Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Mammographic density and associations with breast cancer risk and genetic variation in postmenopausal… Velásquez Garcia, Héctor Alexander 2019

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2019_may_velasquez_garcia_hector.pdf [ 2.55MB ]
Metadata
JSON: 24-1.0378180.json
JSON-LD: 24-1.0378180-ld.json
RDF/XML (Pretty): 24-1.0378180-rdf.xml
RDF/JSON: 24-1.0378180-rdf.json
Turtle: 24-1.0378180-turtle.txt
N-Triples: 24-1.0378180-rdf-ntriples.txt
Original Record: 24-1.0378180-source.json
Full Text
24-1.0378180-fulltext.txt
Citation
24-1.0378180.ris

Full Text

MAMMOGRAPHIC DENSITY AND ASSOCIATIONS WITH BREAST CANCER RISK AND GENETIC VARIATION IN POSTMENOPAUSAL WOMEN:  A CAUSAL INFERENCE APPROACH  by  Héctor Alexander Velásquez García   M.D., Universidad del Rosario, 1996 M.S., The University of California, 2002 M.P.H., Johns Hopkins Bloomberg School of Public Health, 2012    A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in The Faculty of Graduate and Postdoctoral Studies (Population and Public Health)   THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  April 2019  © Héctor Alexander Velásquez García, 2019 ii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled: Mammographic density and associations with breast cancer risk and genetic variation in postmenopausal women: a causal inference approach  submitted by Héctor A. Velásquez García  in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Population and Public Health  Examining Committee: John J. Spinelli, Population and Public Health Supervisor  Kristan J. Aronson, Department of Public Health Sciences, Queen's University Supervisory Committee Member  Carolyn C. Gotay, Population and Public Health Supervisory Committee Member Denise Laronde, Craniofacial Science University Examiner Ying MacNab, Population and Public Health University Examiner  Additional Supervisory Committee Members: Christine M. Wilson, Department of Radiology Supervisory Committee Member  Supervisory Committee Member iii  Abstract The association between mammographic density and breast cancer risk is well known, however, the role of absolute non-dense area remains unclear. Furthermore, even though the associations between breast cancer and known risk factors differ by tumor characteristics, this has not been clearly demonstrated for mammographic density. In addition, not much is known about how the variation in breast cancer related single nucleotide polymorphisms (SNPs) is associated with mammographic density, but given its high heritability, it is possible common genetic determinants could affect both mammographic density and breast cancer.  The objectives of this work were to (1) determine the association between non-dense area and breast cancer, as well as confirming the association between dense area and breast cancer, (2) assess the discriminating power captured by the non-dense area parameter on models forecasting breast cancer, (3) estimate breast cancer risk for mammographic density parameters by breast cancer tumor characteristics, and (4)  evaluate the relationship between polygenic risk scores (PRS) generated developed to predict genetic risk of breast cancer with mammographic density parameters. A population-based case-control study conducted in Vancouver, BC, was used. Detailed questionnaire and clinical information, as well as measured breast density from screening mammography films, were collected. In Chapter 2 estimates of the effects of mammographic density parameters on the risk of breast cancer are computed, using iv  causal inference methods for observational studies. Chapter 3 details the assessment heterogeneity in the relationships between mammographic density parameters and breast cancer risk, according to tumor characteristics. Chapter 4 describes the evaluations of the associations between loci linked in genome-wide association studies to breast cancer risk, and mammographic density parameters, by using recently developed PRS. Non-dense area was found to be an independent risk factor, inversely related to breast cancer risk; however it did not improve prediction over the information given by dense area or percent dense area alone. Mammographic density parameters were not associated to breast cancer tumor characteristics. Finally, limited evidence of shared genetic factors between breast cancer risk and mammographic density was observed. These findings provide important information about the association between mammographic density and breast cancer risk.v  Lay Summary Mammographic or breast density is an important breast cancer risk factor. However, it can be measured in many ways, and one way of measuring it, non-dense area, has not been greatly studied. It also is unclear if the increased risk due to density differs by breast cancer tumor characteristics. Finally, little is known about the genetic determinants of mammographic density.  The present work makes use of modern methods to examine the first two topics. It also evaluates the associations between a combined group of genetic polymorphisms known to be associated to breast cancer, and the different ways to measure mammographic density.  The findings provide further evidence about the importance of mammographic density in the risk of breast cancer, as well as insights regarding potential future methods of breast cancer prevention.  vi  Preface The research chapters in this document were conceived as scientific manuscripts that were published in peer-reviewed journals, or will be submitted for peer-review. This statement certifies that the work contained in this thesis was conducted, analyzed, written and disseminated by Héctor Alexander Velásquez García. The ethical approval for the research described here was provided by the University of British Columbia BC Cancer Research Ethics Board (reference #H14-01614). A summarized version of Chapter 2 was published as “Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, Lohrisch CA, Lai AS, Aronson KJ, Spinelli JJ. Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study. Breast Cancer Research and Treatment. 2018 Jul 7;170(1):159–68”. HA Velásquez García cleaned and organized the data provided by co-author JJ Spinelli, digitized the screening mammographic films made available by the Screening Mammography Program of BC, took mammographic density measurements from the resulting digital images, performed the statistical analyses, and wrote the manuscript. All co-authors took part in reviewing and revising the resultant text. From Chapter 3, a condensed report was accepted for publication, as “Velásquez García HA, Gotay CC, Wilson CM, Lohrisch CA, Lai AS, Aronson KJ, Spinelli JJ. Mammographic density parameters and breast cancer tumor characteristics among vii  postmenopausal women. Breast Cancer: Targets and Therapy. In press 2019”. HA Velásquez García cleaned and organized the data provided by co-author JJ Spinelli, and executed the statistical analyses. All co-authors were involved in the review and revision of the manuscript. A summarized account of Chapter 4 has been prepared with the intent of submission to a peer-review journal, as “Velásquez García HA, Gotay CC, Wilson CM, Lai AS, Aronson KJ, Spinelli JJ. Assessment of associations between breast cancer risk loci and mammographic density parameters using 313-SNPs polygenic risk scores”. HA Velásquez García tabulated the polygenic risk scores developed by the Breast Cancer Association Consortium, computed all statistical analyses, and wrote the manuscript. All co-authors were involved in the review and revision of the manuscript.    viii  Table of Contents Abstract ........................................................................................................................... iii Lay Summary .................................................................................................................. v Preface ............................................................................................................................ vi Table of Contents .......................................................................................................... viii List of Tables .................................................................................................................. xii List of Figures ................................................................................................................ xiv List of Abbreviations ....................................................................................................... xv Glossary ....................................................................................................................... xvii Acknowledgements ....................................................................................................... xix Dedication ..................................................................................................................... xxi 1     Introduction .............................................................................................................. 1 1.1     Overview of Mammographic density .................................................................. 1 1.2     Early density assessment: Wolfe’s parenchymal patterns ................................. 1 1.3     Qualitative assessment of mammographic density ............................................ 3 1.3.1     Boyd’ Six-Category Classification (SSC) ..................................................... 3 1.3.2     Tabár classification ...................................................................................... 