Effect of Preventive Hormonal Therapy on Breast Density: A Systematic Qualitative Review

Breast density (BD) is recognized as one of the strongest independent risk factors of breast cancer (BC). Unlike most other risk factors, BD can be modified, suggesting that it may be a biomarker for preventive interventions. We conducted a qualitative systematic review to address the effect of preventive hormonal therapy on BD. Among the 26 relevant studies, 10 assessed the effect of tamoxifen on BD (TAM: n = 2 877), 9 that of raloxifene (RLX: n = 1 544), and 7 that of aromatase inhibitors (AI: n = 416). The studies were characterized by a large heterogeneity in designs and in methods of BD measurement. BD could be reduced by TAM (10 studies/10). However, the effect of RLX and AI on BD remains unclear due to conflicting results between studies. Consequently, it is crucial to develop practical, accurate, and reproducible methods of measurement in order to be able to compare the effect of preventive hormonal agents on BD and to determine whether change in BD can be used as a predictor of response to therapy.


Introduction
Breast density (BD) is that proportion of breast occupied by radiological dense tissue reflecting breast tissue composition. Dense areas represent fibroglandular tissue when nondense areas correspond to fatty tissue [1]. BD is recognized as one of the strongest independent risk factors of breast cancer (BC) apart from age and genetic mutations [2,3]. Women in the highest categories of BD have a 4-to 6-fold increased BC risk compared to women in the lowest categories [4]. The association of BD and BC risk is present in all ages and is not an artifact of masking bias [5]. Although aging and overweight are risk factors of BC, BD is negatively correlated with age as well as with body mass index (BMI) [6]. To explain this paradox, it has been hypothesized that BD reflects the cumulative exposure to factors that stimulate growth of breast cells since puberty and influence BC incidence [7][8][9]. Details on available methods of BD measurement have been extensively described including qualitative, semiquantitative, and quantitative computerized, fully or not, automated methods [10,11]. The first visual classification of the appearance of the breast was described by Wolfe in four categories: N1, P1, P2, and DY with density increasing from N1 to DY [1]. The most widely used qualitative classification is the BI-RADs system developed by the American College of Radiology in four descriptive categories: (1) almost entirely fatty, (2) scattered fibroglandular tissue, (3) heterogeneously dense, and (4) extremely dense. The new (fourth edition) BI-RADs involves combined qualitative and quantitative assessments with corresponding quartile of dense areas on the film from <25% to >75% [12]. In the last decade, more studies have been conducted with computer-assisted techniques using digitized copies of the mammogram, full digital mammography, and more recently, magnetic resonance imaging (MRI) in order to obtain more objective assessment. Despite these recent inputs, nowadays it remains unclear whether BD is best expressed in terms of absolute dense area or percentage dense area [10]. Although the mechanisms by which BD affects BC risk are not well understood, an estimated 16% of all BC have been attributed to BD higher than 50% [2]. Unlike most other risk factors for BC, BD can be modified, suggesting that it may be a biomarker for preventive interventions [13]. Postmenopausal hormonal therapy (HT) with combined estrogen and progesterone has been shown to increase BC 2 The Scientific World Journal risk and BD. Recently, it has been suggested that the risk of BC and advanced disease is higher among postmenopausal HT users when they have high BD [14]. Since postmenopausal HT may increase BD, one may also wonder to what extent preventive hormonal agents could reduce BD. Furthermore it has been recently shown that the 12-to 18-month change in BD could be a predictor of response to tamoxifen in the preventive setting suggesting that reducing BD may translate into decreased BC risk [13]. Two groups of hormonal agents have proven efficacy in reducing BC risk in large prospective randomized trials. These include selective oestrogen receptor modulators (tamoxifen, raloxifene) and aromatase inhibitors (AI) (exemestane) [15]. Tamoxifen and raloxifene have been approved by the Food and Drug Administration for reducing BC risk but not by the European Medicines Agency. This paper reviews systematically available data concerning the influence of preventive hormonal therapy on BD.

