (kg)/height (m)2.
c
Percent mammographic density expressed as the percentage of dense area relative to the total breast area.
family history of breast cancer and when the analyses were
conducted using women with mammograms taken within 5
years since baseline (results not shown).
Figure 2 shows the predicted relative risks for a woman
with given DA and NDA values compared with a woman
with DA, NDA, and BMI values all equal to the medians
in the general population (calculated as the medians for the
control group); for values of NDA in the third quintile,
which is close to the population median, all models provided
similar estimates of the relative risks irrespective of DA,
whereas for more extreme values of NDA, they gave substantially different predictions. For example, compared to those
for a woman with DA and NDA values equal to the population medians, the relative risks for a woman with DA and
NDA values in the bottom quintile (very small breasts) varied
between 1.14 (95% CI: 0.96, 1.36) in model 1 and 0.81 (95%
CI: 0.76, 0.86) in model 3; for a woman with an NDA value in
the top quintile and a DA value in the bottom quintile ( predominantly fatty breasts), the relative risks varied between
0.53 (95% CI: 0.43, 0.66) in model 1 and 0.81 (95%: 0.76,
0.86) in model 3. Excluding ever users of hormone replaceAm J Epidemiol. 2014;179(4):475–483
ment therapy, excluding cases diagnosed within 2 years from
mammogram, or restricting the analyses to invasive cases did
not make major changes to the findings above (results not
shown).
DISCUSSION
Our data show that, consistent with prior reports in the literature, absolute DA and PMD are both positive risk factors
for breast cancer. By using an approach based on causal path
diagrams, we have shown that absolute NDA is inversely associated with breast cancer risk under the assumption that fat
in the body and fat in the breast cause breast cancer through
independent mechanisms, whereas it is not associated with
breast cancer risk under the assumption that they are both
proxies of the same risk factor (i.e., adiposity).
Our results for DA and PMD are consistent with those previously reported (1, 15–20). Also, the finding that the estimates for DA are not influenced by the inclusion of NDA
or BMI in the model is in agreement with a previous report
(8). There is no consensus in the literature about the
480 Baglietto et al.
Table 2. Relative Risk of Breast Cancera for BMIb, DA, NDA, and PMD From a Nested Case-Control Study Within the Melbourne Collaborative
Cohort Study (Recruitment 1990–1994; Follow-up to End of 2007)
Diagram Ac
e
Risk Factor
Diagram Bd
f
Model 1
Model 2
Model 3
g
RR
95% CI
RR
95% CI
RR
95% CI
At age 50 years
1.14
0.93, 1.41
1.08
0.88, 1.32
0.98
0.80, 1.19
At age 70 years
1.62
1.38, 1.90
1.56
1.35, 1.79
1.43
1.25, 1.64
Model 4h
RR
95% CI
BMI per 1 SD
P for interactioni
0.007
0.004
0.003
DA, quintile
First
1.00
Referent
1.00
Referent
1.00
Referent
1.00
Referent
Second
1.14
0.80, 1.63
1.73
1.33, 2.25
1.11
0.78, 1.57
1.04
0.73, 1.47
Third
1.56
1.10, 2.21
2.88
2.15, 3.86
1.52
1.08, 2.14
1.35
0.96, 1.89
Fourth
2.07
1.47, 2.91
4.43
2.64, 7.41
2.03
1.45, 2.83
1.76
1.26, 2.45
Fifth
2.73
1.95, 3.83
1.57
1.39, 1.77
2.72
1.96, 3.78
2.42
1.74, 3.36
1.50
1.32, 1.70
1.53
1.35, 1.74
1.45
1.28, 1.64
DA per 1 SD
NDA, quintile
First
1.00
Referent
1.00
Referent
Second
0.68
0.50, 0.93
0.70
0.52, 0.95
Third
0.58
0.42, 0.80
0.67
0.49, 0.91
Fourth
0.49
0.35, 0.69
0.68
0.50, 0.92
Fifth
0.48
0.33, 0.71
0.87
0.64, 1.18
0.75
0.65, 0.86
0.95
0.85, 1.06
NDA per 1 SD
Abbreviations: AUC, area under the receiver operating characteristic curve; BIC, Bayesian information criterion; BMI, body mass index; CI,
confidence interval; DA, mammographic dense area; NDA, mammographic nondense area; PMD, percent mammographic density; RR, relative
risk; SD, standard deviation.
a
All of the estimates from conditional logistic regression are adjusted for age at mammogram and the variables included into the model.
b
Weight (kg)/height (m)2.
c
DA, NDA, and BMI are risk factors for breast cancer.
d
DA and adiposity are risks factors for breast cancer. NDA and BMI are proxies of adiposity.
e
Model 1: ln(RR) ∼ BMI + BMI × age + DA + NDA; BIC = 1,508; AUC = 0.68.
f
Model 2: ln(RR) ∼ BMI + BMI × age + PMD; BIC = 1,506; AUC = 0.68.
g
Model 3: ln(RR) ∼ BMI + BMI × age + DA; BIC = 1,517; AUC = 0.67.
h
Model 4: ln(RR) ∼ NDA + DA; BIC = 1,533; AUC = 0.65.
i
Likelihood ratio test for the interaction with age at diagnosis.
association between NDA and breast cancer risk; both positive (5) and negative (7–9) associations have been reported.
