Differences in Measured Mammographic Density in the Menstrual Cycle

Mammography is the most effective means of breast cancer screening. However, it has low sensitivity in women ages <50 years (1, 2); at least partially due to extensive mammographic density in younger women (3). There is some evidence that the amount of dense breast tissue is more extensive in the luteal phase (15-28 days) than in the follicular phase (1-14 days) of the menstrual cycle (4) and the sensitivity of mammography has been reported to be lower during the luteal phase of the cycle (5). Variations in serum levels of estrogen and progesterone over the menstrual cycle, which are associated with variations in the proliferative activity of breast epithelial cells, have been hypothesized to be responsible for variation in mammographic density (4, 6, 7). Combined estrogen/progestin hormone replacement therapy in postmenopausal women increases breast density, whereas estrogen alone does not (8). An increase in mammographic density in the luteal phase of the cycle, when the breast is exposed to both endogenous estrogen and progesterone, is thus a plausible explanation for a reduction in the sensitivity of mammography in premenopausal women. Previous studies that examined the effect of menstrual cycle on mammographic density had either small sample size or used qualitative measures to describe breast density (4, 9, 10). All previous work was done on film mammograms, yet an increasing number of younger women undergo digital mammography, which was shown to have a superior performance over film mammography in premenopausal and perimenopausal women ages <50 years (11).

An improved understanding of the effects of the menstrual cycle on the breast may lead to the development of optimal methods of examining premenopausal women with film or digital mammography and to an improved understanding of the factors that influence mammographic density. If mammographic density is greater during the luteal phase of the menstrual cycle, timing of screening mammography in premenopausal women may be recommended.

The specific aim of the present study was to determine if the time of mammography in relation to menstrual cycle was associated with mammographic density in premenopausal women examined with film or digital mammography.

We identified premenopausal women ages 40 to 49 years who had mammograms in two mammography units in Toronto. After an explanation of the study and screening for eligibility, the subjects were asked for permission to access their mammography files. The mammograms were obtained and measured to determine mammographic density, which was then analyzed in relation to the time in the menstrual cycle at which the mammograms were taken. Ethics approvals were obtained from the University Health Network, Mount Sinai Hospital, and Sunnybrook/Women's College Hospital.

The participants were premenopausal women, ages 40 to 49 years with regular menstrual cycles, without a history of breast disease, who had mammograms in Toronto at Sunnybrook/Women's College Hospital or Mount Sinai Hospital between June 2000 and March 2001. Women at Mount Sinai Hospital were examined either by film or digital mammography, where the decision to examine an individual with film or digital mammography was at the discretion of the radiology technologists. The X-ray detector of the digital mammography system had a smaller area than the cassettes used for film mammography; therefore, technicians preselected leaner women for the examination by this modality.

Women with menstrual cycle length <21 and >35 days, breast reduction or augmentation mammoplasty, hysterectomy, and/or bilateral oophorectomy, or were using hormonal contraceptives or hormone replacement therapy were excluded. In total, 1,000 subjects were recruited into the study, 660 were examined by film and 340 by digital mammography. In total, 64 subjects were excluded due to missing data on important variables (n = 56) or poor quality of images (n = 8), resulting in 620 and 316 subjects for the film and digital subgroups, respectively.

Eligible women provided information on their body weight and height. They were asked to record the first date of their last menstrual period (LMP) and next menstrual period (NMP) after mammography on calendars given to them and to mail the calendar to the study center. A follow-up call was made if the calendar was not received within a month after mammography. In addition, data on reproductive history and personal and family history of breast cancer were obtained by a questionnaire used in the mammography units. During the study period, one of the mammography units discontinued collection of data on parity, which resulted in missing data in about half of the subjects (n = 354, 45.7%) examined by this unit.

To estimate the time when the mammogram was taken in relation to the menstrual cycle, the number of days between the mammogram date and the first day of LMP was calculated and classified into one of four intervals: 1 (1-7 days after LMP), 2 (8-14 days after LMP), 3 (15-21 days after LMP), and 4 (21-35 days after LMP). Data on the first day of NMP were available for 395 (64%) and 214 (68%) subjects examined by film and digital mammography, respectively, which were used to estimate the time of the mammography relative to menstrual cycle using backward calculation from NMP: interval 1 (>21 days before the NMP), interval 2 (15-21 days before NMP), interval 3 (8-14 days before NMP), and interval 4 (≤7 days before NMP).