Material and Methods
Using online databases (Medline, PubMed, Cancerlit, Cochrane Controlled Trials Register, and Google), we conducted searches to identify all published reports dealing with changes in BD associated with preventive hormonal therapy. Since different patterns of BD were identified by Wolfe in 1976, we looked for articles published between 1976 and 2012 [1].
Preventive agents included in this review were tamoxifen (TAM), raloxifene (RLX), and exemestane (EXM) [15]. Results on the effects of anastrozole (ANAS) in the preventive setting are not yet available. However, since ANAS and letrozole (LET) have shown a stronger reduction in the risk of contralateral tumours than Tam in the adjuvant setting, we also included both agents in our review [16,17]. The search strategy included in various combinations the following keywords: "name of the preventive hormonal agent and breast density, mammographic breast density, mammography, MRI, dense breast, breast cancer risk, prevention, hormonal therapy. . .. " If reports identified according to these criteria referred to other papers not identified in the initial search, these were also reviewed when relevant to the main questions. The search was complemented by consulting review articles and BC conference proceedings. All available abstracts were reviewed and the full text of an article was consulted by a different reader (FL) when eligibility was ambiguous.
To be included in this review, studies had to be written in English or French, to provide details on the methods used to determine BD either by mammography (screen analog film and/or digitized) or MRI and to give information on at least one of the following confounding factors: age, BMI, menopausal status, family history, HRT use, previous benign breast biopsy, age at menarche, and age at first birth. Since BD may be influenced by radiotherapy and chemotherapy [18][19][20], studies in BC patients providing no data on BC treatment were excluded.
We also excluded case reports and review studies. Data were extracted from the manuscripts using a standardized methodology and according to the checklist of the PRISMA guidelines [21][22][23].
Quality assessment was performed on a data extraction form developed for the review and based on previously reported method [21]. The following trial "quality characteristics" was assessed: (1) design: appropriateness of the design to evaluate the endpoint (coded as "adequate", "unclear" or "inadequate"); (2) patients selection and eligibility: clear eligibility criteria mentioned and similar baseline characteristics between groups (coded as "done, " "partly done, " and "not done"); (3) missing data: that is, missing mammograms (coded as "yes, " "no, " "unclear"); (4) compliance: report of the measure to which participants complied with taking the preventive hormonal agents (coded as "done" and "not done"); (5) outcome assessor blinding: that is, whether the mammogram reader was blinded to the intervention (coded as "yes, " "no, " "unclear"); (6) reproducibility: data provided on the reliability of the BD measurement at the different time sequence of the treatment (coded as "done, " "partly done, " and "not done"); (7) duration: appropriateness of the duration of the intervention to assess outcome (coded as "yes, " "no, " "unclear"); (8) confounding factors: appropriate testing of major factors that could interact with treatment effect: BD at baseline, age, menopause status, and BMI (coded as "done, " "partly done, " and "not done").
These quality criteria were independently assessed by all investigators except one (CB). For each criterion we obtained five evaluations. Based on these evaluations, we calculated the number of studies that met appropriately each quality criterion. We retrieved 164 abstracts using the keywords; 130 were found to be irrelevant to the topic and 34 full-text papers were selected from the remaining abstracts ( Figure 1). We then excluded 4 case reports studies [24][25][26][27], one study with no data on BC treatment [28], one study with no information on methods of BD measurement [29], one review study [30], and one study with full text in unselected language [31]. We retrieved 26 relevant studies. Among them, 10 studies assessed the effect of Tam [32][33][34][35][36][37][38][39][40][41], 9 that of RLX [42][43][44][45][46][47][48][49][50], and 7 that of AI [51][52][53][54][55][56][57] on BD ( Figure 1). We evaluated the methodology, the characteristics of the studied populations, confounding factors, and the results of these 26 remaining studies. The details of study design were submitted in June 2012 to the Prospero database and registered under the number CRD42012002536.  baseline characteristics between groups were provided in 40%. Data on missing mammograms were detailed in 30% of the studies. Compliance with medication was reported in 17%. A clear and realistic attempt to mask the mammogram reader about the treatment was detailed in 58%. Data on the reliability of the BD measurement were mentioned in 17%. Study duration was quoted as adequate in 44%. Finally, description of confounding factors was adequately assessed in 10% of the reviewed trials. (Table 1). Among the ten studies having assessed the effect of Tam on BD, we found one randomized trial (RT) [33] and two post hoc studies of a subset of mammograms from patients included in RT [32,35] ( Table 1). One study was a nested case control analysis from a prospective RT [41]. Five studies were retrospective, among them three were matched with controls [36,37,39]. Finally, since one study retrospectively selected patients from another protocol to form a control group, we classify this study as retrospective [34].