3 1.3.3     Breast Imaging Reporting and Data System (BI-RADS) density categories 4 1.4     Quantitative assessment of mammographic density .......................................... 5 1.4.1     Planimetry .................................................................................................... 5 1.4.2     Thresholding ................................................................................................ 6 1.4.3     Automated methods ..................................................................................... 8 1.5     Breast cancer and mammographic density ........................................................ 9 ix  1.6     Epidemiology of breast cancer ......................................................................... 10 1.6.1     Descriptive epidemiology ........................................................................... 10 1.6.2     Reproductive factors .................................................................................. 11 1.6.3     Socioeconomic and lifestyle factors ........................................................... 12 1.6.4     Environmental factors ................................................................................ 13 1.6.5     Hormonal factors and their relationship with mammographic density ........ 14 1.6.6     Genetic and familial susceptibility .............................................................. 15 1.6.7     Specific genetic variation associated with breast cancer ........................... 16 1.7     Objectives ........................................................................................................ 18 1.8     Figures ............................................................................................................. 19 2     Estimating treatment (exposure) effect of non-dense mammographic density over breast cancer risk in postmenopausal women .............................................................. 26 2.1     Methods ........................................................................................................... 26 2.1.1     Study Population ........................................................................................ 26 2.1.2     Questionnaire ............................................................................................ 27 2.1.3     Menopausal status assessment................................................................. 28 2.1.4     Mammographic density measurement ....................................................... 28 2.1.5     Statistical analysis ..................................................................................... 29 2.2     Results ............................................................................................................. 33 2.3     Discussion ........................................................................................................ 36 2.4     Conclusions ..................................................................................................... 39 2.5     Tables .............................................................................................................. 41 2.6     Figures ............................................................................................................. 51 x  3     Associations between mammographic density parameters and breast cancer tumor characteristics ............................................................................................................... 53 3.1     Methods ........................................................................................................... 54 3.1.1     Study Population ........................................................................................ 54 3.1.2     Mammographic density measurement ....................................................... 55 3.1.3     Breast tumor characteristics assessment .................................................. 56 3.1.4     Statistical analysis ..................................................................................... 57 3.2     Results ............................................................................................................. 59 3.3     Discussion ........................................................................................................ 60 3.4     Conclusion ....................................................................................................... 62 3.5     Tables .............................................................................................................. 63 4     Genetic variation associated with breast cancer risk and mammographic density parameters .................................................................................................................... 80 4.1     Methods ........................................................................................................... 82 4.1.1     Study Population ........................................................................................ 82 4.1.2     Mammographic density measurement ....................................................... 83 4.1.3     Genotyping ................................................................................................ 84 4.1.4     Statistical analysis ..................................................................................... 85 4.2     Results ............................................................................................................. 85 4.3     Discussion ........................................................................................................ 86 4.4     Conclusions ..................................................................................................... 88 4.5     Tables .............................................................................................................. 89 5     Conclusion ............................................................................................................. 93 5.1     Summary of Study Findings ............................................................................. 93 xi  5.1.1     Estimation of the causal effect of mammographic density parameters over breast cancer in postmenopausal women .............................................................. 93 5.1.2     Associations between mammographic density parameters and breast cancer tumor characteristics among postmenopausal women ............................... 95 5.1.3     Genetic variation associated with breast cancer risk and mammographic density parameters ................................................................................................. 96 5.2     Strengths and Limitations ................................................................................. 97 5.3     Implications and Future Research .................................................................. 103 5.4     Concluding Remarks ...................................................................................... 106 Bibliography ................................................................................................................ 108 Appendices ................................................................................................................. 155 Appendix A: Causal relationships between mammographic density parameters and breast cancer ........................................................................................................... 155 xii  List of Tables Table 2-1: Inter-rater agreement (ICC) for mammographic density measurements ...... 41 Table 2-2: Characteristics of study population .............................................................. 42 Table 2-3: Marginal odds ratios (augmented inverse-probability weighting estimators) for mammographic density measures ................................................................................ 44 Table 2-4: Comparison of model accuracy for breast cancer ........................................ 45 Table 2-5: Interactions between mammographic density parameters and selected breast cancer risk factors ......................................................................................................... 46 Table 2-6: Marginal odds ratios (inverse-probability weighting estimators) for mammographic density measures; no imputation ......................................................... 47 Table 2-7: Comparison of model accuracy for breast cancer; no imputation ................. 48 Table 2-8: Marginal odds ratios (inverse-probability weighting estimators) for mammographic density measures, eliminating 92 controls with mammograms generated after study enrollment ................................................................................... 49 Table 2-9: Comparison of model accuracy for breast cancer, eliminating 92 controls with mammograms generated after study enrollment ........................................................... 50 Table 3-1: Distribution of tumor characteristics on cases .............................................. 63 Table 3-2: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women .............................................. 64 xiii  Table 3-3: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women (no imputation) ..................... 70 Table 3-4: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women (no late controls) .................. 75 Table 4-1: Characteristics of study population .............................................................. 89 Table 4-2: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population ...................................................... 