Mammographic density and breast cancer risk: current understanding and future prospects

Variations in percent mammographic density (PMD) reflect variations in the amounts of collagen and number of epithelial and non-epithelial cells in the breast. Extensive PMD is associated with a markedly increased risk of invasive breast cancer. The PMD phenotype is important in the context of breast cancer prevention because extensive PMD is common in the population, is strongly associated with risk of the disease, and, unlike most breast cancer risk factors, can be changed. Work now in progress makes it likely that measurement of PMD will be improved in the near future and that understanding of the genetics and biological basis of the association of PMD with breast cancer risk will also improve. Future prospects for the application of PMD include mammographic screening, risk prediction in individuals, breast cancer prevention research, and clinical decision making.


Mammographic density and risk of breast cancer
Th e radiographic appearance of the breast on mammography varies among women, as illustrated in Figure 1, and refl ects variations in breast tissue composition and the diff erent x-ray attenuation characteristics of these tissues [3]. Fat is radiologically lucent and appears dark on a mammogram. Connective and epithelial tissues are radiologically dense and appear light. Th is appearance is usually expressed as a percentage of the breast area, or (as referred to here) as percent mammographic density (PMD).
In a systematic meta-analysis of data for more than 14,000 cases and 226,000 non-cases from 42 studies, McCormack and dos Santos Silva [1] reviewed the data on the association of PMD with risk of breast cancer. Th e authors found that PMD was consistently associated with risk of breast cancer. Associations were stronger in studies in the general population rather than symptomatic women, in studies of incident rather than prevalent cancer, and for percent density rather than Wolfe's classifi cation. Wolfe was the fi rst to describe diff erences in breast cancer risk associated with variations in the mammographic appearance of the breast [4,5]; he used four categories: N1 (predominately fat), P1 and P2 (ductal prominence in less than 25% or more than 25% of the breast, respectively), and DY (extensive 'dysplasia'). A quantitative method of measur ing breast density, Cumulus, is illustrated in Figure 1. Th resholds placed at the edge of the breast (red line) and the edge of density (green line) are used to calculate PMD [6]. Table 1 summarizes selected features of the cohort studies, or studies nested within cohorts, that used quanti tative methods to classify PMD [7][8][9][10][11][12][13][14][15]. Th e 10 studies shown were carried out in the US, Europe, or Canada and all found a statistically signifi cant increase in risk associated with more extensive PMD after adjustment for other risk factors, and the increase in risk persisted for at least 8 to 10 years from the date of the mammogram used to classify PMD [9,15]. Th ere is also evidence of a dose-response relationship (that is, of risk increasing with increasing PMD).
Other qualitative classifi cations, such as the fourcategory system developed by the American College of