Tamoxifen: Populations and Design
The number of women included in these ten studies ranged from 16 to 1065 (total = 2.877). Four studies included women at high risk of developing BC, based on family history of BC or proliferative benign breast disease diagnosis (ductal carcinoma in situ, lobular carcinoma in situ, or atypical hyperplasia) or a Gail 5-year risk for BC higher or equal to 1.7% [34,35] or 1.3% [33]. In one of these studies, patients who developed small invasive BC within a period of 3 years before randomization were also included [33]. One additional nested case-control study within a randomized prevention trial of Tam versus placebo (IBIS 1) aimed to investigate the relationship between change in BD under treatment and known BC risk factors. This trial included women diagnosed with BC (cases), but no matching occurred for the control subjects [41].
The five remaining studies evaluated BC patients who received Tam alone or Tam combined with chemotherapy and/or radiation therapy after surgery. Among them, two were case control studies with various control groups [37,39]. None of these two studies provided data on whether BC patients were matched for treatment such as chemotherapy and/or radiation therapy or not ( Table 1). The mean age ranged from 43 to 67 years and in four studies, patients were mainly younger than 50 years of age. Most studies (9 out of 10) included some postmenopausal patients, but none explained clearly the criteria used to define menopause. The total duration of the Tam studies ranged from 1.5 to 6 years. (Table 1). Most studies evaluated change in BD under Tam treatment using visual qualitative or semiquantitative methods ( = 7, Table  1). In three studies, the authors described various visual methods to score BD on analog mammograms [35,37,38]. Only four studies mentioned data on missing mammograms at baseline ranging from 10% to 45% [32,33,35,41].

Tamoxifen: Methods of BD Measurement
In six out of eight studies assessing BD in BC patients or in patients who had undergone biopsy for high risk lesions, the unaffected breast free of any surgical intervention was selected for BD measurement [34,36,37,[39][40][41], but information on the side of breast assessed was not available in the two remaining studies. BD was determined in nine studies either on screen film ( = 6) or on digitized mammograms ( = 3) ( Table 1). Six out of these nine studies provided information on incidence used: craniocaudal (CC: = 2) [34,39]; mediolateral oblique (MLO: = 2) [37,41]; both ( = 2) [35,38]. Data on BD at baseline were provided in various manners: the proportion of the total breast area that was composed of dense tissue in percent ( : 5) [32][33][34][35]41], the BI-RADs or Wolfe classification ( = 3) [36,37,39], a ratio dense/fat ( = 1) [38], and MRI volumetric changes ( = 1) [40]. When available, BD at baseline ranged from 31.9% to 60.5% (Table 1). BD was determined by one reader in three studies [33,36,41] and by at least two readers in six [32,34,35,[37][38][39]. Three authors provided no information on whether readers were blinded or not to the trial information. In four studies, readers were blinded to treatment arm [32,33,39,41]. Among them, only one mentioned blinding for time sequence of the mammograms [33]. Finally, in two studies, the authors mentioned that "the readers did not have knowledge of the outcome of the first review" [34,35]. The reproducibility of BD measurement was available in seven studies using various endpoints (Table 1). These endpoints include: interobserver agreement (correlation between different observers) in two studies [32,41]; intra-observer agreement (correlation between different readings of the same observer) in two studies [35,36], and agreement between methods of BD measurement in one [39]. One study provided reproducibility data according to the method of BD measurement. For the computer-aided calculation, reproducibility was determined as the differences between the rescored and original measurements [34]. Finally, one study [38] mentioned that parallelism in the findings of the two radiologists occurred in 24 out 4 The Scientific World Journal For 40 women BD was not measured at baseline. DB with digital measurement: 16% lower compared to those with analog film screen. DCIS: ductal carcinoma in situ. LCIS: lobular carcinoma in situ. NA: nonavailable or not applicable; SD: standard deviation. Wolfe classification: N1: nondense, no ducts visible; P1: prominent ductal pattern occupying less than one-fourth of the breast; P2: prominent ductal pattern occupying more than one-fourth of the breast; DY: homogenous, plaque-like areas of density. 4 Personal calculation for percentages. 5 For 24 cases: results revealed parallelism in the findings of the two radiologists. MD: mammographic density. BI-RADS breast categories: (1) almost entirely fat, (2) fatty with scattered fibroglandular densities, (3) heterogeneously dense breast tissue, and (4) extremely dense breast tissue. 6 i.e., a mother or sister who developed BC before the age of 50.

