Prenatal and Perinatal Correlates of Adult Mammographic Breast Density

Background: Adult mammographic percent density is one of the strongest known risk factors for breast cancer. In utero exposure to high levels of endogenous estrogens (or other pregnancy hormones) has been hypothesized to increase breast cancer risk in later life. We examined the hypothesis that those factors associated with higher levels of estrogen during pregnancy or shortly after birth are associated with higher mammographic breast density in adulthood.

Methods: We analyzed data on 1,893 women from 360 families in the Minnesota Breast Cancer Family Study who had screening mammograms, risk factor data, over age 40, and no history of breast cancer. Prenatal and perinatal risk factor data were ascertained using a mailed questionnaire. Mammographic percent density and dense area were estimated from the mediolateral oblique view using Cumulus, a computer-assisted thresholding program. Linear mixed effects models incorporating familial correlation were used to assess the association of risk factors with percent density, adjusting for age, weight, and other breast cancer risk factors, all at time of mammography.

Results: The mean age at mammography was 60.4 years (range, 40-91 years), and 76% were postmenopausal. Among postmenopausal women, there was a positive association of birthweight with percent density (P trend <0.01), with an adjusted mean percent density of 17.1% for <2.95 kg versus 21.0% for ≥3.75 kg. There were suggestive positive associations with gestational age (mean percent density of 16.7% for preterm birth, 20.2% for term birth, and 23.0% for late birth; P trend = 0.07), maternal eclampsia/preeclampsia (mean percent density of 19.9% for no and 14.6% for yes; P = 0.16), and being breast-fed as an infant (mean percent density of 18.2% for never and 20.0% for ever; P = 0.08). There was no association of percent density with maternal age, birth order, maternal use of alcohol or cigarettes, or neonatal jaundice. Except for being breast-fed, these associations showed similar but attenuated trends among premenopausal women, although none were statistically significant. The results for dense area paralleled the percent density results. The associations of gestational age and being breast-fed as an infant with percent density attenuated when included in the same model as birthweight.

Conclusions: Birthweight was positively associated with mammographic breast density and dense area among postmenopausal women and more weakly among premenopausal women, suggesting that it may be a marker of this early life exposure. These results offer some support to the hypothesis that pregnancy estrogens or other pregnancy changes may play a role in breast cancer etiology, and suggest that these factors may act in part through long-term effects on breast density.

Trichopoulos (1) hypothesized that breast cancer might originate from in utero exposure to elevated concentrations of estrogens, and empirical tests of this hypothesis have taken the form of evaluating factors thought to be reasonable surrogates of low or high estrogen concentration during pregnancy or shortly after birth as breast cancer risk factors. For example, greater birthweight and neonatal jaundice, both surrogates of higher estrogen exposure to the fetus/newborn (2, 3), may be positively associated with breast cancer risk, although these findings have not been universal (4-6). Women whose mothers had preeclampsia/eclampsia, a surrogate of decreased estrogen exposure (7), may have a decreased risk of breast cancer (8). There is also some evidence that other prenatal or perinatal factors (which may be surrogates for estrogen levels) may also be associated with breast cancer risk, but data are only available from a limited number of studies and are somewhat conflicting (4-6).

Whether defined by the parenchymal pattern or percent mammographic density, the radiographic appearance of the breast has been shown to be a major risk factor for breast cancer (9-11). Many breast cancer risk factors have been shown to be risk factors for higher mammographic density, although the strength and consistency of these associations have varied (10, 11). However, only two studies have evaluated prenatal or perinatal factors with mammographic features (12, 13). We hypothesized that those factors associated with higher levels of estrogen during pregnancy or shortly after birth (i.e., neonatal jaundice) would be associated with higher levels of mammographic breast density in adulthood.

The baseline enrollment (14) and first follow-up (15) for this study population have been previously described. Briefly, breast cancer probands seen at the Tumor Clinic of the University of Minnesota Hospital between 1944 and 1952 (n = 544) were enrolled into a family study. From 1990 to 1996, 426 families were updated, and each proband's first and second degree female relatives and spouses of male relatives were contacted, and extensive risk factor data were collected by telephone interview on 6,194 women.

In 2001, questionnaires were mailed to all female blood relatives and spouses of male blood relatives in the 426 pedigrees who completed the first follow-up survey. Nonresponders were contacted by telephone to complete priority questions and obtain consent to retrieve mammograms. Of the 6,194 eligible women from the first follow-up, 604 were deceased (9.8%), 654 were lost to follow-up (10.6%), 1,109 refused (17.9%), and 84 required a next of kin (1.4%) to complete the questionnaire. A total of 3,743 women completed the 2001 questionnaire, giving a response rate of 77.1% of those contacted and competent to complete a survey, and an overall participation rate of 60.4% of those who participated in the first follow-up. The 2001 questionnaire ascertained updated cancer information, new exposure data on early childhood and adolescence exposures, and authorization for release of mammograms. Authorization for mammogram retrieval was unavailable for 295 women and not attempted for 539 (because of few longitudinal mammograms available), leaving 1,893 available for study (69.4% of women providing authorization and 61.5% of the age-eligible women without breast cancer who responded to the follow-up 2 survey).