Abstract
Variations in percent mammographic density (PMD) refl ect variations in the amounts of collagen and number of epithelial and non-epithelial cells in the breast. Extensive PMD is associated with a markedly increased risk of invasive breast cancer. The PMD phenotype is important in the context of breast cancer prevention because extensive PMD is common in the population, is strongly associated with risk of the disease, and, unlike most breast cancer risk factors, can be changed. Work now in progress makes it likely that measurement of PMD will be improved in the near future and that understanding of the genetics and biological basis of the association of PMD with breast cancer risk will also improve. Future prospects for the application of PMD include mammographic screening, risk prediction in individuals, breast cancer prevention research, and clinical decision making.
Radiology (Breast Imaging-Reporting and Data System, or BI-RADS), also create groups with diff erent risks of breast cancer [16,17]. Th e BI-RADS classifi cation of mammographic density has four categories: (1) almost entirely fatty, (2) scattered fi broglandular densities, (3) heterogeneously dense, and (4) extremely dense. BI-RADS is the only classifi cation of mammographic density currently in clinical use in the US but, of the available methods, appears to be the least reliable. Reliability between readers is modest (kappa statistic = 0.56) [18], whereas the interclass correlation coeffi cient for trained readers using Cumulus is more than 0.9 [15]. None theless, the BI-RADS classifi cation does distinguish women at diff erent risks for the development of breast cancer, and a summary by Cummings and colleagues [17] estimated a fourfold gradient in risk between BI-RADS categories 1 and 4.
As shown in Table 2, PMD is associated with risk of breast cancer both at screening and between screening examinations. In the three Canadian studies shown in Table 1 [15], the method of breast cancer detection was recorded by each of the programs. We used those classifi cations to subdivide the breast cancers into those detected at screening, those detected within 12 months of a negative screen, and those detected more than 12 months after a negative screening examination. In a compari son of those with less than 10% density and those with more than 75% density, the odds ratio was 4.74 (95% confi dence interval (CI) 3.0, 7.4) for all cancers. In the 717 cases of breast cancer detected at screening, the odds ratio was 3.52 (95% CI 2.0, 6.2). In the 124 cases of breast cancer detected within 12 months of the last screening examination, the odds ratio for risk of breast cancer in those with more than 75% density was 17.81 (95% CI 4.8, 65.9). For cancers detected more than 12 months after the last screen, the odds ratio for those with more than 75% density was 5.68 (95% CI 2.1, 15.5). Within each category of detection, there was a monotonic increase in risk with each category of density, and the trend tests were all highly signifi cant. Similar results were seen in each of the three screening programs.
More extensive PMD was thus associated with an increased risk of breast cancer at screening, in the presence of potential masking by density. Th e marked elevation in risk associated with PMD in the 12 months after a negative screening examination does, however, probably refl ect the masking of tumors by density. Th e annual incidence of breast cancer associated with diff erent degrees of density may be best estimated by combining the incident cancers detected at screening with those found by other methods in the 12 months following screening [15].

Relative risk
Among other menstrual, reproductive, and familial risks of breast cancer, only age and BRCA carrier status are associated with larger relative risks of breast cancer than PMD (for example, [19]). Th e relative risk associated with density is substantially larger than the relative risk of Figure 1. Examples of mammographic density. (a) 0% mammographic density, (b) less than 10%, (c) less than 25%, (d) less than 50%, (e) less than 75%, and (f) greater than 75%. On the right is an illustration of Cumulus in the measurement of mammographic density. The red line outlines the breast, and the green line outlines the area of density. Republished with permission from [2].
breast cancer associated with a family history of the disease or any of the menstrual and reproductive risk factors.

Attributable risk
Because extensive PMD is common in the population and associated with a large relative risk, if the association with breast cancer risk is causal, the proportion of the disease attributable to this risk factor is expected to be substantial. According to data from three Canadian screen ing programs [15], the risks of breast cancer attributable to density of 50% or more were 16% for all cancers, 12% for screen-detected cancers, 40% for cancers detected within 12 months of a negative screen, and 16% for cancers detected more than 12 months after a screening examination.
For women below the median age of 56 years, the prevalence of density of 50% or more was about three times greater than in older women, in each category of detection, and the attributable risks of breast cancer were 26% for all cancers, 21% for screen-detected cancers, 50% for cancers detected within 12 months of a negative screen, and 28% for cancers detected more than 12 months after a screening examination. Similar estimates of attributable risk have been calculated for PMD in the Breast Cancer Detection and Demonstration project [9]. Th ese estimates of attributable risk are larger than for any other risk factor for breast cancer, including BRCA carrier status, which is estimated to be responsible for 5% or less of all breast cancer [20,21].