8
The Scientific World Journal of 27 cases. In four out of seven studies, reproducibility was determined only in a subset of mammograms [32,34,36,41]. (Table 4). Most studies showed that Tam reduced BD significantly. Four studies showed a statistically higher decrease in percentage of BD in the TAM group as compared with placebo or control groups: −13.7% at 54 months [32], −19.6% at 24 months [33], −4.3% estimated reduction per year [34], and −9.4% at 62 months [35] ( Table 4). The number of women experiencing a decrease in BD was determined in five studies after different treatment durations and ranged from 20.59% to 59.8%. Finally, using MRI technology in a cohort study, the authors found a significant absolute reduction of −5.8% ± 3.8% in %BD after a mean duration of 17.5 ± 5.7 months [40]. Data on some confounding factors of BD at baseline were mentioned in five studies (Table 4) [32][33][34][35]41]. Among them, BMI was negatively correlated with BD at baseline in two studies [32,33], and young age was more frequently associated with higher BD in four studies [32][33][34][35], although, not statistically significant in two [34,35]. Premenopause, low predicted familial risk, and never smoking status were significantly associated with higher BD at entry in one study [32]. History of past breast biopsy was associated with higher baseline BD in two studies [32,41].

Tamoxifen: Effects on BD
Some potential confounding factors that could interact with TAM effect were detailed in seven studies in which the authors attempted to evaluate the reduction in BD in different subgroups [32-35, 37, 39, 41] (Table 4). In multivariable analysis, greater reductions in BD were significantly associated with higher BD at entry [32], BMI < 25 kg/m 2 or less [32], stopping HRT during the study [32,41], smoking during the study [32], young age [32,39], past breast biopsy [41], and premenopausal status [32,37]. (Table 2). Among the nine studies having assessed the effect of RLX on BD, we found three RT [42,43,49] and three studies that were either post hoc analyses [46,47] or a substudy [45] performed within RT. Two studies were retrospective without control groups [44,50], and finally one study was prospective controlled [48] ( Table 2). Most of the reviewed studies (8/9) were performed in postmenopausal women. Four out of these eight studies evaluated women with osteopenia and/or osteoporosis [42,46,48,50]. Among them, one study prospectively enrolled women with osteoporosis who received RLX whereas women with osteopenia were enrolled as controls and matched for age at entry and age at menopause [48]. One additional study included women declared to be at risk of osteoporosis and cardiovascular diseases but details regarding these risk factors were missing [49]. Only one study included premenopausal women at high risk of developing BC, based on family history of BC or proliferative benign breast disease diagnosis or a Gail 5-year risk higher or equal to 1.7% [44] ( Table 2).

Raloxifene: Populations and Design
The nine RLX studies totalized 1 544 patients ranging from 37 to 442. The mean age (±SD) of postmenopausal women varied between 50.4 ± 3.6 years and 66.9 ± 5.3 years. In six out of these nine studies, the effect of RLX on BD was compared with the effect of various regimens of hormone replacement therapy (HRT) including combined estrogen progesterone hormone therapy at different doses (cHRT) [42], estrogen only therapy [45,47], tibolone or cHRT [43], tibolone alone [49], and bazedoxifene [46] ( Table 2). Information regarding the menopause definition was provided in five out of eight trials including (1) time since last menstrual period ranging from 6 months to 5 years [42,43,46] and (2) serum levels of follicle stimulating hormone and/or estradiol [48,49]. In one additional study the authors included some patients based on surgical menopause but provided no details on how they confirmed the menopause status of the patients who had not undergone hysterectomy [50]. The interval between HRT discontinuation and study inclusion was mentioned in six out of eight studies [42,43,45,46,49,50] and ranged from 1 months to 24 months. The duration of the RLX studies ranged from 3 months to 36 months (Table 2). (Table 2). In three out of nine studies, calculation of BD was performed using visual qualitative methods alone [42,49,50] (Table 2). Computer-assisted segmentation of digitized mammograms using an interactive thresholding software was utilized in other three studies [45,46,48]. Two authors described personal methods of BD measurement based on (1) the relative volume of dense tissue [43] and (2) an estrogen-specific heterogeneity radiograph score [47]. Finally, in the remaining study, semiautomated calculation of change in breast volume based on MRI technology was used in association with an interactive thresholding software aimed at determining percent BD [44]. Five authors mentioned information on missing or not technically acceptable mammograms at study entry [42,[44][45][46][47] ranging from 0.45% to 48.6%. The number of mammograms readers was available in six out of nine studies and ranged from one to three. Readers were blinded to treatment arm in four studies [42,45,47,49] and both to treatment arm and to time sequence only in one study [46]. Four authors published some data on reproducibility of BD calculations using different endpoints: interobserver reliability at baseline (weighted kappa score: range 0.57-0.70) and after 12 months (range 0.51-0.66) [42], interradiologist correlation at each year of assessment for 3 years and one year after cessation (range 0.63 for 1 year assessment to 0.39 for post treatment calculation) [44], intra-observer variability between baseline and 2-year assessment (range 0.70-0.86) [47], and finally one author mentioned that, in case of discrepancy (9,2-13%), films were reevaluated by the 2 radiologists together for consensus [49]. Seven studies provided information on mammography incidence used: MLO alone [43,47], CC alone [44,46], and both [45,48,49]. When available, the BD at baseline in postmenopausal [45][46][47] and premenopausal women [44] ranged from 8.1% to 27.6% and from 7% to 78%, respectively. (Table 5). Two out of nine studies showed that RLX significantly decreased BD ( Table 5).