For this cross-sectional analysis, we excluded women with breast cancer and used the mammogram with the date closest to the date of the first follow-up (1990-1996), which was within 1 year for 39% of the women, and within 5 years for over 90% of the women. In a comparison of women with and without percent density estimates, there were no differences in education level, weight, age at menarche, adult alcohol use, birthweight, mother's age at first birth, eclampsia/preeclampsia and neonatal jaundice, and trivial differences in parity/age at first birth, birth order, gestational age, breast-fed as an infant, and maternal use of alcohol or tobacco during pregnancy. However, women with density measures were on average older (58 versus 53 years), had more longitudinal mammograms (another aim of the study), were more likely to be postmenopausal (76% versus 59%), and had ever used hormone replacement therapy (HRT; 47% versus 34%).

Mammograms obtained on the 1,693 women from the 2001 follow-up were digitized on a Lumiscan 75 scanner with 12-bit grayscale depth. The pixel size was 0.130 × 0.130 mm² for both the 18 × 24 cm² and the 24 × 30 cm² films. Percent mammographic breast density (dense area divided by total area multiplied by 100) and absolute dense area (square centimeter) were estimated from the left mediolateral oblique view of the mammogram using Cumulus, a computer-assisted thresholding program (16). All images were read by a single trained technician with a high intrareader reliability (r > 0.90 for rereading of over 500 duplicate images randomly included throughout the readings).

We modeled the association of prenatal and perinatal factors with density measures using linear mixed effects models. Tests for trends were computed by ordering the categorical variable and performing a Wald test with 1 degree of freedom. These models adjusted for potential intrafamilial correlation by accounting for degree of relationship using a random effect based on a familial kinship matrix (17). We used a four-stage approach to model building. First, we fit a simple model which adjusted for age at mammography, weight at the first follow-up, HRT use (never, former, current), and menopausal status, variables strongly associated with mammographic breast density in this study. This step in the modeling process eliminated 126 women with missing data (effective sample size, 1,767). Second, we fit a full model that included potential confounders: education, age at menarche, parity/age at first birth, oral contraceptive use, alcohol use, and smoking history (all coded as in Table 1). Further consideration of time between survey and mammogram, age at menopause, years since menopause, time since HRT start/stop, and HRT formulation (only available from a later survey) did not alter the results (data not reported). Third, we added a frailty score to account for familial breast cancer risk.⁴

For each woman, a specific frailty score was obtained based on the degree of relationship (kinship) among the women in the family and the pattern of breast cancer in the pedigree. The median of the frailty scores was obtained and used as the cutpoint to classify each woman's frailty score as either high or low. Because the addition of this frailty score did not change the estimates derived by the first two models, they are not reported here. Fourth, we simultaneously modeled several prenatal and perinatal factors that were individually associated with percent density.

Adjusted means were calculated for each of the models using a smearing estimate (18). Means for the simple model were adjusted to reflect the following characteristics: a postmenopausal, 60-year-old woman, with a weight of 150 lbs, and no HRT use. Adjusted means reported from the full model reflect the additional characteristics of a woman having a high school education, being 13 years of age at menarche, having one to two births over age 20, and reporting no oral contraceptive use, no alcohol use, and never having smoked. In addition, separate models were built for pre- and postmenopausal women, with the premenopausal model adjusted means to the same reference values except for age, which was referenced to a 45-year-old woman.

The mean age at mammography was 60.4 years (range, 40-91 years) and 76% of the women were postmenopausal. Table 1 provides descriptive characteristics of the sample with respect to traditional breast cancer risk factors. Of note, postmenopausal women, compared with premenopausal women, were more likely to have three or more births and to have used HRT, but they were less likely to have used oral contraceptives.

There was no association of percent density with maternal age, birth order, or maternal smoking or alcohol use during pregnancy, and these results were similar among pre- and postmenopausal women (Table 2).

There was a positive association of birthweight, modeled as a continuous variable, with percent density in both the simple (P = 0.01) and full (P = 0.01) models (Table 3). Similar associations were apparent when birthweight was divided into quartiles. In the full model, women with a birthweight of <2.95 kg had a mean density of 20.1% compared with 23.0% for women with a birthweight of ≥3.75 kg (P trend <0.01). This association was strongest among postmenopausal women (P trend <0.01), and a weaker and nonsignificant association was seen among premenopausal women (P trend = 0.19).