Biological plausibility of the association of mammographic density and breast cancer risk
Hypotheses concerned with the biological basis of the association of PMD with risk of breast cancer have been reviewed elsewhere [22] and will be discussed only briefl y here. Th e change in PMD with age refl ects the reduction in glandular tissue and accompanying increase in fat which occur with increasing age. Th is decline in the risk factor of density with age may seem paradoxical, as breast cancer incidence increases with age. However, cumulative exposure to PMD refl ects cumulative exposure of breast stroma and epithelium to hormonal and growth factor stimuli to cell division. Cumulative exposure to PMD increases with age and may be related to the age-specifi c incidence of breast cancer [23]. As reviewed in [22], PMD is also less extensive in women who are parous and in those with a larger number of live births and is reduced by menopause. After adjustment for age and other potential infl uences, a family history of breast cancer is associated with a more extensive PMD [24]. PMD has consistently been found to be inversely asso ciated with body weight. Greater birth weight and adult height have been shown to be positively associated with PMD [25,26] and with an increased risk of breast cancer [27]. With the exception of weight, PMD may be on the causal pathway for breast cancer for some or all of these other risk factors.
Many of the factors that are associated with PMD are also associated with alterations in exposure to hormones that may infl uence the number and proliferative state of epithelial and stromal cells in the breast. To date, most studies of blood levels of ovarian hormones have found either no association or an inverse association with PMD in premenopausal or postmenopausal women (reviewed in [22]). Positive associations with PMD have been found between serum levels of growth hormone and breast water (a surrogate for PMD) in young women from 15 to 30 years old [28], and serum insulin-like growth factor I (IGF-I) levels in premenopausal women and postmenopausal women, and with serum levels of prolactin in postmenopausal women (reviewed in detail in [22]).
Radiologically dense breast tissue -in addition to greater amounts of collagen and cells and greater stained area (on immunohistochemistry) of IGF-I -also have greater amounts of the tissue inhibitor metalloprotease 3 (TIMP-3) [29]. Aromatase immunoreactivity is also associated with dense breast tissue [30]. Th e proteoglycans lumican and decorin have been associated with breast cancer and have also been found to be present in greater amounts in women with extensive PMD [31].

Mammographic density and risk of histological precursors to breast cancer
Mammographic density refl ects the proportions of fat, stromal, and epithelial tissue in the breast and does not denote any histological abnormality [32,33]. Extensive mammographic density is, however, associated with increased risks for the development of most of the histological abnormalities that are non-obligate precursors of breast cancer. Th e breast lesions of ductal carcinoma in situ (DCIS), atypical hyperplasia, hyperplasia without atypia, and columnar cell lesions (CLLs) are, to diff erent degrees, associated with an increased risk of breast cancer, and, as discussed below, risk of each type of lesion is also increased by extensive PMD.
In the Multiethnic cohort, women with more than 50% PMD had, compared with those with less than 10% PMD, an increased risk of both invasive breast cancer (OR = 3.58; 95% CI 2.3, 5.7) and DCIS (OR = 2.9; 95% CI 1.4, 5.9) [26]. A case control study in the Canadian National Breast Screening Study showed that, in women with more than 75% density, compared with those with no density, risk of in situ breast cancer and atypical hyperplasia combined was greater (OR = 9.7; 95% CI 1.7, 53.9), as was risk of hyperplasia without atypia (OR = 12.2; 95% CI 2.9, 50.1) [34]. Additional studies have also shown PMD to be associated with risk of DCIS [35,36].
CLL, thought to be the earliest recognizable histological feature that is a non-obligate precursor to breast cancer, has been found to be more frequent (OR = 2.2; 95% CI 1.03, 4.8) in biopsies from breasts with more than the median density of 30%. CLLs were also strongly positively associated with the percentage of the biopsy occupied by collagen (P = 9.2 × 10 −5 ) and glandular area (P = 2 × 10 −5 ) [37]. Age-related atrophy of breast lobules (lobular involution) has been found to be inversely associated with risk of breast cancer [38], and it appears that PMD and lobular involution are independently associated with risk of breast cancer [39].