90 Table 4-3: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population, stratified by menopausal status (premenopausal N=268, postmenopausal N=446) ........................................................ 91 Table 4-4: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population, stratified by family history of breast cancer (negative N=578, positive N=112)* .................................................................... 92       xiv  List of Figures Figure 1-1: Routine screening mammogram views ....................................................... 19 Figure 1-2: Mammograms showing the four Wolfe grades ............................................ 20 Figure 1-3: Six Category System (Boyd’s SSC) for assessment of mammographic density ........................................................................................................................... 21 Figure 1-4: Tabár classification of mammographic parenchymal patterns .................... 22 Figure 1-5: BI-RADS breast composition categories ..................................................... 23 Figure 1-6: Measurement of mammographic density via planimetry ............................. 24 Figure 1-7: Computer-assisted thresholding method (Cumulus software) for the determination of mammographic density ....................................................................... 25 Figure 2-1: Minimally sufficient adjustment set between mammographic density parameters and breast cancer risk in postmenopausal women (causal diagram) ......... 51   xv  List of Abbreviations (Alphabetical order)  AIC: Akaike information criterion BC: British Columbia BCAC: Breast Cancer Association Consortium BI-RADS: Breast Imaging Reporting and Data System BMI: Body mass index CBCS: Canadian Breast Cancer Study CC: Cranial-caudal view CI: Confidence interval DA: Mammographic dense area DAG: Directed acyclic graph ER: Estrogen receptor FFDM: Full-field digital mammography FISH: Fluorescence in-situ hybridization HER2: Human epidermal factor receptor 2 HRT: Hormone replacement therapy  IARC: International Agency for Research on Cancer ICC: Intraclass correlation coefficients xvi  IGF-1: Insulin-like growth factor 1 IHC: immunohistochemistry IPW: Inverse probability weighting LRT: Likelihood ratio test MAF: minor allele frequency MICE: multiple imputation by chained equations MLO: Mediolateral-oblique view NDA: Mammographic non-dense area OR: Odds ratio PDA: Mammographic percent dense area PRS: Polygenic risk score PR: Progesterone receptor ROC: Receiver operating characteristic curve  SD: Standard deviation xvii  Glossary Intraclass correlation coefficient (ICC): Statistic that describes how strongly units in the same group resemble each other. The assessment of consistency of quantitative measurements made by different observers measuring the same quantity is one of the its most important applications1. Inverse probability weighting (IPW): Most of the times in statistics the same importance in given to each observation. Conversely, there are situations when it is fitting to confer modified weights to observations, for instance, when there is missing information, or as a way to deal with confounding2. In this last case, in observational studies, the effect of the exposure of interest can be influenced by factors (confounders) that also affect the outcome, rendering it biased. There are different approaches to attempt to eliminate bias from the estimation of the causal effect of interest. The conventional way to do it is via multiple regression; another possibility is the use of inverse probability of treatment (or more properly “exposure”, in the context of this thesis) weighting. In the last case, the probability of having a particular exposure given the presence of a set of confounders, defines the propensity score (PS), which is used to compute the weights; 1/PS for exposed individuals, 1/(1-PS) for unexposed ones3. Kappa (Cohen’s): Statistic used to assess reliability, comparing the observed accuracy versus the expected accuracy (agreement given by chance) in categorical scales. Cohen’s kappa considers total agreement only; partial agreement is ignored, which may xviii  matter when the evaluated scoring system is ordered (e.g. Glasgow Coma Scale). An extension to this method, the weighted kappa, addresses this issue. Weighted kappa uses a predefined table of weights measuring the degree of disagreement between raters, the higher the disagreement the higher the weight. When the weighting scheme corresponds to the square of the amount of discrepancy, the weighted kappa is equal to the intraclass correlation coefficient4.  xix  Acknowledgements I would like to start by thanking the members of my supervisory committee: Drs. John J. Spinelli, Kristan Aronson, Carolyn Gotay, and Christine Wilson. My sincerest thanks to them for their instrumental contributions, and for providing mentorship and guidance throughout my dissertation. I cannot thank you each enough for what you did for me and my research. I hope the future brings new opportunities to collaborate with and learn from each one of you. Dr. Spinelli deserves special mention because he took a chance on me when I contacted him seven years ago, enquiring about the possibility of having him as my supervisor. Thanks for welcoming me into the PhD program, for providing priceless advice and opportunities, for the endless patience, for help me to improve my skills as researcher, writer, and communicator. Above all, thank you for your friendship. I gratefully acknowledge the funding received towards my doctoral studies provided by the Four Year Doctoral Fellowship Award from the University of British Columbia, and a Canadian Breast Cancer Foundation Fellowship (award #319404). I would also like to thank Dr. Boris Sobolev, whose expertise in the area of causal inference in observational studies was essential to achieve the work I have done. I am grateful for the time and kind assistance contributed by Drs. Gertraud Maskarinec, Jennifer Stone, and Martin Yaffe to ascertain the consistency of my mammographic density readings. I would like to express my deep gratitude to Ms. Karen Locken and Mrs. Christine Lam (BC Cancer, Diagnostic Images), as well as the staff of the Screening Mammography xx  Program of British Columbia, particularly Mrs. Carla Brown-John for their invaluable help. I highly appreciate the extensive support provided by Ms. Anoma Gunasekara, Mr. Gord Mawdsley (Sunnybrook Research Institute), Ms. Agnes Lai, Ms. Zenaida Abanto, Ms. Magali Coustalin (BC Cancer, Cancer Control Research), and the staff of the BC Cancer Registry and Breast Cancer Outcomes Unit. Finally, I would like to thank the British Columbia participants of the Canadian Breast Cancer Study for their inestimable involvement.  xxi  Dedication To my parents, Flor and Héctor, for their unconditional love throughout my life, and whose good example have taught me to do always my best. Without your absolute support I wouldn't be the person that I am today. You are ever-present in my thoughts, and even though during the past six years we have been separated by half a world, I have never felt far away from you. To my mother and father-in-law, Tamara and Yury, for their wise advice and constant encouragement since I decided to return to graduate school. To all my friends, both present locally, as well as hailing from abroad, who have been good listeners, and even better counselors. Lastly, this thesis is dedicated to my beloved wife and daughter. Elena, without your constant support, this work wouldn't have been possible. You are my soul mate, my best friend, my dauntless companion. Thank you for joining me in this ongoing journey, from Carcosa to Celephaïs and beyond. Vasilisa, you continually provided the necessary breaks from my research, as well as the motivation to finish my dissertation. You are the foremost inspiration in my life to keep evolving as a person each day.  1  1 Introduction 1.1  Overview of Mammographic Density Radiographically, breasts consist of two types of tissues: fibroglandular and adipose. Fibroglandular tissue is made up by the combination of connective tissue (stroma) and glandular epithelial cells lining the mammary ducts (parenchyma)5. The mammographic representation of breasts is determined by the different level of X-ray attenuation that each structure has. Fibroglandular tissue has a higher X-ray attenuation coefficient than fat; for that reason it is radiologically dense or radiopaque and it is recorded as light on the mammogram, while fat, being consequently more transparent to X-rays, is radiographically non-dense, and therefore, it is rendered as darker or radiolucent6. The relative presence of the two types of tissue composing the breast can be deduced from the brightness pattern present in a mammographic image7.  Routine screening mammography consists of taking two mammograms of each breast from different views (angles): from above (known as cranial-caudal or CC) and from an oblique position (mediolateral-oblique or MLO) (figure 1-1). Since the goal of the procedure is cancer detection before the onset of clinical signs, it is practiced in women with no symptomatology or evident breast abnormalities.   1.2  Early density assessment: Wolfe’s parenchymal patterns 2  John Wolfe was the first person to realize that the risk of breast cancer was associated to the “parenchymal patterns” that he identified on mammograms, describing this concept in 19768,9. These patterns, known today as Wolfe grades, are, in order of increasing risk10 (illustrated in figure 1-2):  N1 (normal): Breast is mammographically radiolucent, being composed mostly by fat, representing the lowest risk of breast cancer.  