Comparison of the two methods in the subset where both LMP and NMP were available for estimation of the time of mammography showed that density measurements over the menstrual cycle were similar for each method. The results shown are based on the larger subset and uses forward calculation from the date of the LMP described above.

Film Images. Mammographic density was determined using previously described methods (12). In brief, one cranio-caudal view of the film mammograms was digitized using a Lumisys 85 digitizer. The total breast area and dense area were measured using computer-assisted methods and nondense breast area was calculated by subtracting dense area from total breast area. Average reliability for measuring breast density was assessed by re-reading a 10% random sample of images, and the correlation coefficient was >0.9 both within and between reads.

Digital Images. For the subjects examined by digital mammography, the cranio-caudal view of the digital images processed by the digital mammography machine was used for breast density measurements. All of the digital images were done on a single digital mammography machine (Senographe 2000D; GE Medical Systems). The density measurements were done using the Cumulus software. Average reliability for breast density measurements was assessed by re-reading a 10% random sample of images and was >0.88 both within and between reads.

X-ray tube voltage (kV), product of tube current and exposure time (mAs), compression force (N), and compressed breast thickness were routinely and automatically recorded on the mammography film and digital images.

Data were analyzed using SAS statistical package (SAS Institute). Characteristics of the subjects examined by digital and film mammography were compared using t test and Wilcoxon rank-sum test for continuous and χ² test for categorical variables. t test results are presented here, because they were consistent with Wilcoxon rank-sum test.

Simple and multiple linear regressions were applied to the data. Data from women examined by film and digital mammography were analyzed separately, as there were significant differences between the density measurements of women examined by these methods. Log transformation was applied to nondense area, because it had a skewed distribution. The results from untransformed analysis are shown here, because log transformation did not affect the results. Age and body mass index (BMI) were included in the multiple regression models. Although parity is known to affect mammographic density, it was not included in the multivariate models, as the data were not available for abstraction in more than one-third (n = 360; 38.5%) of the subjects. The overall difference between intervals is tested by P value from ANOVA. P value for linear trend was also calculated and reported to test the trend for change in mammographic density across intervals. Significance level was set at α = 0.05; all tests were two-sided.

Selected characteristics of the study subjects are shown in Table 1. The mean age of the subjects was 45 years and was similar in those examined by film and digital mammography. Compared with women examined by digital mammography, women examined by film mammography weighed, on average, 7 kg more, had significantly higher BMI, and were more likely to be nulliparous. Mean age at menarche, age at first birth, menstrual cycle length, and proportion in first-degree relative with breast cancer were not significantly different between the groups.

Mammographic measures for women examined by film and digital mammography are compared in Table 2 and the distributions of the mammographic measures of dense area, nondense area, and percent density for film and digital subgroups are shown in Fig. 1. The mean values for all mammographic measures were greater for women examined by film mammography.

Mean dense area was 53.1 and 41.3 cm², mean nondense area 95.1 and 82.5 cm², and percent density 40.3% and 35.9% for subjects examined by film and digital mammography, respectively. These differences were only partially explained by the differences in age and BMI. After adjustment for age and BMI, the difference between women examined by film and digital mammography was no longer significant for nondense area (−5.9 cm²; P = 0.06) but was larger for dense area (13 cm²; P < 0.0001) and percent density (8.8%; P < 0.0001).

Breast thickness and the imaging characteristics for the subjects examined by film and digital mammography are shown in Table 2. Mean breast thickness was 52.5 and 49 mm for women examined by film and digital mammography, respectively (P < 0.001). Compression force and the product of X-ray tube current and exposure time were considerably higher for film images (P < 0.001), whereas voltage was slightly higher for digital images (P < 0.001). After adjustment for age and BMI, the difference in the breast thickness between film and digital subgroups was no longer significant (P = 0.39); however, all imaging parameters remained statistically significantly different (P < 0.001).