Potential improvements in measuring breast tissue composition
All of the methods currently used to assess breast density by mammography have limitations. None takes into account the thickness of the breast, and all are based on the projected area, rather than the volume, of breast tissue. All current methods depend upon a trained observer and thus are subjective. Th ese potential sources of error in measurement are likely to attenuate the observed associations between percent PMD and other risk factors for breast cancer and risk of the disease itself.
To date, three published case control studies have examined the association between percent PMD and risk of breast cancer by measuring breast tissue volumes. One used standard mammography form (SMF) software that uses information about the non-fat tissue in the breast, in conjunction with the thickness of the compressed breast and the breast imaging variables of tube voltage and exposure time, to generate estimates of breast tissue volumes [40]. In an alternative approach to the measurement of tissue volumes, we acquired images prospectively from mammography machines cali brated to allow examination of the relationship between the image signal in each pixel (that is, optical density or blackness of the processed fi lm value), the exposure factors (that is, kilovoltage, milliampere-seconds, tube target, and beam fi lter), and the amount of radiation transmitted by the breast. Corrected breast tissue thickness and breast tissue volumes were calculated [41].
In two of these studies, the volume-based measures of percent density were associated with breast cancer risk, though less strongly than the area-based measures of percent density. It is not yet clear whether these results refl ect as-yet-uncorrected errors in the measurement of breast tissue volumes or the failure to capture additional breast risk information that is present in the area-based measures. An alternative method of measuring percent fi broglandular tissue volumes by using single x-ray absorptiometry has been shown to more accurately predict breast cancer risk than percent dense area [42] but has not yet been replicated or applied to digital mammograms. Other methods of measuring tissue volumes are under development [43,44].
Potential alternatives to the assessment of breast tissue composition by mammography include measurement of the breast water (refl ecting the stromal and epithelial tissue) and fat content by magnetic resonance (MR) and ultrasound tomography (UST). Both have been discussed elsewhere as alternatives to mammography in measuring density [2]. Percent PMD in the mammogram is strongly correlated both with percent water by MR (Spearman r = 0.85; P <0.001) [45] and average sound speed by UST (Spearman r = 0.77; P <0.001) [46].

Etiology of mammographic density
Because PMD is strongly associated with risk of breast cancer, factors that infl uence PMD may also contribute to the causes of breast cancer, and the identifi cation of factors that change PMD may lead to the identifi cation of factors that can reduce the incidence of breast cancer.
Age, parity, and menopausal status (see 'Biological plausibility of the association of mammographic density and breast cancer risk' section above) account for only 20% to 30% of the PMD variation observed in the population [47], and genetic factors might explain a proportion of variation (that is, the heritability) of PMD. Two large, twin studies have added to the evidence that PMD is a heritable quantitative trait. In one, 951 twin pairs (age range of 40 to 70 years) in Australia and North America were recruited, and mammograms and information on the factors associated with variations in PMD were collected. After adjustment for age and other covariates, the proportion of the residual variation in PMD accounted for by additive genetic factors (heritability) was estimated to be 63% (95% CI 59% to 67%) in the combined studies [48]. In a second study, with 553 twin pairs, the propor tion of the residual variation in PMD heritability was estimated to be 53% [49]. Research now in progress seeks to identify genetic variants associated with PMD, and, of the 12 single-nucleotide polymorphisms reproducibly asso ciated with risk of breast cancer, at least 3 have been found to be also associated with PMD [50,51].

Understanding of biological mechanisms
Epithelial and stromal cells, collagen, and fat are the tissue components that contribute to variations in PMD. Th e twin studies described in the previous section indicate that the quantities of these tissue components in the breast are determined largely by heritable factors. Further more, each component has properties that may infl uence the risk and progression of breast cancer.
Breast cancer arises from epithelial cells and the number and proliferative state of these cells may infl uence both the radiological density of the breast and the probability of genetic damage that can give rise to cancer. In addition, collagen and the stromal matrix are products of stromal cells, which may, through mechanical and other properties, facilitate tumor invasion. Interactions between stroma and epithelium are known to infl uence breast development and the changes in breast structure that take place during pregnancy, lactation, and involution and during tumorigenesis. Th e extracellular matrix, which comprises collagens, fi bronectin, laminins, polysaccharides, and proteoglycans, plays a key role in these processes, and there is a large and rapidly growing body of literature on the molecules that mediate how the extracellular matrix infl uences the epithelium (see [52][53][54][55] for reviews). Proteoglycans (see 'Biological plausi bility of the association of mammographic density and breast cancer risk' section above) bind growth factors, contribute to the mechanical integrity of tissues, may refl ect the stiff ness of breast tissue, and can modify tissue behavior [55]. To date, there has been limited application of these basic science fi ndings to understanding the association between PMD and risk of breast cancer. Animal models now being developed may clarify the biological mechanisms that underlie the association of PMD with breast cancer risk.