There was also a suggestive positive association of gestational age with percent density, with a younger gestational age associated with a lower percent density (P trend = 0.07). This association was also stronger among postmenopausal women (P trend = 0.07) than among premenopausal women (P trend = 0.25). Women whose mother had eclampsia/preeclampsia during their pregnancy had a lower percent density (17.7%) compared with women whose mother did not have this condition (21.5%), although this was not statistically significant (P = 0.25). Among postmenopausal women, there was also a suggestive inverse association, although this was not statistically significant (P = 0.16). There was no association among premenopausal women.

There were weak positive associations of jaundice after birth with percent density overall and among pre- and postmenopausal women, although none of the differences approached statistical significance. Overall, if a woman was breast-fed as an infant, there was no significant association with percent density (P = 0.52). However, there was a suggestive association among postmenopausal women (P = 0.08), such that there was a higher density among women who were breast-fed in the full model (20.0% versus 18.2%); the pattern, however, was opposite among premenopausal women.

We also evaluated the association of all of these factors with dense area, and the results strongly paralleled the results seen for percent density (data not shown).

We next evaluated whether the association of birthweight with percent density among postmenopausal women was influenced by other early life factors. When we included birthweight and gestational age in the same full model, the association with gestational age was eliminated (P trend = 0.25), whereas the association with birthweight was unchanged (P trend <0.01). Similarly, when birthweight and having been breast-fed as an infant were included in the same model, the suggestive association with being breast-fed attenuated (P trend = 0.14), whereas the association with birthweight was unchanged (P trend <0.01). There was no evidence that birth order or preeclampsia/eclampsia confounded the association of birthweight with percent density (P trends for birthweight were both <0.01). Finally, we evaluated birthweight and adult height in the same model; the association for birthweight was only slightly attenuated (P trend = 0.04).

Our major finding was that birthweight was positively associated with adult mammographic percent density and dense area, particularly among postmenopausal women. These associations did not seem to be confounded by traditional breast cancer risk factors or other early life factors. Among postmenopausal women, gestational age was also positively associated with percent density, but this association was eliminated after adjustment for birthweight. Strengths of this study include the use of a defined, community-based study population. There were detailed risk factor data collected over the life span and around the time of mammography, and the statistical approach included careful adjustment for these potential confounders. Percent mammographic density was estimated using a semiautomated method assessed by a single trained reader with excellent intrareader reliability. Mammographic percent density has been shown to have a genetic influence (19, 20), and our family-based design permitted adjustment for this effect.

There are also limitations. All of the exposures evaluated here were self-reported and there was no attempt at validation. Validity of self-reported early life characteristics, including birthweight (21-24), gestational age (22), being breast-fed as an infant (21), and maternal smoking (22), has been shown to be reasonably high and of sufficient accuracy for epidemiologic studies. One study found that being younger at time of interview and the eldest in the family was an important predictor of reporting an accurate birthweight (24). However, adjustment for these factors did not alter our results. We also had a large amount of missing data for many of these variables, with the most extreme being birthweight, where 47% of women in the sample did not know their birthweight. Whereas this is a large amount of missing data, other studies of similarly aged women have reported between 20% and 72% of the participants were unable to self-report their birthweight (23-27). In our study, women with missing birthweight had mean percent density similar to the overall mean of women with birthweight data after accounting for multiple other factors (Table 3). Thus, we have no clear evidence that missing data has introduced a large and systematic bias.

The design of the study allowed us to compare differences in adult and early life risk factors between women who did and did not provide a mammogram. These were generally small and suggest that our sample was representative of an older, more frequently screened population. The mammograms were taken as a part of routine clinical practice from hundreds of different facilities over time, raising concerns about variability in acquisition variables and film quality on density estimation. However, most of the mammograms were taken in the late 1980s and early 1990s, when film technology and more rigorous accreditation standards were in place. Furthermore, the variability introduced across radiologic practices would be expected to be independent of birth characteristics and mammographic density, and thus would most likely attenuate associations to the null. Finally, although the study is generalizable to white women of European descent, generalizability to other racial/ethnic groups is unknown.

To our knowledge, this is the first study to evaluate percent mammographic density and dense area with prenatal and perinatal factors. The only other studies to evaluate these associations (12, 13) used the subjective Wolfe classification, which is a categorical measure based on both percent density and parenchymal pattern (prominent ducts and dysplasia; ref. 28). Current evidence suggests that percent density is a better predictor of future breast cancer risk than the Wolfe pattern (9, 29, 30).

