Study Population:
All women were consecutively enrolled in the Study of Health in Pomerania (SHIP),
a prospective population-based cohort study in Northeast Germany (11).
SHIP aims to estimate the prevalence and incidence of risk factors and diseases and to investigate the complex associations among risk factors,
subclinical disorders,
and diseases.
Since 2008,
whole-body MRI has been part of SHIP,
including an optional MRM (12).
Women with known allergies to any kind of contrast agent or drugs were excluded from MRI,
as were pregnant or breastfeeding women.
Of 1475 female study participants,
aged 20 to 83,
who were enrolled in the MRI examination between June 2008 and September 2011,
a total of 651 (44.1%) underwent MRM.
All women underwent a structured interview to obtain information on history of breast diseases,
history of breast surgery including breast implants,
menopausal status,
day of the menstrual cycle,
and medication history including postmenopausal hormone therapy (HT) and oral contraceptives (OC).
Menopause was defined as cessation of menstrual bleeding for at least 12 months.
Body height and weight were measured,
and the body mass index (BMI) was calculated.
SHIP was approved by the institutional review board,
and written informed consent was obtained from each participant prior to enrollment.
Inclusion and Exclusion Criteria:
Inclusion criteria for analysis of normal breast parenchyma were: females > 20 years.
Exclusion criteria were:
- history of recent or previous breast disease or history of breast surgery including breast implants (n=12),
- breasts with complete involution precluding measurement of representative parenchyma (n=68),
- breasts with mass lesions according to the BI-RADS MRI lexicon (n=97),
- perimenopausal women with cessation of menstrual bleeding for less than 12 months (n=15),
- Because previous studies found a relationship between contrast enhancement and use of HT (7) and and OC (4),
these women (n=33,
n=81) were also excluded.
Therefore,
a total of 306 (47.0%) subjects were excluded from the analysis of anthropometric measures and menopausal status,
resulting in a final study population of 345 women.
For the analysis of the influence of hormone use groups of HT and OC users were compared with groups of non-HT and non-OC users.
Dynamic Contrast-enhanced MR Mammography Examination:
MR imaging was performed at 1.5 Tesla on a whole-body MR imager (Magnetom Avanto; Siemens Medical Solutions,
Erlangen,
Germany).
An intravenous access was established,
and the woman was placed prone with the uncompressed breasts suspended in a commercial circularly polarized bilateral breast phased-array receiver coil (Siemens AG Healthcare Sector,
Erlangen,
Germany).
The protocol was identical for all participants and included axial dynamic,
T1-weighted,
time-resolved angiography with stochastic trajectories (TWIST) and three-dimensional imaging (8.86 / 4.51 [repetition time msec / echo time msec]; 25° flip angle; 340 mm field of view; 0.9 mm x 0.7 mm x 1.5 mm voxels).
Following acquisition of the first unenhanced sequence,
an intravenous gadobutrol bolus (Gadovist,
Bayer Healthcare,
Leverkusen,
Germany) was administered with a power injector at a dose of 0.1 mmol/kg body weight at a rate of 1.0 mL/sec,
followed by a saline flush (20 mL) injected at the same rate.
The sequence was repeated five times without time gaps.
Each sequence took 58.27 sec.
Quantitative Analysis of Baseline T1 Signal Intensity and Contrast Enhancement in Normal Breast Parenchyma:
First,
images were postprocessed for quantitative analysis using the Syngo 2008A MultiModality Workplace (Siemens Medical Solutions,
Erlangen,
Germany).
Image subtraction was done to identify any non-mass-like enhancement.
To limit possible bias in the reproducibility of measurement resulting from variable repartitioning of fibroglandular tissue throughout the breast,
measurements were performed in two slices above and two slices below the nipple,
where breast tissue is usually constant and more homogeneous (6,7).
Second,
a region of interest (ROI) was drawn manually to include all fibroglandular tissue of the breast in the four slices selected,
while excluding visible fat,
cysts,
or non-mass-like enhancement.
Third,
a time-signal intensity curve was created automatically for each ROI on a pixel-by-pixel basis representing mean values of T1 signal intensity (SI) and standard deviations for all dynamic frames.
Percent contrast enhancement was calculated as [SI(t1-5) – SI(t0)] / SI(t0)*100,
where SI(t0) is the signal intensity before and SI(t1-5) after gadobutrol administration (6,7,10).
To exclude interreader variability only one radiologist performed all readings.
Statistical Analysis:
Two-level random effects models (13) were applied to analyze BSI and CE,
using the STATA xtmixed routine with six time points at level 1 and individuals at level 2.
Regression model building was based on deviance tests.
Restricted maximum likelihood estimates were applied to determine variance components of the model.
A random intercept model strongly outperformed a model without random effects (Chi²=3701.63,
df=1: p<0.001),
as did a linear time random slope model versus the random intercept model (Chi²=984.11,
df=1: p<0.001).
An unstructured covariance matrix performed better than an independent matrix in the random slope model (Chi²=5.42,
df=1: p=0.02) and was therefore chosen.
All continuous variables were checked for nonlinear associations based on fractional polynomials (FPs) (14).
Second-degree FPs were used with the following selection of power terms {-2,-1,-0.5,0,0.5,1,2,3}.
Selection of powers was based on a cut-off at a=0.10.
Regarding our four continuous indicators,
age,
body weight,
body mass index (BMI),
and body height,
linear models performed best and were used in the analyses.
Full maximum likelihood estimation was used to decide on the inclusion of fixed-effects terms into the models.
In addition to time,
which was included in all models,
we used two sets of predictors: (1) age and the anthropometric measures height and weight,
and (2) menopausal status.
To assess the robustness of our results regarding potential selection bias all analyses were repeated using statistical inverse probability weights.
A p-value <0.05 was considered statistically significant.
Analysis was performed using STATA 12 (StataCorp LP,
College Station,
Texas,
USA) and SPSS 15.0.1 (SPSS GmbH Software,
Munich,
Germany).