P1 (prominent ducts): Mammogram shows prominent ducts that can be found in up to 25% of the breast.   P2 (prominent ducts): Presence of prominent duct pattern that occupies more than 25% of the breast.    DY (dysplasia): Mammary parenchyma is radiopaque, given the presence of dense sheet-like conglomerations of fibroglandular tissue. Denotes the highest level of breast cancer risk. The risk of breast cancer was found to increase from 2 to 3 times when comparing the DY pattern to the N1 pattern, according to the reviews of Wolfe parenchymal patterns performed by Saftlas and Szklo (1987)11, and Goodwin and Boyd (1988)12. The review by Oza and Boyd (1993)13 found that the interobserver agreement of this classification oscillated between 52 to 97%, and its intraobserver agreement ranged from 69 to 97%. 3  Most of later research in this area has tried to quantify objectively mammographic density, given that the increment in breast cancer risk is related to the amount of fibroglandular tissue present7.   1.3 Qualitative assessment of mammographic density 1.3.1  Boyd’ Six-Category Classification (SSC) Proposed by Boyd (1995)14, this semi-quantitative score of six categories describes the percentage of mammary fibroglandular tissue present as adjudicated by an expert observer. Figure 1-3 shows its definitions:  SCC 1 (0%), SCC 2 (>0 − 10%), SCC 3 (>10 − 25%), SCC 4 (>25 − 50%), SCC 5 (>50 − 75%), and SCC 6 (> 75%). The relative risks of developing breast cancer for SCC 3 to 6 are respectively 1.9, 2.2, 4.6, and 7.1 compared to SSC 114. Good agreement between readers (weighted kappa 0.78) has been reported for this classification method; its intra-reader reliability has also been found to be very good (weighted kappa 0.89)15.   1.3.2 Tabár classification Tabár (1997)16 proposed a revised version of Wolfe’s parenchymal patterns, containing five categories or patterns (I to V; figure 1-4), based on a histologic-mammographic correlation based on a three-dimensional, subgross (thick-slice) method. In Tabár’s 4  Pattern I, the breast features borders delineated with semicircles ("scalloped"), regularly dispersed nodular densities (1-2 mm), and oval-shaped radiolucent areas; Pattern II is characterized in turn by complete breast involution (fibroglandular tissue replaced by fat); Pattern III resembles Pattern II, differing in the presence of prominent retroareolar ducts, often associated with periductal fibrosis; Pattern IV shows predominantly nodular and linear densities, due to the presence of proliferating glandular structures in the former and periductal elastosis in the latter; Pattern V exhibits widespread fibrosis with no noticeable characteristics. Patterns I to III are considered low-risk, while patterns IV and V represent high-risk17; the assessment by Winkel et al. (2015) reported age-adjusted odds ratios of 2.61 and 2.51 (two independent reviewers) for combined patterns IV and V compared to pooled patterns I, II and III18. The agreement statistics (unweighted kappa) for Tabár reported by Jamal et al. (2006) are 0.63 for inter-observer agreement and 0.75 for intra-observer agreement19.   1.3.3 Breast Imaging Reporting and Data System (BI-RADS) density categories The Breast Imaging Reporting and Data System (BI-RADS, or BIRADS)20, a scheme used by clinicians to report mammographic findings, is of the most widespread mammographic density classification systems in use presently. It was not designed to estimate risk; it was conceived to declare the degree of concern that a malignant lesion may be masked on a mammogram by dense tissue, and consequently may possibly be missed. The mammographic sensitivity is decreased in dense breasts scenarios21,22; an elevated BI-RADS score informs the physician that tests more robust to density (such 5  as diagnostic mammography) may be in order when breast cancer is a likely concern (like in family history in 1st degree). Four categories are considered, labeled with the first letters of the alphabet since the 5th edition23 (figure 1-5): BI-RADS “a” corresponds to a breast that is nearly entirely adipose tissue (fatty breast); BI-RADS “b” indicates presence of some dispersed fibroglandular tissue; BI-RADS “c” designates a heterogeneously dense breast; BI-RADS “d” denotes a breast that is extremely dense (therefore, a lesion could be masked). The weight-adjusted odds ratios for breast cancer between increasing BI-RADS categories versus the “a” category in postmenopausal women have been found to be respectively 1.6, 2.3, and 4.524. The BI-RADS classification has allowed the analysis of large study populations with relative ease and swiftness24. The subjectivity of its categories gives the BI-RADS classification for density only a moderate level of interobserver agreement (weighted kappa=0.43-0.59)25,26. However, the degree of intraobserver reliability reported by Kelikowski et al. (1998)26 was good (weighted kappa=0.73).   1.4  Quantitative assessment of mammographic density 1.4.1 Planimetry Used by Wolfe and his team in their research10,27, this technique entails direct measurement of the area on the mammogram that is occupied by dense tissue. Perimeters are traced on an acetate overlay (laying over the mammogram) with a wax pencil around the sectors containing dense tissue; the areas isolated in this way are 6  combined with a computerized compensating planimeter to estimate the total area of dense tissue10,27,28. The same process is used to find the total breast area; subsequently the division of the first area by the second one determines the fraction of the breast that is dense10,27 (figure 1-6). The process to obtain the measurements is not difficult, but the time required to complete the task will escalate with increasing number and complexity of areas containing dense tissue present on the mammogram7. Saftlas et al. (1991)27 reported odds ratios for breast cancer compared to a category defined as less than 5% breast density of 1.7, 2.2, 2.8, and 3.5 for percent density categories of 5-24.9%, 25-44.9%, 45-64.9, and 65% or more, respectively, controlling for age; further adjustment by including weight and parity gave estimates of 1.7, 2.5, 3.8, and 4.3, correspondingly. The intraobserver agreement of this technique has been described as good (unweighted kappa=77% for the five mammographic density categories)27.   1.4.2 Thresholding This method, also known as interactive or computer-assisted thresholding, developed by a team from the University of Toronto and the Sunnybrook Health Science Centre29, is a more time-efficient alternative to planimetry. It can be used on any mammographic digital image; such picture can result from digitization of mammograms (scanning), or from digital mammography; however, the images to be analyzed should come from the same source. 7  A grey-level value denotes the brightness of each picture element (pixel) composing the digital image30. In interactive thresholding, the boundaries of specific areas of the breast can be matched by a reader using a computer application (Cumulus), to a particular threshold grey level. In an interactive manner, with the adjustment of the threshold-grey level, the pixels of the image with values of brightness beyond the designated threshold are highlighted graphically on a color overlay; in this way the reader can discern on the screen the best level that isolates the area of interest. The determination of two threshold-grey levels is required (figure 1-7). The first one is used to define the breast perimeter, extricating the total breast area from the background. A second threshold level is used subsequently to optimally select the mammographically dense region(s); all the pixels above this threshold grey level are considered dense area. Furthermore, if the pectoralis muscle is present in the mammogram, it can manually be excluded from the quantification with a built-in tool. The extensions of the total breast and dense areas are computed by quantification of the pixels contained within each grey-level threshold; the ratio of these areas is the percent density (also known as percent dense area)7,30, which has been the traditional way to quantitatively express mammographic density31. The difference between these measurements corresponds to the non-dense area32,33. According to the meta-analysis by Pettersson et al. (2014)34, the reported age-adjusted summary odds ratio was 1.45 in premenopausal women, and 1.37 among postmenopausal women, for a one standard deviation increment in percent density. Thresholding inter- and intraobserver agreements, measured with intraclass correlation coefficients have regularly been greater than 0.930. Typically a mammographic examination generates four images (two projections per breast); however, it has been 8  demonstrated that representative mammographic density values can be obtained from a single mammographic image35.   