Raloxifene: Effects on BD
The Scientific World Journal 9

10
The Scientific World Journal  Weekly patch delivering 0.014 mg estradiol/day. 5 Image Mean Index (IMI): computer-assisted algorithm calculation including identification and delimitation of the same region of interest for each image and the differentiation of gray levels into 10 classes. 6 Wolfe classification: N1: parenchymal pattern composed almost entirely of fat with trabeculae but no visible ducts, P1: pattern composed mainly of fat with fibroglandular tissue that constitutes 25% of the breast, P2: pattern composed of fibroglandular tissue appearing as a heterogeneously dense breast that occupies more than 25% of the breast, and DY: extremely dense tissue. 7 Tibolone versus controls: = 0.32 or RLX versus controls: = 0.18.

The Scientific World Journal
Using MRI technology in high risk premenopausal women, the median relative MRI volume (MRIV) decreased after 1 year and 2 years by 17% (95% CI, −28 to −9; = 0.0017) and 16% (95% CI, −31 to −14; = 0.0004), respectively [44]. In postmenopausal women using a computer-assisted evaluation, the image mean index (IMI) decreased significantly by 1.9% ( < 0.5) after 2 years of RLX therapy [48]. In the seven other studies, no significant modification of BD was observed. Only two studies provided some confounding factors of BD at baseline with discordant results: one study found that BMI was negatively correlated with BD at baseline [47] while this was not the case in the other study [44]. Only one study looked at therapy-by-subgroup interactions and found no statistically significant interaction for BMI, age at entry, menopausal status, use of HRT, baseline BD, and smoking status [45]. (Table 3). We found seven studies assessing the effect of AI on BD (LET: = 5, ANAS: = 1, and EXE: = 1). Among these seven studies, two were RT [51,57] and one was a subgroup analysis of a subset of mammograms from patients included in an RT [52]. Three were prospective single arm trials [53,54,56]. The remaining study was retrospective and case controlled [55]. The number of women included ranged from 16 to 104 (total = : 416).