In a study by Stone et al. (8), the negative association disappeared after adjustment for DA. Our data suggest that the estimate of the association between NDA and breast cancer
depends on the underlying causal model determining the covariates to be included in the model. After taking into account
the age-dependent association of BMI with breast cancer risk,
we observed an increasing strength of association between
BMI and increasing age at diagnosis that was consistent
across the models and has been previously observed using
the MCCS data (3).
Our data slightly favor a causal process in which NDA and
BMI are both determinants of breast cancer risk, suggesting
that in postmenopause, peripheral fat would be positively associated with breast cancer risk, the more so for later-onset
breast cancer, whereas fat in the breast would play a protective effect. The distribution of fat in the body as a risk factor
for breast cancer has some biological plausibility. Peripheral
fat after menopause could increase the risk of breast cancer by
maintaining high levels of estrogens through the aromatization of androgens (21). Instead, fat in the breast tissue could
contribute to a decreased risk of breast cancer through its contribution to vitamin D3–induced growth regulation of ductal
epithelium (10, 22). Also, having an adipose breast could be
indicative of adiposity during childhood (23), which is negatively associated with breast cancer risk possibly because of
an effect mediated by insulinlike growth factor 1 (24, 25). Insulinlike growth factor 1 is a recognized breast cancer risk
factor (26), and its level in adulthood has been found to be
lower in women with a history of obesity at menarche (27).
Moreover, obesity during childhood could result in a decreased cumulative exposure to endogenous estrogens because of increased anovulatory cycles with a consequent
decreased risk of postmenopausal breast cancer (24). It has
also been speculated that the negative association between
Am J Epidemiol. 2014;179(4):475–483
Mammographic Density and Risk of Breast Cancer 481
B)
A)
C)
3.5
3.5
3.0
3.0
2.5
2.5
2.5
2.0
Relative Risk
3.0
Relative Risk
Relative Risk
3.5
2.0
2.0
1.5
1.5
1.5
1.0
1.0
1.0
0.5
0.5
1
2
3
4
5
Dense Area, quintiles
Model 1
Model 2
Model 3
Model 4
0.5
1
2
3
4
5
Dense Area, quintiles
1
2
3
4
5
Dense Area, quintiles
Figure 2. Relative risk of breast cancer for a woman with body mass index (BMI) (weight (kg)/height (m)2) equal to the population median by levels
of dense and nondense mammographic areas compared with a woman with dense and nondense areas equal to the population median. Models 1,
2, 3, and 4 are those described in Table 2. A) Nondense area = quintile 1; B) nondense area = quintile 3; and C) nondense area = quintile 5.
nondense breast tissue and breast cancer risk could be due to
the positive correlation between NDA and breast involution,
which is a protective factor for breast cancer independent of
mammographic density (28).
It has been noted that studies reporting a negative association between NDA and breast cancer risk determined breast
area from the craniocaudal view, whereas those reporting a
direct association used the mediolateral oblique view (11).
The mediolateral oblique view could potentially measure a
greater nondense breast area by including body subcutaneous
adipose tissue that would reflect an increase in BMI rather
than an increase in breast adipose tissue. We measured mammographic density using a craniocaudal view.
We found that, as a linear combination, the best predictor
of breast cancer risk (after adjustment for BMI as an increasing function of age at diagnosis) is F = DA – 0.7 × NDA,
where DA and NDA were fitted as pseudocontinuous variables corresponding to 1 standard deviation increase, and
BMI was adjusted for as a standardized variable. However,
DA and NDA are negatively correlated (Spearman’s rank
correlation = −0.3), so in a statistical sense, the estimates of
association could be reflecting the same underlying phenomenon. If the true causal area (i.e., the area containing the causally
related tissues) is a subset of what is considered by the observer to be “dense” area, the positive association with DA
will be underestimated. This could then lead to a spurious
negative association with NDA.