The positive association of birthweight with mammographic breast density in this study is in agreement with a study of 370 Swedish women, ages 40 to 79 years, with no history of breast cancer, where there was a weak but suggestive positive association of birthweight with the high-risk (i.e., P2/DY) Wolfe pattern [odds ratio, 1.39; 95% confidence interval (95% CI), 0.56-3.47, for >4.0 versus 2.5-2.9 kg; P trend = 0.53; ref. 12]. In a study of 1,298 British women, age 53 years, there was also a weak and not statistically significant association of birthweight with higher risk Wolfe pattern (odds ratio, 1.03; 95% CI, 0.92-1.15, for each SD increment in birthweight; ref. 13). Birthweight may be less strongly associated with the Wolfe classification because there are four discrete categories rather than a continuous measure of percent density or dense area. Of other pregnancy characteristics previously evaluated (12), birth length was weakly associated with the high-risk parenchymal pattern, whereas placental weight was strongly related to this pattern; we did not have data on these factors.

Birthweight has been positively associated with pre- and postmenopausal breast cancer risk in the majority of studies reported to date, although findings are strongest and most consistent for premenopausal breast cancer (4-6). Greater birth length has been weakly but positively associated with breast cancer risk (8, 31-33), whereas gestational age, a strong correlate of birthweight, has shown no association with breast cancer risk in most (25, 26, 34, 35) but not all (32, 36) studies. The biological mechanisms linking birthweight to both mammographic breast density and breast cancer risk in adulthood are not known. The strongest evidence implicates estrogens, which play a role in both breast cancer (37) and breast density (38, 39), although the latter association is not consistent (40) and progestins are also likely involved (41, 42). There is a positive association of maternal endogenous estrogen levels with birthweight (2, 43-45), although fetal levels may differ (44, 46, 47). Experimental data from animal models support a role for prenatal estrogen levels influencing mammary gland structure and function (48, 49) and the development of breast tumors (50). Epidemiologic data are more limited, but the strongest support for the prenatal estrogen hypothesis is that exposure to diethylstilbestrol during pregnancy seems to increase breast cancer risk among daughters (22, 36, 51, 52).

Besides estrogens, other aspects of the hormonal milieu correlated with birthweight may be relevant. Progesterone and sex hormone binding globulin (37, 40, 45), prolactin (40, 45, 53), and testosterone and dehydroepiandrosterone (37, 44, 46) could link birthweight, mammographic density, and breast cancer risk, but data remain limited. Insulin-like growth factor I has also been positively associated with birthweight (47, 54), mammographic density (40, 55), and breast cancer risk (56), but only among premenopausal women.

The prenatal factors evaluated in our study (i.e., maternal age, birth order, maternal smoking or alcohol use during pregnancy) were not associated with mammographic breast density. Maternal parity was weakly and positively associated with high-risk parenchymal pattern in one study (odds ratio, 1.24; 95% CI, 0.74-2.07), whereas maternal age was not associated with risk (12). Although these factors have been associated with altered maternal estrogen levels (2, 44, 57-59), they have not shown any consistent association with breast cancer risk (4-6).

We found slightly higher percent density among postmenopausal women whose mother had eclampsia/preeclampsia, but this was not statistically significant and the association was in the opposite direction of the association observed among premenopausal women. However, we had limited power to assess this association. A single other study reported no association of maternal eclampsia/preeclampsia with high-risk parenchymal pattern (12). Maternal eclampsia/preeclampsia, which is associated with lower levels of estrogen (7), has been inversely associated with breast cancer risk in most (3, 36, 60) but not all (34) studies, although only one of these studies achieved statistically significance (3). Larger studies with a sufficient number of women whose mothers had eclampsia/preeclampsia will be needed to fully address this hypothesis.

Of the perinatal factors evaluated, neonatal jaundice was associated with a small and not statistically significant higher percent density. Neonatal jaundice may be associated with both higher estrogen levels in neonates as well as later breast cancer risk, although this has been reported in only a single study (3). After adjustment for birthweight, we found no association of being breast-fed as an infant and breast density; no previous study has evaluated this association. Being breast-fed as an infant has shown no association (36, 61) or an inverse association (62-64) with breast cancer risk.

In conclusion, birthweight was positively associated with mammographic breast density and dense area among postmenopausal women and more weakly among premenopausal women. These differences were small and of unknown clinical significance, but we were still able to detect these differences during the postmenopausal years. These data suggest that mammographic density may be a marker of the biological effect of birthweight. In addition, these results offer some support to the hypothesis that pregnancy estrogens or other pregnancy changes may play a role in breast cancer etiology, and suggest that these factors may act in part through long-term effects on breast density.

We thank Fang-Fang Wu for estimation of percent density, Erin Carlson and Zach Fredericksen for assistance in data analysis, and Mary Jo Eversman and Jessica Curry for assistance in manuscript preparation.