1.4.3 Automated methods In recent years, full-field digital mammography (FFDM) has been substituting film-based mammography given its higher image quality36,37 , as well as its benefits in terms of data storage, processing, and retrieval36. This method is the standard in countries like the United Kingdom38, and Canada39. Various software applications able to measure mammographic density in a fully automated manner on FFDM images, have become available. Two of such programs are Quantra (Hologic Inc, Bedford, MA, USA) and Volpara (Volpara Health Technologies, Wellington, New Zealand). By taking pixel depth into account, these applications estimate the breast tissue volume40. Proprietary algorithms are used in each case41,42 on the “raw” FFDM image (that is, before any processing) to measure x-ray attenuations together with breast thickness; dense and non-dense tissue volume estimates are then generated for each pixel, information that each software uses to determine volumetric mammographic density quantities. These can be obtained as absolute measures of dense and non-dense tissue, or as a percentage (volumetric percent density)38. In a case-control study by Eng et al (2014)43, logistic regression models with log-transformed density measures rendered breast cancer risk estimates of 1.4 for Quantra and 1.83 for Volpara per one standard deviation increase in volumetric percent density, adjusted for age, body mass index, menopausal status and parity. In reliability analyses by Alonzo-Proulx et al (2015)40, a random-9  effects model estimated the within-subject variability of two mammographic density measurements taken from the same breast; for each method, an intraclass correlation coefficient was generated based on the within-subject and between-subject variance components of the random-effects model. The resulting intraclass correlation coefficient was 0.98 for both automated software applications.  1.5 Breast cancer and mammographic density Even though the biological mechanism is not well-understood, elevated mammographic density, specified either qualitatively or quantitatively, either as percent density or absolute dense area, is widely recognized as important breast cancer risk factor44-54. On the other hand, a small number of studies have evaluated the role of absolute non-dense area in breast cancer risk, and their conclusions have not reached a consensus, having observed conflicting results. For instance, Torres-Mejía et al.55 concluded non-dense area and breast cancer risk were negatively associated; however they did not adjust for dense area. Stone et al.56 found a weak inverse association which did not persist after controlling for dense area. Lokate et al.32 found a positive association (OR=1.7), while Petterson et al.33 reported a strong negative association (OR= 0.46), comparing 5th versus 1st non-dense area quintile among postmenopausal women, with the results of both studies being independent of dense area. A 2014 meta-analysis34 reported that non-dense area was inversely associated with the risk of breast cancer; however, it remained unclear whether this relationship is independent of dense area. In studies with a large negative correlation between dense area and non-dense area, the 10  association between non-dense area and cancer risk was no longer apparent after dense area adjustment. Baglietto et al. (2014)52 concluded that non-dense area was either negatively related with breast cancer risk assuming that the breast adipose tissue and the body fat are independent causal factors, or there was no association with breast cancer risk under the premise that both types of measures represent adiposity; the evidence slightly favored the former conclusion. The 2015 meta-analysis by Bertrand et al.57 concluded that, even though in most studies the associations for non-dense area were weakened after controlling for dense area, both parameters are independent risk factors associated with breast cancer risk, with the non-dense area being significantly associated with lower risk.   1.6  Epidemiology of breast cancer 1.6.1 Descriptive epidemiology Breast cancer is the most common malignant neoplasia in Canadian women; it is estimated that 26,300 women will be diagnosed in 2017 with invasive breast cancer, representing one fourth of all new cancer female cases in that year (excluding non-melanoma skin cancers)58. It is projected that 5,000 Canadian women will die from breast cancer in 2017, representing 13% of all female cancer deaths in the country58. Based on its invasiveness, breast cancer is classified as either in situ or invasive. The number of in situ breast cancers cases each year in Canada are unknown, but data 11  from the US and UK have estimated that 20% of all breast cancers are in situ59. In situ breast cancer denotes a neoplasia that confined to the site from which it originated in mammary ducts or lobules. The latter, also known as infiltrating cancer, has the same origin in ducts or lobules of the breast, but when found, it has invaded the surrounding breast adipose tissue. Histopathologically speaking, both types of tumors are similar60; likewise their risk factors are homologous61,62. In the U.S. and Canada, the most common type of breast cancer, originating from mammary glandular cells, is ductal adenocarcinoma, known also as infiltrating ductal carcinoma, or invasive ductal carcinoma58,63,64, being responsible for nearly four fifths of all invasive breast cancers in Canada58. In situ tumors are successfully treated by surgical excision; most are detected via screening mammography65,66. Invasive breast cancers differ in terms of treatment and prognosis according to the stage of the disease (tumor size, lymph node status, presence of metastasis), as well as immunohistochemistry (IHC) markers (estrogen (ER), progesterone (PR), and human epidermal growth factor 2 (HER2) receptors), histologic type, nuclear grade, and proliferation rate67. IHC expression of ER, PR and HER2 correlates with treatment response and subsequently with survival, with the triple negative (no expression of any of the three proteins) subtype having the worst overall survival, and positive subtypes having better outcomes68.   1.6.2 Reproductive factors Ultimately, the precise causes of a breast cancer in a woman are uncertain; however, multiple risk factors are known to be involved in its genesis, varying according to 12  molecular and clinical subtypes69,70. The risk of breast cancer is altered by several reproductive factors such as nulliparity71 (it is positively associated with breast cancer risk, with the reported relative risk being 1.7 to 1.9 compared with women who had their first live birth before age 20)72, menarche (there is a positive relationship between breast cancer risk and early menarche; the risk increases three times with menarche onset before age 11)73, menopause (late age at menopause is related with greater risk of breast cancer, increasing two-fold in women experiencing menopause after the age of 54 versus women who have their menopause before age 45)73, age at first full term pregnancy (having the first living birth after age 30 increases risk nearly two times compared to women having it before age 20)73, and duration of breastfeeding (longer duration is related with diminished risk; it has been reported that for every 12 months of lactation, the relative risk of breast cancer decreased by 4.3%74)75-77.   1.6.3 Socioeconomic and lifestyle factors Other associated risk factors include postmenopausal obesity78 (the risk doubles in postmenopausal women with BMI greater than 35)73, dietary fat79 (the reported relative risk in women with high intake of saturated fat is 1.5)73, sedentarism (the estimated significant pooled odds ratio for sedentary behavior and breast cancer is 1.08)80, and alcohol consumption81 (excessive intake increases the risk 1.3 times73). Tobacco use is not a well-established risk factor but its importance has sometimes been noted in epidemiological studies82,83. Among women belonging to the highest levels of income84 and education85, the occurrence of breast cancer has been observed to be consistently 13  higher. The association between high socioeconomic status and breast cancer risk is considered to be mediated by differential exposure to reproductive (such as older age at first childbirth, and greater use of hormone replacement therapy) and lifestyle factors (like alcohol consumption)86.  1.6.4 Environmental factors Several toxicants have been identified as human breast tissue carcinogens; for instance, the International Agency for Research on Cancer (IARC) considers polychlorinated biphenyls (PCB), widely used in the past as dielectric material in electric equipment87, as a breast carcinogenic agent, based on limited evidence on humans88. The same type of evidence was indicated by IARC to declare the organochloride dieldrin (used as insecticide)89 as a cause of breast cancer88. Polycyclic aromatic hydrocarbons (PAH) could potentially favor the development of breast malignant neoplasias90,91, but the epidemiological evidence is inconclusive92. Long-term night shift work for more than 30 years has been associated to greater breast cancer risk across diverse occupations (for women with more than 30 years of long-term night shifts the estimated odds were about two times the odds for women with no exposure to shift work)93; the meta-analysis of night-shift work has also been found to have a positive dose-effect association with breast cancer94. Cadmium, a common pollutant that can be found in contaminated food and tobacco smoke95, that has shown to have estrogenic activity96, has also been suggested to be associated with increased risk of breast cancer97. It was initially linked to increased mammographic density98; however, 14  subsequent evidence by the same researcher found no association99. Recent observations have suggested a link between high levels of air pollution (ozone, and particulate matter < 2.5 μm in diameter) and elevated mammographic density100.   