Potential clinical applications of mammographic density Mammographic screening
Th e evidence given above shows that women undergoing screening for breast cancer with mammography are heterogeneous with respect to cancer risk and the ease with which breast cancer can be detected by mammography. Women with extensive PMD are doubly disadvan taged as they are both at higher risk of developing breast cancer and at greater risk that cancer will not be detected by mammography, because of 'masking' by density of the radiological signs of cancer. In the presence of this underlying heterogeneity in the population undergoing screening, it does not seem likely that screening with a single modality and a single screening frequency will be optimal. It seems possible that, for women with extensive PMD, screening more often than once every 2 to 3 years and with modalities such as MR or UST in addition to mammography would improve cancer detection rates at screening and reduce the frequency of interval cancers. For women with radio-lucent breast tissue and a negative screening mammogram, in whom risk is lower and detection easier, re-screening less frequently than every 2 to 3 years might be safe. Research is required into opti miz ing screening frequency and modality according to the breast tissue characteristics of women. An approach to mammographic screening that starts at age 40 and that bases the frequency of screening on a women's age, breast density (by BI-RADS score), and other risk factors was recently advocated and shown to be cost-eff ective [56]. However, in an editorial accompanying that paper, a number of potential limitations of this approach were raised [57]. Th ese limitations include lack of knowledge of the biological basis of the risk associated with mammo graphic density and of the eff ects of density on the risk and detection of breast cancer subtypes (see 'Breast cancer characteristics and clinical outcomes' section below).

Individual risk prediction
Currently, the most widely used method of predicting risk of breast cancer in individuals is the Gail model [58], which takes into account a woman's age, age at menarche, age at fi rst live birth, number of previous benign breast biopsies, and number of fi rst-degree relatives with breast cancer. Breast density is more strongly associated with breast cancer risk than the other variables included in the Gail model, and the addition of breast density, measured by a manual method tracing, to the Gail model increased predictive accuracy, as shown by the concordance statistic, from 0.607 to 0.642 [59]. Tice and colleagues [60] developed a predictive model for breast cancer by using the BI-RADS classifi cation; the model had a concordance statistic of 0.66. Th e Gail and Tice models have only moderate levels of risk prediction that might be improved by the improvements in measuring breast density described above.