The main strengths of our study are its size and prospective
design, as well as detailed information on the main risk factors for breast cancer and highly repeatable measurements of
Am J Epidemiol. 2014;179(4):475–483
mammographic density. A possible limitation of our study is
not having measurements of mammographic density at baseline; in fact, all of the measurements refer to mammograms
taken between MCCS baseline and age at diagnosis of matched
cases. However, mammographic density tracks strongly with
age with correlations of 0.8 or more between mammograms
taken 10 years apart (29), and we have adjusted all measurements for age at mammogram. Moreover, the results do not
change when restricting the study sample to women with
mammograms taken within 5 years from baseline. Another
possible limitation is that the study sample has been extracted
from the MCCS participants who attended breast screening
services at least once and might not be fully representative
of the general population.
Determining the most appropriate causal model underlying the relationship between mammographic density, BMI,
and breast cancer risk is essential before including mammographic density in risk prediction models. Our data are better
fitted by a model in which body and breast adiposity are independent risk factors for breast cancer than by a model in
which they are proxies of the same risk factor. However, in
terms of AUC and BIC, the differences between the fitted
models are minimal and do not allow us to draw a definitive
conclusion. We have shown that, for women with NDA equal
to the median in the population, the models are similar in
their predictions, whereas for women with very low or very
high values of NDA, they provide quite different predictions.
A validation study conducted on a sample of women with extreme values of NDA might help to determine which of these
models provides a better fit to the data.
482 Baglietto et al.
Our study adds to the evidence from the literature regarding the positive association between DA and breast cancer
risk, showing that the estimate of the association does not depend on the choice of the causal model. The role played by
NDA is more complex, and we have shown that its association with breast cancer risk is negative or null depending on
the causal model underlying the relationships between DA,
NDA, BMI, and breast cancer risk. Elucidating the biological
mechanisms regulating the distribution of adipose tissue in
the body and the breasts and breast cancer risk might help to
determine the proper causal model and, consequently, to provide better estimates of an individual’s risk of breast cancer.
ACKNOWLEDGMENTS
Author affiliations: Cancer Epidemiology Centre, Cancer
Council Victoria, Melbourne, Australia (Laura Baglietto,
Kavitha Krishnan, Melissa C. Southey, Dallas R. English;
Graham G. Giles); Centre for Molecular, Environmental, Genetic and Analytic Epidemiology, School of Population and
Global Health, University of Melbourne, Melbourne, Australia (Laura Baglietto, Kavitha Krishnan, Jennifer Stone,
Carmel Apicella, Dallas R. English, John L. Hopper; Graham
G. Giles); Centre for Genetic Origins of Health and Disease,
University of Western Australia, Perth, Australia (Jennifer
Stone); Genetic Epidemiology Laboratory, Department of
Pathology, University of Melbourne, Melbourne, Australia
(Melissa C. Southey); Seoul National University, Seoul,
Korea (John L. Hopper); and Department of Epidemiology
and Preventive Medicine, Monash University, Melbourne,
Australia (Graham G. Giles).
Cohort recruitment was funded by VicHealth and the Cancer
Council Victoria. This study was funded by the Breast Cancer
Research Consortium, the National Health and Medical Research Council (grants 251533, 209057, and 504711), and the
National Breast Cancer Foundation and was further supported
by infrastructure provided by the Cancer Council Victoria.
We thank the Victorian Cancer Registry, BreastScreen
Victoria, and the Australian Mammographic Density Research Facility. J.L.H. is a Principal Research Fellow of the
National Health and Medical Research Council. M.C.S. is a
Senior Research Fellow of the National Health and Medical
Research Council.
Conflict of interest: none declared.
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. World Cancer Research Fund/American Institute for Cancer
Research. Food, Nutrition, Physical Activity, and the
Prevention of Cancer: A Global Perspective. Washington, DC:
American Institute for Cancer Research; 2007.
3. Macinnis RJ, English DR, Gertig DM, et al. Body size and
composition and risk of postmenopausal breast cancer. Cancer
Epidemiol Biomarkers Prev. 2004;13(12):2117–2125.
4. van den Brandt PA, Spiegelman D, Yaun SS, et al. Pooled
analysis of prospective cohort studies on height, weight,
and breast cancer risk. Am J Epidemiol. 2000;152(6):
514–527.
5. Lokate M, Peeters PH, Peelen LM, et al. Mammographic
density and breast cancer risk: the role of the fat surrounding
the fibroglandular tissue. Breast Cancer Res. 2011;13(5):
R103.
6. Stone J, Warren RM, Pinney E, et al. Determinants of
percentage and area measures of mammographic density. Am J
Epidemiol. 2009;170(12):1571–1578.
7. Pettersson A, Hankinson SE, Willett WC, et al. Nondense
mammographic area and risk of breast cancer. Breast Cancer
Res. 2011;13(5):R100.