1.6.5 Hormonal factors and their relationship with mammographic density Oral contraceptives (OC), specifically recent or current use of estrogen-progestogen combinations, are considered by IARC as a cause of breast cancer88. Current OC users, compared to never-users, are at 24% greater risk of breast cancer; younger age of oral contraceptives initiation is associated to greater risk101. Longer exposure does not seem to increase the risk; it has been determined that the excess risk is eliminated after interruption of OC for 10 years101,102. Ethnicity, family history, and BRCA carrier status does not seem to modify breast cancer risk associated to OC exposure101-103. Similarly, IARC classifies current use of estrogen-progestogen hormone replacement therapy (HRT) as breast cancer cause, and estrogen-only HRT as possible cause of breast cancer88. Compared to never-users of estrogen-progestogen HRT, the risk of breast cancer is estimated to be 55 to 100% higher in current HRT users104,105; in estrogen-only HRT the risk is higher but at lower extent than in estrogen-progestogen users104,106,107. The risk is increased in current HRT users with duration and lower body mass index (BMI). The associated risk disappears after cessation of HRT for 5 years or longer104,108.  15  Both, endogenous (menarche-menopause) and exogenous (such as HRT) sources of estrogen can make more likely the risk of breast carcinogenesis109; these exposures affect breast cellular proliferation by increasing it, a process that escalates the probability of mutations during cellular division110. Longer lifetime exposure is present when there is early menarche, late menopause, OC, and HRT60; in contrast, parity generates changes in levels of circulating estrogen, and breastfeeding can reduce the number of ovulatory cycles, decreasing the cumulative exposure of breast epithelium to estrogen110. In the presence of post-menopausal obesity, estrogen production is increased, resulting from androstenedione to estrone conversion, increasing breast cancer risk; obesity is also related to lower production of sex hormone-binding globulin, increasing the amounts of albumin-bound and free estrogen110. The association between nulliparity and elevated mammographic density (BI-RADS scale) has been described, as well as the association between current HRT use and density, with stronger association in women aged 65 or older111. Discrepancies in percent density in association with differences in exogenous hormone exposure (both OC and HRT) have been reported, with percent density being 6% greater in HRT users than in never users, and the same parameter being on average 5.3% greater in past OC users than in never users112.   1.6.6 Genetic and familial susceptibility 16  Breast cancer risk is greater in women with one first-degree relative with the disease, as compared with women with no first-degree relatives affected; the family history evidence suggests a genetic predisposition113,114 (relative risk is one and a half to three times higher in women with a first-degree relative affected by breast cancer)115; however since the evidence comes from familial aggregation, shared environment as well as genes could be responsible. Breast cancer risk has not been associated with breast cancer in an adoptive parent, likewise, the time since diagnosis of an affected family member does not modify breast cancer risk; this evidence suggests that genetic factors rather than shared environment are responsible for familial clustering of breast cancer cases116,117. Mutations in the tumor suppressor genes BRCA1 and BRCA2 (BReast CAncer genes 1 and 2) are the best-known high penetrance genetic determinants of breast cancer118,119; however the population prevalence of these mutation is relatively low (about 2 to 5% of incident breast cancers are estimated to be the result of these mutations)120. It has been recognized that from 5 to 10 % of breast cancer cases can be accounted by genetic factors121.  Dense area, non-dense area, and percent density are all mammographic characteristics that are highly heritable (60-70%)122,123; nonetheless, few genetic loci in relation to mammographic density have been identified124-127.   1.6.7 Specific genetic variation associated with breast cancer 17  The overall impact of genetic mutations in the population over the risk of breast cancer is characterized by how frequent the mutation is, and to what degree can the mutation increase breast cancer risk (penetrance); for instance, an uncommon mutation with high penetrance will be responsible for a low proportion of breast cancer cases in the population (like in Cowden, and Li-Fraumeni syndromes) 128.  Polymorphisms in estrogen-related genes such as CYP17, CYP19, CYP1A1, CYP1B1, COMT, and GSTP1, have been reported to influence the activity, regulation, and transcription of estrogen enzymes, as well as to impact endogenous estrogen exposure and the development of breast cancer129. A single nucleotide polymorphism (SNP) is the simplest and most frequent form of mutation, consisting in the variation of a single nucleotide occurring at a specific position in the genome130; even though many polymorphisms are low penetrance (therefore, seldom generate disease traits), considerable influence can be held by SNPs that affect the protein sequence (nonsynonymous SNPs). Although, as stated before, not many loci for mammographic density characteristics have been recognized, some genetic loci have been found to be linked to both breast cancer and mammographic density. For instance, rs12665607 is a SNP present in ESR1, the gene in chromosome 6 that is translated into estrogen receptor alpha (ERα); it has been found to be associated with breast cancer, as well as with elevated dense area in a large meta-analysis127.   18  1.7 Objectives The objectives of the present work are to examine whether information from screening mammograms could be utilized in a more optimal way by better understanding the association between the different components of mammographic density (dense and non-dense areas, percent density), and relate these to breast cancer tumor characteristics, and novel breast cancer risk factors (such as polygenic risk scores), the results of this study could be used to improve breast cancer screening strategies, as well as to help guide clinicians in diagnosing the disease (for example, by facilitating the definition of risk groups requiring tailored screening programs). The specific objectives are as follows: 1. To determine the association between mammographic non-dense area and breast cancer, as well as confirming the more well-studied association between dense area and breast cancer. 2. To assess the discriminating power provided by the information captured by the non-dense area parameter on models forecasting breast cancer.  3. To estimate breast cancer risk for mammographic density parameters by breast cancer tumor characteristics. 4. To evaluate the relationship between polygenic risk scores developed to predict genetic risk of breast cancer with mammographic density parameters. 19   1.8 Figures  Figure 1-1: Routine screening mammogram views. 20   Figure 1-2: Mammograms showing the four Wolfe grades. From left to right: N1, P1, P2 and DY). Reproduced from Kostas et al.131 with permission from Springer Nature ©2003. 21   Figure 1-3: Six Category System (Boyd’s SSC) for assessment of mammographic density: (a) SCC 1 (0%), (b) SCC 2 (>0 − 10%), (c) SCC 3 (>10 − 25%), (d) SCC 4 (>25 − 50%), (e) SCC 5 (>50 − 75%), and (f) SCC 6 (> 75%). Reproduced from Boyd et al. 19986 with permission from American Association for Cancer Research. 22   Figure 1-4: Tabár classification of mammographic parenchymal patterns. Top row shows MLO views; lower row contains CC views. Columns with labels A through E illustrate patterns I through V, in that order. Reproduced from Winkel et al. 201518 with permission from BioMed Central Ltd.23   Figure 1-5: BI-RADS breast composition categories. a) Almost entirely fatty, b) Scattered areas of fibroglandular density, c) Heterogeneously dense, d) Extremely dense. Reproduced with permission from BC Cancer132.24  Figure 1-6: Figure 1-6 has been removed due to copyright restrictions. It was an image illustrating the procedure to measure mammographic density via planimetry, showing the breast perimeter and the outline of the area containing dense tissue, traced on an acetate overlay placed over a mammogram. Original source: Wolfe JN, Saftlas AF, Salane M. Mammographic parenchymal patterns and quantitative evaluation of mammographic densities: A case-control study. Am J Roentgenol. 1987;148(6):1087–92. 25   Figure 1-7: Computer-assisted thresholding method (Cumulus software) for the determination of mammographic density. (a) Digitized mammogram. (b) The first threshold level isolate the breast from background (red overlay); the second threshold level indicates the dense area (green overlay).  26  2 Estimating treatment (exposure) effect of non-dense mammographic density over breast cancer risk in postmenopausal women In observational studies, the estimation of causal effects is difficult due to potential bias generated by non-randomly selected samples (such as in case-control studies)133 and potential residual confounding134. It is possible that, even after the application of standard design and adjustment techniques, the aforementioned problems could be responsible for the discrepancies seen in the results of studies estimating the effect of non-dense area on breast cancer risk. An analytic alternative is the application of the potential outcomes (counterfactual/Neyman-Rubin) framework of causal inference135, which can be used to address these issues136. The main purpose of these analyses is to estimate the marginal effect of non-dense area as well as the other mammographic density parameters on breast cancer risk using the potential outcomes framework, with data from the British Columbia (BC) component of the Canadian Breast Cancer Study (CBCS)93.  