Breast cancer prevention trials
In contrast to most other risk factors for breast cancer, mammographic density can be changed (as described below), suggesting that MD might be used as a surrogate marker in clinical trials of potential approaches to breast cancer prevention. Clinical trials of breast cancer prevention require large numbers of subjects and long periods of observation and thus are expensive. Th e number of subjects required in a breast cancer prevention trial can, however, be reduced by the selection of subjects at increased risk of breast cancer. We have carried out a long-term dietary intervention study in 4,690 women selected because they had mammo graphic density in 50% or more of the breast. During an average follow-up of 10 years (range of 7 to 17 years), invasive breast cancer was detected in 220 women, an observed age-specifi c incidence twice that of women of the same age in the Canadian population followed for the same length of time. However, a potential limitation of the selection of a high-risk group is that the results of such a trial may not be applicable to women who are not at increased risk [61].
It would make possible smaller, shorter, and less expensive trials of breast cancer prevention strategies if there were a breast cancer surrogate that after a short period of observation would allow the identifi cation of interventions that would reduce breast cancer incidence. To be used as a surrogate for breast cancer, a biomarker such as PMD should meet the criteria proposed by Prentice [62] and further by Schatzkin and Gail [63]. Th ese are that (a) the marker should be associated with risk of breast cancer, (b) the marker should be changed by the intervention, and (c) the change in the marker should mediate the eff ect of the intervention on breast cancer risk.
In a case control study nested within the fi rst International Breast Cancer Intervention Study (IBIS), a random ized prevention trial of tamoxifen versus placebo, Cuzick and colleagues [64] showed that, compared with all women in the placebo group, those in the tamoxifen group who experienced a 10% or greater reduction in breast density had a 63% reduction in breast cancer risk, whereas those who took tamoxifen but experienced a reduction in PMD of less than 10% had no risk reduction. In the placebo arm, breast cancer risk was similar in subjects who experienced less than a 10% reduction in PMD and those who experienced a greater reduction. Th e authors conclude that the change in PMD 12 to 18 months after starting treatment is an excellent predictor of response to tamoxifen in the preventive setting [64].
Th ese results (and others) show that PMD is associated with risk of breast cancer and is changed by intervention with tamoxifen. However, although the change in PMD was associated with the eff ect of tamoxifen on breast cancer risk, no evidence is given that the change in PMD mediated the eff ect of tamoxifen on breast cancer risk.
Even if it were convincingly shown that change in PMD did mediate the eff ects of tamoxifen on breast cancer risk, it should not be concluded that all other causes of a reduction in PMD will reduce risk of breast cancer. For example, as discussed above, average PMD decreases with increasing age whereas breast cancer incidence increases with age. A randomized controlled trial of physical activity for 1 year in postmenopausal women, which may reduce breast cancer risk, showed that PMD was increased as a result of the weight loss associated with the intervention [65].
Other interventions that are known to infl uence PMD and breast cancer risk include combined hormone therapy (but not estrogen alone), which increases PMD and risk of breast cancer [66][67][68], and a gonadotrophinreleasing hormone agonist reduces PMD in premenopausal women [69]. It is not yet known whether PMD can be used as a surrogate for breast cancer in any of these settings. In the IBIS trial, the association observed between change in PMD and reduc tion in breast cancer incidence with tamoxifen suggests that change in PMD after the initiation of hormone therapy might be useful in the prediction of eff ect in therapeutic settings. Tables 3 and 4 show, respectively, summaries of published studies that have examined the associations of breast density with tumor characteristics and the clinical course of breast cancer. To date, most studies examining the association of breast density with tumor characteristics have used a qualitative measure of density (for example, BI-RADS), lacked information on covariates, and diff ered in whether and how the cancer was detected (by screening or other means).

Risk of second breast cancer
Studies that have examined risk of a second invasive or in situ breast cancer are summarized in Table 4. Four [79,[90][91][92] of the fi ve [79,[90][91][92][93] studies show an increased risk of a second cancer in the ipsilateral breast, and three [90,91,93] of the fi ve show an increased risk in the contralateral breast. Only one [79] of the three [79,91,93] studies to examine the potential modifying role of radiation therapy found evidence that risk of a second breast cancer was higher in those who did not receive radiation.
Women with higher density have been shown to have a higher risk of dying from breast cancer compared with those with lower density, but this is due largely to the increased breast cancer incidence associated with density [73,94]. In terms of survival after a breast cancer diagnosis, one study reported a non-signifi cant trend to better survival in women with dense breasts [68], and another reported that women with mixed/dense breasts had a signifi cantly lower risk of death from any cause or from breast cancer specifi cally (case fatality rates of 60% and 53%, respectively) compared with women with fatty breasts [73].

Summary
Th ere is now extensive evidence that extensive PMD is a strong risk factor for breast cancer and is associated with large relative and attributable risks for the disease. As discussed above (in the 'Breast cancer prevention trials' section), unlike most breast cancer risk factors, PMD can be changed. Work now in progress is likely to improve measurement of PMD, understanding of the genetics and biological basis of the association of PMD with breast cancer risk, and the clinical signifi cance of change in PMD. Future prospects for the application of PMD include improvements in mammographic screening, risk prediction in individuals, breast cancer prevention research, and clinical decision making.

Competing interests
MJY is a founding partner of, and holds shares in, Matakina Technology (Wellington, New Zealand), a company that develops software for measurement of breast density. The other authors declare that they have no competing interests.