8. Stone J, Ding J, Warren RM, et al. Using mammographic
density to predict breast cancer risk: dense area or percentage
dense area. Breast Cancer Res. 2010;12(6):R97.
9. Torres-Mejia 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.
10. Pettersson A, Tamimi RM. Breast fat and breast cancer. Breast
Cancer Res Treat. 2012;135(1):321–323.
11. Shepherd JA, Kerlikowske K. Do fatty breasts increase or
decrease breast cancer risk? Breast Cancer Res. 2012;14(1):102.
12. Giles GG, English DR. The Melbourne Collaborative Cohort
Study. IARC Sci Publ. 2002;156:69–70.
13. Lohman T, Roche AF, Martorell R. Anthropometric
Standardisation Reference Manual. Champaign, IL: Kinetics
Books; 1988.
14. Hernán MA, Hernandez-Diaz S, Werler MM, et al. Causal
knowledge as a prerequisite for confounding evaluation: an
application to birth defects epidemiology. Am J Epidemiol.
2002;155(2):176–184.
15. Boyd NF, Guo H, Martin LJ, et al. Mammographic density and
the risk and detection of breast cancer. N Engl J Med. 2007;
356(3):227–236.
16. Chiu SY, Duffy S, Yen AM, et al. Effect of baseline breast
density on breast cancer incidence, stage, mortality, and
screening parameters: 25-year follow-up of a Swedish
mammographic screening. Cancer Epidemiol Biomarkers Prev.
2010;19(5):1219–1228.
17. Heusinger K, Loehberg CR, Haeberle L, et al. Mammographic
density as a risk factor for breast cancer in a German casecontrol study. Eur J Cancer Prev. 2011;20(1):1–8.
18. Maskarinec G, Meng L. A case-control study of
mammographic densities in Hawaii. Breast Cancer Res Treat.
2000;63(2):153–161.
19. Wong CS, Lim GH, Gao F, et al. Mammographic density and its
interaction with other breast cancer risk factors in an Asian
population. Br J Cancer. 2011;104(5):871–874.
20. Yaghjyan L, Colditz GA, Collins LC, et al. Mammographic
breast density and subsequent risk of breast cancer in
postmenopausal women according to tumor characteristics.
J Natl Cancer Inst. 2011;103(15):1179–1189.
21. Baglietto L, Severi G, English DR, et al. Circulating steroid
hormone levels and risk of breast cancer for postmenopausal
women. Cancer Epidemiol Biomarkers Prev. 2010;
19(2):492–502.
22. Ching S, Kashinkunti S, Niehaus MD, et al. Mammary
adipocytes bioactivate 25-hydroxyvitamin D(3) and signal via
vitamin D(3) receptor, modulating mammary epithelial cell
growth. J Cell Biochem. 2011;112(11):3393–3405.
23. Sellers TA, Vachon CM, Pankratz VS, et al. Association of
childhood and adolescent anthropometric factors, physical
Am J Epidemiol. 2014;179(4):475–483
Mammographic Density and Risk of Breast Cancer 483
activity, and diet with adult mammographic breast density. Am J
Epidemiol. 2007;166(4):456–464.
24. Fagherazzi G, Guillas G, Boutron-Ruault MC, et al. Body
shape throughout life and the risk for breast cancer at adulthood
in the French E3N cohort. Eur J Cancer Prev. 2013;22(1):
29–37.
25. Fuemmeler BF, Pendzich MK, Tercyak KP. Weight, dietary
behavior, and physical activity in childhood and adolescence:
implications for adult cancer risk. Obes Facts. 2009;
2(3):179–186.
26. Baglietto L, English DR, Hopper JL, et al. Circulating insulin-like
growth factor-I and binding protein-3 and the risk of breast cancer.
Cancer Epidemiol Biomarkers Prev. 2007;16(4):763–768.
Am J Epidemiol. 2014;179(4):475–483
27. Poole EM, Tworoger SS, Hankinson SE, et al. Body size in
early life and adult levels of insulin-like growth factor 1 and
insulin-like growth factor binding protein 3. Am J Epidemiol.
2011;174(6):642–651.
28. Ghosh K, Vachon CM, Pankratz VS, et al. Independent
association of lobular involution and mammographic breast
density with breast cancer risk. J Natl Cancer Inst. 2010;
102(22):1716–1723.
29. Stone J, Dite GS, Giles GG, et al. Inference about causation
from examination of familial confounding: application to
longitudinal twin data on mammographic density measures that
predict breast cancer risk. Cancer Epidemiol Biomarkers Prev.
2012;21(7):1149–1155.