2.1  Methods 2.1.1  Study population The participants for this analysis came from the Canadian Breast Cancer Study, a population-based case-control study with 1,142 breast cancer cases and 1,178 controls, 27  from Greater Vancouver, British Columbia (BC), and Kingston, Ontario93. Women aged 40-80, diagnosed with incident in situ or invasive breast cancer between 2005 and 2009 were recruited from the BC Cancer Registry, and controls were cancer-free women from the Screening Mammography Program of BC, from the same geographic area, frequency-matched to cases in five-year age groups. A study package, including a questionnaire and a consent form, was delivered to potential participants. Questionnaire information included personal, medical and reproductive history, and lifestyle characteristics. In addition, participants consented to access to their medical records for information related to breast health and breast tissue blocks that were collected as part of regular care.  Since the etiology of premenopausal breast cancer is different than that of postmenopausal breast cancer, especially in regard to adiposity137,138, and given that the disease is more common after menopause, this study was restricted to postmenopausal participants from BC (606 cases and 595 controls); participation rates were 54% among cases and 57% among controls. Furthermore, the study was restricted to cases who participated in the Screening Mammography Program (n=524), and only subjects for whom screening film mammograms were identified; 477 cases (91%) and 588 controls (99%) were utilized in this analysis.  2.1.2  Questionnaire 28  Participants completed a questionnaire that was either answered by telephone interviews, or written by the participant and mailed. Questions covering education, ethnicity, reproductive and contraceptive history, family history of cancer, lifestyle characteristics (e.g. alcohol consumption and smoking), and lifetime work history were included in the questionnaire.   2.1.3   Menopausal status assessment A similar approach to Friedenreich et al.139 was followed; women were considered postmenopausal if either: (1) more than 1 year had passed with no menstruation, (2) they had underwent bilateral oophorectomy, (3) they were older than 50 years with unknown date of last menstruation and their menstruation had stopped spontaneously, or (4) menstruation had stopped following different factors (such as endometriosis treatment with leuprolide acetate) and they were over the age of 55.  2.1.4 Mammographic density measurement  The most recent normal mammogram prior to study enrollment was selected for each participant. For 92 controls, it was not possible to locate mammograms from before enrollment, so the mammogram closest to that date was chosen (mean time=2.3 years after enrollment, SD=0.7). For cases, the contralateral breast was chosen, and for 29  controls, the side was randomly determined. All views were craniocaudal, and mammograms were digitized using the same device (iCAD TotalLook Mammo Advantage). Total breast area and dense area were determined with the Cumulus software140, using the well-defined interactive thresholding technique29 by a single reader (HAVG), trained at the Sunnybrook Health Sciences Centre, blinded to the participants' information. Non-dense area was calculated subtracting dense area from breast area, and percent mammographic density was obtained as the ratio between dense area and breast area. The mammograms were randomly allotted to reading batches of 50, with 10% of the images repeated to assess reliability within the set, plus 20% of the images of the first set repeated every five reads to estimate consistency between the sets. The within-set intraclass correlation coefficients (ICC) were above 0.99 for breast area and 0.98 for dense area, and the between-sets ICC were 0.98 and 0.97. As a validation exercise, a random sample of images was sent to three experts in Cumulus assessment to evaluate the reliability of the reader; the resulting between-reader ICC was between 0.98 and 0.99 for breast area, and 0.90 to 0.98 for dense area (table 2-1).  2.1.5 Statistical analysis A causal diagram141,142 (directed acyclic graph, or DAG; figure 2-1) was used to identify minimally sufficient adjustment sets of variables, using DAGgity143 v.2.2, under the assumption of independent effects of non-dense area and dense area, which combined define percent mammographic density (details about the DAG’s nodes and 30  relationships can be seen in Appendix A). This assumption was tested by examining a model containing the adjustment set plus dense area (model BASE+DA+NDA) evaluated against a reduced model excluding non-dense area (model BASE+DA), using likelihood ratio tests (LRT). The same methodology was used to assess the independent effects of non-dense area and BMI on the risk of breast cancer.  Following the methodology described by Didelez et al144, a sampling indicator node was included in the causal diagram to illustrate the sampling used in a case-control study: that is, observations are conditional on being sampled given age and the presence or not of breast cancer. The DAG was also used to assess the presence of the collapsibility of the conditional odds ratio for the exposures of interest and the outcome over the sampling node, given the identified adjustment set. This method is employed to confirm that the estimated odds ratios for mammographic density parameters conditional on the adjustment set in the sampled population are consistent for the entire target population, and this situation is known as collapsibility under outcome-dependent sampling144. The minimally sufficient adjustment set to identify the effect of mammographic density parameters over breast cancer risk included: body mass index (BMI) two years before study enrollment, age, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, use of oral contraceptives, family history of breast cancer, use of hormone replacement therapy (HRT), lifetime smoking, and average alcohol consumption over the 10-year time period when the mammogram was 31  generated. The adjustment set predicting non-dense area was comprised of age, family history of breast cancer, parity, age at first full term pregnancy, education, ethnicity, HRT, lifetime smoking, and alcohol consumption. The set of variables predicting dense area and percent mammographic density included age, education, ethnicity, parity, age at first full term pregnancy, lifetime breastfeeding, family history of breast cancer, HRT, lifetime smoking, and alcohol consumption. Values were missing for some variables in 0.5-5.6% of the cases, and in 0.1 to 3.3% of controls (for details, see table 2-2), but since no obvious pattern was seen to suggest that missing information depended on unobserved values, missing values were assigned using predictive mean matching, a semi-parametric imputation approach implemented in the R package mice (Multivariate Imputation by Chained Equations v.2.25)145. Analyses were also conducted for the complete dataset after eliminating participants with missing values. Logistic regression models including variables from the minimally sufficient adjustment set were used to estimate the association between mammographic measures (dense area, non-dense area, percent mammographic density) and the risk of breast cancer. Trend tests were conducted by entering the relevant ordinal variable as a continuous variable into the model. Based on the literature, family history of breast cancer146, HRT147 and parity148 were considered as potential effect modifiers; tests for effect modification were explored by inclusion of interaction terms in the models. 32  Based on the Akaike information criterion (AIC), the best model was achieved with the adjustment set represented as BMI (continuous), age (continuous), education (high school or less / college or trade certificate / undergraduate degree/ graduate or professional degree), ethnicity (European / East Asian / Filipino / South Asian / Mixed or Other), age at menarche (continuous), age at first full-term pregnancy (never / younger than 20 years/ 20 – 29 years / 30 – 39 years / older than 40 years), parity (yes / no),  lifetime breastfeeding (continuous), use of oral contraceptives (never / <4.5 years / 4.5 – 10 years / >10 years), family history of breast cancer (positive / negative), HRT (never / <5 years / 5 – 12 years / >12 years), lifetime smoking (continuous), alcohol consumption (continuous). Also, as proposed by Baglietto et al.52, an age by BMI interaction (continuous) was included in all models, allowing the associations of breast cancer risk and BMI to be subject to age. The presence of collapsibility under outcome-dependent sampling was determined graphically, ascertaining that consistent estimation of conditional odds ratio was possible144. Since marginal odds ratios can be directly estimated with observations weighted by inverse probability of treatment (exposure) allocation149, marginal odds ratios between mammographic parameters and breast cancer risk were computed using the augmented inverse-probability weighted (IPW) treatment-effect estimator, implemented in Stata150. This estimator makes use of weighted regression coefficients to calculate averages of exposure-specific predicted outcomes where the estimated inverse probabilities of exposure are the weights. It also introduces an augmentation term in the outcome model to amend the estimator in case of misspecification of the 33  treatment model (double-robust property). The positivity assumption151 was assessed using plots of the calculated probability densities of receiving each exposure level for each subject152. Observations with propensity scores <0.05 were removed prior to the computation of marginal effects153. Treatment-effect point estimates, standard errors, significance levels, and confidence intervals were computed using the delta method154.  