Aromatase Inhibitors: Populations and Design
Two studies included women at high risk of developing BC based on history of proliferative benign breast disease diagnosis, proven BRCA 1/2 mutation, or a Gail 5-year risk higher or equal to 1.67% [54] or to 8% [53]. In one of these studies, high risk women received HRT [54]. Three studies addressed the issue of BD in BC patients receiving LET or ANAS. Among them, BC patients were selected if they had completed 5 years of TAM therapy [52] or if they were receiving an AI as their only adjuvant systemic therapy [56] or if they had an estimated baseline BD of at least 25% [51]. The two remaining studies determined BD in postmenopausal women receiving EXE [57] or in postmenopausal women receiving HRT alone compared to women receiving HRT + LET [55]. In two out of four controlled studies, the treatment arm and the control group were not perfectly balanced in regard to age, history of benign breast disease, and family history of BC [51] and BMI, number of tumors, and node positive disease [52]. The median age of the selected patients in the seven studies ranged from 50 years (39-68) to 64.6 years . Most studies (6 out of 7) included solely menopausal patients with various menopause definitions according to patients' age. For younger women (less than 50 years of age [53,57] or less than 55 years of age [51]), menopause was defined as no spontaneous menses for at least 12 months and FSH levels in the menopausal range. Two additional studies included patients with hysterectomy and/or bilateral oophorectomy or radiation castration with more than 6 months of amenorrhea [54,56]. Finally, in one study, patients were declared in menopause based on an accepted age for menopause but no additional criteria [55]. Only one author provided data regarding time since menopause [55]. The total duration of the reviewed studies ranged from 6 to 24 months (Table 3). (Table  3). Four studies evaluated change in BD on digitized mammograms using a computer-assisted thresholding program [52][53][54]56]. In three studies, different computer-assisted methods were associated with visual qualitative or semiquantitative measurements based on BI-RADs and/or Boyd classifications [51,55,57]. Data on missing mammograms were provided in five studies [51,53,[55][56][57] and ranged from 7.4% to 28.5%. In all studies assessing BD in BC patients, the unaffected breast was selected for BD measurement [51,52,56]. The CC incidence was used to determine BD in all studies. Data on BD at baseline were provided in various manners. Some used the proportion of dense tissue expressed in percent ( : 5 [51,53,54,56,57]), the Boyd classification scale ( : 1 [52]), and the total integrated pixel intensity (IPI) ( : 1 [55]). When available, BD at baseline ranged from 13.4% to 40% (Table 3). BD was determined by one reader in all studies ( Table 3). Most of the AI studies (6/7) provided information on blinding. In two single arm prospective studies, readers were blinded to time sequence of the mammograms [54,56]. In the remaining studies, readers were declared blinded to time sequence and patients' treatment [51,53,55,57]. Concerning reproducibility of BD measurement, only one author detailed intra-observer reproducibility for qualitative BI-RADs assessment [55] (Table 3). For computerassisted methods only one study provided information on the reproducibility of this analysis and mentioned an intraobserver variation not greater than 10% [52]. (Table 5). Two studies using different methodologies showed that AI significantly decreased BD. Based on a computer-assisted method, one author showed that a statistically significant reduction in integrated pixel index (IPI) occurred in the women who received HT plus LET, whereas no significant change was observed in the women receiving HT alone [55]. In another study using LET, eight out of 16 patients exhibited decreased BD at six months, and eleven out of 16 at 12 months [53] ( Table 5). No data on confounding factors of BD at baseline were mentioned in any AI studies. Potential confounding factors that could interact with treatment effect were detailed in four studies [51,52,56,57] (Table 5). The authors found no statistically significant therapy-by-subgroup interactions for BMI and age at entry to the trials.

Discussion
There is considerable evidence establishing a clear relationship between BD and risk of BC [2,5,[58][59][60]. It has been estimated that an increment of 2% in BC risk exists for each percentage increase in BD [61]. In addition, strong scientific data suggest that TAM, RLX, and EXE are reasonable options for reducing BC risk in women who are at increased risk for developing the disease [15]. Despite all these available data, we found few RT ( : 6) assessing the effect of preventive hormonal agents on BD. Most studies ( : 20) had some collection and selection biases. In addition, there was an important heterogeneity between studies with respect to different confounding factors, selected population, The Scientific World Journal 13 The Scientific World Journal  Thirty-five of 42 subjects were taking estrogen alone (various preparations) and seven estrogen + progesterone. Fifty-nine percent were on transdermal preparations. 4 Study participants: all received letrozole (Femara) 2.5 mg orally 3 times weekly with the exception of 2 women who were given anastrozole (Arimidex) 1 mg/day because of headaches with letrozole. IPI: computerized calculation of integrated pixel intensity. 5 Anastrozole 1 mg daily.

16
The Scientific World Journal  Using analog mammograms only. 3 Using digitized mammograms: SD ± 6.6%, range = −21.5 to 10.1. 4 Because of the small numbers of patients in these groups, the power to detect mean differences (6-8% between them) is low. NMTA: no multivariable analysis of interaction with treatment effect available. 5 There was too much missing data in order to assess the effect of previous use of HRT on change in BD. NA: data non available. 6 Calculated in 27 cases. 7 Evaluated by 3D MRI (magnetic resonance imaging). 8 Personal calculation for percentages. LCIS: lobular carcinoma in situ. 9 After adjusting for age and BD at IBIS-I entry, body mass index at IBIS-I entry, age at menarche, menopausal status at entry to IBIS-I, smoking, family history, benign breast disease (atypical hyperplasia or LCIS), and treatment arm, the mean reduction in breast density was smaller in women who started HRT during the study compared with those who had never taken HRT (4.3% versus 6.3%), mean difference in breast density reduction = 2.02%, 95% CI = 0.02% to 3.83%, = 0.03.

18
The Scientific World Journal  1 The mean density within the placebo group and both raloxifene groups decreased statistically significantly from baseline ( < 0.02 for each group), while the mean density in the ERT (estrogen replacement therapy) group increased but not statistically significantly. 2 E2-HER: computer-based (E2-specific) heterogeneity examination of radiographs. 3 None of the patients in either treatment group demonstrated a decrease in breast density; using area % measurement: no significant increase from baseline. NA: not available. 4 Only in one patient did the BI-RADS classification of 2 change to 3 after 12 months of therapy. 5 HRT: hormonal replacement therapy. 6 IPI: computerized calculation of integrated pixel intensity.