Table 3 shows the results of mammographic measures, adjusted for age and BMI, for the film mammography data. Mean breast dense area was 55.12 cm² in women in the fourth interval and was higher than the means of the other intervals (51.1-53.9); however, the differences were small and not statistically significant. Mean nondense area was 100.9 cm² during the fourth interval and was larger than the means of the other intervals (91.2-94.8), but the differences between intervals were not statistically significant. The variation in percent density over the menstrual cycle was small and not statistically significant.

The distribution of mammographic measures across the menstrual cycle for the digital mammography subgroup is given in Table 3. Mean dense area was 41.3, 38.3, 41.8, and 44.6 cm² during the first, second, third, and fourth intervals, respectively; however, P values for linear trend and overall P were not statistically significant. In contrast, mean nondense area was 11 cm² higher in the first interval compared with the fourth interval; overall P value and P value for linear trend across intervals were both statistically significant. Percent density was lower in the first two intervals than in the third and fourth intervals; the overall P was not significant, but the P value for linear trend was marginally significant.

The distribution of imaging parameters for film images according to the menstrual cycle showed no statistically significant difference between intervals for the mAs and kV; neither overall P value nor P value for linear trend were significant (Table 4). However, the mean compression force was higher in interval 4 compared with the other intervals; P value for linear trend was significant (overall P = 0.08; P_trend = 0.02). There was no statistically significant difference in any of the imaging parameters according to menstrual cycle for digital data. Adjustment for compression did not affect the results of the mammographic measurements according to menstrual cycle (data not shown).

Meta-analysis and systematic reviews of randomized clinical trials show that mammographic screening reduces breast cancer mortality (2, 13). However, this effect is lower in women ages <50 years (14, 15), which may be at least partially due to the lower sensitivity of mammography in younger women (1). On average, younger women have higher mammographic density, which may interfere with the ability of the radiologist to detect a cancer, thus explaining at least part of the decreased sensitivity of mammography in women ages 40 to 49 years (3). There is some evidence that diagnostic accuracy of mammography varies according to menstrual cycle when mammogram was taken (5). Baines et al. reported that the sensitivity of mammography was 60% when done in the follicular phase and 49% in the luteal phase in the subset of the Canadian National Breast Screening Study. The analysis was limited to women ages 40 to 44 years with regular menstrual cycles (5); the overall sensitivity in women ages 40 to 49 years was 77% (1). Variation in mammographic density was suggested to be the reason for observed differences in mammographic performance.

Previous research concerned with the histologic changes across menstrual cycle suggested that most proliferative and apoptotic activity in the breast occurs in the luteal phase of menstrual cycles with the peak ∼25 to 28 days of the cycle (7, 16). Changes in the breast volume across menstrual cycle measured using magnetic resonance imaging or ultrasound have been reported (17); however, few studies have examined changes in breast tissue composition. Fowler et al. reported a substantial increase in parenchymal volume in the luteal phase of the cycle in a study of 8 women using magnetic resonance imaging (18). However, Graham et al. examined changes in breast water content by Dixon three-point method, which is strongly correlated with percent mammographic density (19), in 7 women with magnetic resonance imaging and observed an average change in water content over the cycle of only 3% (20).

Few studies have examined variation in mammographic density across menstrual cycle. White et al. examined whether mammographic density, reported according to the Breast Imaging Reporting and Data System (predominately fat, heterogeneously dense, and extremely dense), varied according to the phase of the menstrual cycle in 2,591 women ages 40 to 49 years (10). It was found that, in the first, second, third, and fourth weeks of the cycle, respectively, 24%, 23%, 28%, and 28% of subjects had “extremely dense” mammograms, whereas 34%, 29%, 32%, and 31% had “predominately fat” mammograms (P = 0.04). Two studies examined quantitative changes in breast density characteristics during menstrual cycle using a quantitative technique, similar to the one we used in this study, to measure dense, nondense area, and percent density. Ursin et al. examined changes in percent density in 11 subjects ages 30 to 45 years by obtaining two mammograms per subject, one in follicular phase (7-10 days) and the other in luteal phase (24-27 days) of menstrual cycle (9). They observed a small (1.2%) increase in percent density of borderline statistical significance (P = 0.08) in the luteal phase. Buist et al. examined mammographic density in 204 premenopausal women ages 40 to 55 years with regular periods (4). Two screening mammograms were obtained per participant within 9 to 18 months, with one mammogram taken between days 9 and 14 and the other between days 22 and 35 of menstrual cycle. After adjustment for weight change between exams, there was a small nonsignificant increase in percent density in the latter phase (1.1%; P = 0.09). Small increases were also observed in dense area (P = 0.04) and total breast area (P = 0.09).