20
The Scientific World Journal number of patients, missing mammograms, methods and reproducibility of BD measurement, and duration of study. Despite this, we found in 19 controlled studies that TAM reduced BD in all studies (8 out of 8), RLX in one out of seven, and AI in one out of four. Several explanations may account for this discrepancy. First, the magnitude of BD reduction and the reliability of the assessment may depend on the method of measurement [62]. For instance, Chow et al. [34] found a significant decrease in density in women at high risk of BC who received TAM, measuring BD with a semiquantitative method but not using Wolfe or BI-RADs classification. The only study relating BD decrease to RLX assessed volumetric BD [63] by automated technique in full-field digital mammograms [43], while the other RLX studies used different methods of BD assessment. Mousa et al. [55] observed, when measuring BD by quantitative image analysis software, a significant reduction in women who received HT plus LET as compared to HT alone. This was not the case when BD was visually analyzed by radiologists. Nielsen et al. [47] determined BD changes via BI-RADs and E2-specific heterogeneity scoring; the latter aims to quantify a specific biological effect on mammographic patterns. They found that RLX modified the E2-specific heterogeneity score, but not BD assessed using BI-RADs. Similarly, Eng-Wong et al. [44] studied the effects of RLX on BD calculated by a semiquantitative thresholding technique and MRI-breast volume (MRIV). They found no significant change in BD calculated on mammograms whereas MRIV decreased during RLX treatment. These results suggest that determination of MRIV changes offers a more reproducible and sensitive measure of fibroglandular tissue [64]. Decensi et al. [33] observed that, at baseline, women with digital measurement had a BD that was nearly 16% lower compared to those with analog-film screen. In this study, BD was determined on digitized mammograms using an interactive threshold method (Cumulus software [61,65]). It has been suggested that density measured with this software is a better predictor of risk than density assessed visually [5] and has an established sensitivity of 5%, allowing the detection of relatively small changes in BD [66,67]. Nevertheless, the evaluation of BD based on mammogram entails major problems: tissue overlapping, positioning difference of the breast, variation in the degree of compression as well as calibration of mammography units, and changes in exposure factors and doses used. These represent additional limitations when using mammograms to assess BD changes over time [68,69]. Although controversial, the incidence use to determine BD may also play a role as it has been shown that density estimates on the CC tend to be higher than on the MLO [62,70,71]. Other potential factors that could have confounded the accuracy of BD measurements include the huge heterogeneity of mammogram incidence used, the lack of systematic evaluation of intra-interreaders reproducibility, and the fact that mammogram readers were not always blinded to time sequence of treatment.
Second, the discrepancy in our results may also be explained by differences in selected populations. The greatest decline in BD typically occurs around 45 years of age and plateaus at approximately 60 years of age [72]. In agreement with this, Cuzick et al. [32] found a significant interaction of treatment effect with age. A minimal decrease in BD was observed for women over 55 years of age treated with TAM (1%), compared to women younger than 45 years of age (13%). Similar results were observed by Meggiorini et al. [39]. Tam studies included larger numbers of patients, both preand postmenopausal, who were also at increased risk of BC, whereas the RLX and AI studies mostly included small numbers of mainly postmenopausal patients. This heterogeneity in the age of the populations is underlined by the variety of baseline BD: in the Tam studies, the RLX studies, and in the AI studies, baseline BD ranged from 31.9% to 60.5%, from 8.1% to 27.6%, and from 13.4% to 40%, respectively. This difference raises the question that perhaps the BD in the RLX and AI treated women may not have been elevated enough to detect a significant change. In addition, menopause is thought to have a more important influence than age on the decline in BD [72][73][74]. Although a standardized definition of menopause is frequently missing, the influence of menopause on BD changes is confirmed in several reviewed studies. Hong and Ki [37] found in BC patients that 87% of premenopausal women with BC had a decrease in BD with Tam use, whereas only 29% of postmenopausal women experienced a decrease. A similar trend was observed in the Brisson's study [35], performed on women with high risk for BC. In a retrospective analysis comparing the effects of bazedoxifene and RLX on BD in postmenopausal women with osteoporosis, Harvey et al. [46] observed that neither significantly decreased BD. In contrast Lasco et al. [48] observed a reduction of BD using RLX in a population of postmenopausal women. However, it should be noted that women enrolled in the Lasco's study were younger (mean age: 52 yrs) than those evaluated in the Harvey's study (mean age: 59 yrs). In addition it is well documented that weight gain is common among women diagnosed with BC [75] and among postmenopausal women [76]; as weight increases, breasts tend to become more lucent. Unfortunately, most of the RLX and IA reviewed studies were not sufficiently powered to properly assess the interaction of BMI with treatment effect.
Third, the discordant results may also be attributed to different biological effects of selected preventive agents. Selective estrogen receptor modulators (SERMs: Tam and RLX) are nonsteroidal compounds that elicit estrogen agonist effects in some tissues, such as bone and the cardiovascular system, and estrogen antagonist effects in others, such as the breast. The tissue specificity of SERMs may be related to the existence of (at least) two different isoforms of the estrogen receptor with distinct signaling properties [77,78]. On the other hand, the AI reduce breast and circulating estrogen levels in postmenopausal women by blocking the conversion of androstenedione to estrone and testosterone to estradiol by cytochrome P450 (CYP) 19, aromatase [79]. In addition, growing evidence shows that BD reflects the degree of stromal and epithelial proliferation and may be closely linked, mostly in premenopausal women, to some growth factor activity such as insulin growth factor (IGF)-1 [11,[80][81][82]. Furthermore, a recent study showed that aromatase immunoreactivity is increased in dense breast tissue and that stromal cells from dense regions have higher levels of The Scientific World Journal 21 aromatase expression than epithelium [83]. However, studies correlating BD with serum estrogen levels have been inconsistent with most studies supporting either no association or an inverse association with estrone or estradiol levels [11]. In this review only four studies evaluated whether preventive agents' effects on BD could be mediated by different biological mechanisms. Two SERMs studies addressed the IGF-1 pathway: Decensi et al. [33] showed that, during the 2-year intervention, Tam significantly lowered IGF-I and BD by 12% and 20%, respectively. Lasco et al. [48] observed that longterm treatment with RLX is able to reduce BD. In women treated with RLX, there was a negative correlation between IGF-1/IGFBP-3 ratio and BD. Although these data confirm previous studies showing that SERMs could decrease IGF-1 and increased IGFBP-3 plasma levels, hence reducing the IGF-1/IGFBP-3 ratio [84][85][86], they could not clarify the link between BD changes and IGF-1 pathway. Conversely, Cigler et al. [51] noted a significant increase in serum IGF-1 levels in the LET group compared to the placebo group. Whereas Fabian et al. [54] found no change in IGF-1 and IGF-1/IGFBP-3 ratio in women receiving LET. Interestingly, it has been recently shown that Tam interacts with the mammary stroma and that these interactions dictate epithelial cell function. Given that BD also is influenced by stromal tissue, this finding suggests a specific effect of Tam on BD [87].
Finally, individual polymorphisms in drug metabolizing enzymes such as CYP2D6 (Tam metabolizing genes) and CYP19 (aromatase genes) may play a role in sensitivity or resistance to preventive therapy [88,89]. All these data underscore potential different mechanisms of action between SERMs and AI on BD but need further confirmation.

Conclusion
In the event that preventive hormonal agents protect women from BC by inducing changes in BD, it may be important to identify women in whom BD decreases and who will benefit from these therapies. In this review, we found that BD could be reduced by TAM. However, the effect of RLX and AI on BD remains unclear due to conflicting results between studies. These differing results highlight numerous biases associated with the design and conduct of trials that assessed BD as endpoint. A major one is the variability of methods of BD measurement retrieved. Consequently, it is crucial to develop practical, accurate, and reproducible methods of measurement in order to determine whether BD can be used as a predictor of response to therapy. Moreover, as new therapies become available in the preventive setting for women at high risk of BC, our results support the need for further larger and well conducted studies aimed to understand whether specific characteristics predict BD changes in response to such therapies.

Implications for Practice
Tam could reduce BD. This reduction is higher in young and premenopausal women. Changes in BD may be a marker of response to Tam therapy [41]. Although the clinical relevance of this potential relationship merits further investigations, there are no immediate implications for practice. No definitive results could be found for RLX and AI.

Implications for Research
In the absence of sufficient data on the effects of RLX and AI on BD in women at high risk of BC, it would be helpful to assess these agents in a large, powered and high quality RT. This trial should last at least 2 years, assess BD with a reproducible and reader-blinded method, and determine if some treatment effect is influenced by other BC risk factors such as age, BMI, family history, past breast biopsy, age at menopause, age at first live birth and use of HRT.