Our results for film mammography agree with the previous work by Ursin et al. and Buist et al. in that we did not observe significant variation in percent density over the menstrual cycle (4, 9). In women examined by digital mammography, we observed a marginally significant trend of higher percent density during the later phases of the menstrual cycle; however, this association was due primarily to lower mean nondense area rather than an increase in dense area. We would not expect change in percent density due solely to a decrease in the absolute amount of nondense area to influence cancer detection by “masking” cancers.

Digital mammography has been shown to have superior performance over film mammography in breast cancer diagnosis in younger women (11, 21), and the use of digital mammography is increasing. However, all previous work on the relationship of mammographic density and breast cancer risk was done using film mammograms. In our study, percent density was lower in women examined by digital versus film mammography. Women examined by digital mammography were leaner than women examined by film mammography (see Materials and Methods), and because leaner women tend to have smaller breast size and less fat breast tissue (representing nondense tissue), we would have expected percent density to be higher in those examined with digital mammography. After adjustment for age and BMI, the mean percent density was 8.8% lower in women examined by digital mammography compared with film. This difference is similar in magnitude and direction to that observed in the only other study comparing percent density measured from film and digital mammography (22). Harvey reported lower percent density measured from digital compared with film mammograms done on the same subjects, on average, 19 months apart (mean, 32.2% and 40.3%; P < 0.0001, respectively; ref. 22). The author attributed the observed difference to better recognition of skin line on digital images, which resulted in inclusion of more s.c. fat and therefore lower percent density. In our study, after adjustment for BMI, the nondense area was higher in women examined by digital mammography (borderline statistical significance); however, the lower percent density was primarily a result of lower dense area in women examined by digital mammography. More research is needed to explain the differences in density measurements in women examined by digital and film mammography and caution should be exercised in combining film and digital mammograms.

Our study is the first to report on the variation of imaging parameters according to the menstrual cycle. Overall, there were no statistically significant differences in any of the parameters across the menstrual cycle in women examined by film or digital mammography (overall P > 0.05). Nevertheless, we observed higher compression force during the days 21 to 35 of the menstrual cycle for both women examined by digital and film mammography; P value for linear trend was statistically significant only for women examined by film images. This may be at least partially attributed to the breast engorgement during the luteal phase of menstrual cycle (7). Accounting for the differences in the compression did not significantly affect the mammographic measurements according to the menstrual cycle.

One of the limitations of our study was that we had only one mammogram per subject; therefore, we were not able to observe cyclic variations of density within subjects. Also, we had no definitive information on whether the menstrual cycles were ovulatory in nature. It was mentioned earlier that constant estrogen exposure does not affect the amount of breast dense tissue, whereas, combined with progesterone, estrogen increases breast density. Therefore, it is possible that density variations are present only during ovulatory cycles, yet anovulatory cycles are common in premenopausal women (23). We did made efforts to limit participation of women with anovulatory cycles by excluding those with irregular menstrual cycles and those with extremes of cycle length. However, more objective methods may be needed (hormone levels and basal temperature measurements) to identify women with anovulatory cycles.

Misclassification of time of mammography relative to menstrual cycle may be another source of bias in this study. In previous studies, the subjects were classified into luteal and follicular phase based on the day during menstrual cycle when mammography was taken. For the reasons outlined above, we decided to classify subjects into the intervals rather than luteal and follicular phases. Further, we compared two approaches to estimate time of mammography in relation to menstrual cycle (see Materials and Methods) for a subset of subjects who provided data on both first day of LMP and NMP. Because the luteal phase is more stable with average duration of 14 days (24), we expected that the backward calculation from the menstrual period after mammography would provide more accurate results; however, the variation in density measurements over menstrual cycle were similar between these two methods.

In conclusion, if the sensitivity of mammography varies across the menstrual cycle, it is unlikely to be due to the changes in either mammographic density or imaging parameters. Thus, timing of mammography in menstrual cycle is unlikely to have significant influence in breast cancer detection by screening mammography.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.