AIC and receiver operating characteristic (ROC) curves were used to assess the goodness-of-fit and discriminating power of unconditional logistic models containing the adjustment set and different combinations of the parameters: non-dense area alone, dense area alone, non-dense area and dense area, or percent mammographic density alone. The ROC curves were compared using tests of equality of areas under the ROC curve (adjusted for multiple comparisons via Šidák correction). All statistical tests were two-sided. Analyses were performed using R v.3.0.3155, and Stata v.14.0156.  2.2 Results As seen in table 2-2, compared to controls, cases had a greater proportion of women who had family history of breast cancer among 1st-degree relatives, were East Asians, and were less educated; on average they also were older, had higher BMI, breastfed for a shorter time, had lower exposure to oral contraceptives, and higher alcohol consumption. 34  The assessment of the predictive effect of mammographic density parameters estimating the risk of breast cancer produced the following results: when the model containing non-dense area plus the adjustment set was compared with the model comprised of the adjustment set only (base model), non-dense area was significant (p-LRT=0.0053). Similarly, the comparison between the base model and a model including dense area and the adjustment set, showed dense area was a better fit (p-LRT<0.0001). The evaluation of the model including dense area plus the adjustment set competing with the model including both dense area and non-dense area (plus the base model) found the latter fit better (p-LRT=0.0248). No interaction between the mammographic density parameters and the proposed effect modifiers was observed (table 2-5) After assessment of the positivity assumption, the following sets of observations with propensity scores lesser than 0.05 were removed, previous to the calculation of marginal effects: for non-dense area, 11 observations (4 cases and 7 controls); for dense area, 1 observation (1 control); lastly, for percent mammographic density, 8 observations (5 cases and 3 controls). Table 2-3 presents marginal odds ratios for the three mammographic density parameters. The estimated marginal odds of breast cancer that would be expected if all the participants in the study were in the top quartile of non-dense area are 60% lower than the odds that would be expected if all women were in the bottom quartile (OR=0.40, 95%CI=0.26–0.61, p-trend<0.001). Similarly, the computed marginal odds of breast cancer that would be present if all the study participants were in the top quartile 35  of dense area are nearly twice as high as the odds that would be expected if all the population were in the bottom quartile (OR=1.81, 95%CI=1.19–2.43, p-trend<0.001). Finally, if the entire study population were exposed to the fourth percent mammographic density quartile, the expected marginal odds of breast cancer would be over three times as high as the odds that would be present if all women were in the first quartile (OR=3.15, 95%CI=1.90–4.40, p-trend<0.001).  Table 2-4 indicates the goodness of fit and predictive accuracy for the logistic models using the adjustment set as assessed by AIC. All models with a mammographic density parameter performed better than the model with no parameter (AICBASE=1,379.4). The model with the best fit was percent mammographic density alone (AICBASE+PD=1,331.2), followed by non-dense area plus dense area (AICBASE+NDA+DA=1,350.8), dense area alone (AICBASE+DA=1,351.9), and non-dense area alone (AICBASE+NDA=1,376.4). The model with the best predictive accuracy was again the one with percent mammographic density alone (ROCBASE+PD=0.74), followed by the model using non-dense area and dense area (ROCBASE+NDA+DA=0.73), dense area alone (ROCBASE+DA=0.73), and non-dense area alone (ROCBASE+DA=0.71). However, the test of equality of the AUC did not provide evidence to conclude that the predictive accuracy of the first three models differed. The marginal estimates, as well as the evaluation of the accuracy of the models conducted with the dataset without missing values were very similar (tables 2-6 and 2-7) to those from the main analyses with imputed data. Similarly, a sensitivity analysis 36  eliminating controls with mammograms generated after enrollment did not significantly change the results of the study (tables 2-8 and 2-9).  2.3 Discussion This study shows that non-dense area is an independent factor inversely related with risk of breast cancer, after adjustment for dense area and other covariates. This evidence is concordant with the findings of Baglietto et al.52 showing an inverse association of non-dense area with breast cancer risk, under the assumption of an independent effect of non-dense area on breast cancer risk. In addition, the results showing that both dense area and percent mammographic density are directly related risk factors for breast cancer are consistent with the literature45,47-53. The plausible negative association between non-dense area and breast cancer risk can be explained by various biological mechanisms. Breast adipose tissue may participate in the storage and bioactivation of vitamin D, and its active form could regulate ductal epithelium through an inhibitory factor released by activation of the vitamin D receptor within the mammary adipocytes157. Since the physiological atrophy of the epithelium is directly related to non-dense area158, this mammographic density parameter could represent the level of lobular involution, which is a process inversely related to breast cancer risk. The amount of breast adipose tissue could also be the result of childhood adiposity159, a factor that has been inversely associated with breast cancer risk160,161; correspondingly, high mammographic density (percent mammographic density) has 37  been reported to be more prevalent in women who had low prepubertal weight162, probably through a mechanism mediated by insulin-like growth factor 1 (IGF-1)163. IGF-1 is a recognized breast cancer risk factor164; its levels have been found to be lower in adult women who were obese at menarche165. Similarly, breast aromatase activity takes place mostly in stromal preadipocytes, decreasing after these cells mature into adipocytes166,167, and tumor growth may be influenced by estrogen production taking place at this level. Since this estrogen source is diminished as result of the differentiation of breast adipocytes, the negative association between non-dense area and breast cancer risk could also be a consequence of this process. A peculiar finding of the initial assessment of the study population was the fact that, in contrast to what has been observed consistently in other studies85,86 , cases were better educated than controls. The negative association with education can be the result of selection bias, given the different participation rates present in cases and controls.  A strength of this study is that the effects of the mammographic density parameters were examined in a causal inference framework. IPW estimators generate “pseudo-populations” in which the exposure of interest is independent of measured confounders. Furthermore, with the double-robust nature of the augmented IPW estimators applied in our study, the consistency of the estimator is present as long as at least one of the models (either exposure or outcome) is correctly specified150. The computed effect of the exposure on the outcome in a sample weighted in this way can be used to estimate the causal (marginal) effect of the exposure of interest168-170.  38  Three assumptions described by Cole and Hernán168 are required to calculate causal effects from observational data using IPW estimators: exchangeability (the potential outcomes, given the adjusted confounders, are independent of the observed exposure; it implies also the assumption of absence of unmeasured confounding), positivity (at every category of the confounders, there must be subjects at every exposure level), and correct specification of the model. In practice, it is not possible to guarantee that all the required assumptions for inference of causality are being fulfilled; however, IPW estimators offer some advantages as alternative to regression adjustment methods.  Traditional regression methods are dependent on linear functions between the outcome of interest and its predictors; the required linearity is not always tested, and when non-linear relationships are adjusted using misspecified linear models, the regression adjustment can be unsuccessful at eliminating bias, and it might even increase it171,172. IPW estimators, on the other hand, are not as dependent on distributional assumptions173. The results of our analyses show that (as per AIC and ROC curve point estimates) percent mammographic density provides the best prediction, followed by the combination of dense area plus non-dense area, and lastly dense area alone. There is no evidence to conclude that the information captured by the non-dense area parameter improves the model (non-statistically significant test of equality of ROC areas, ROC confidence intervals overlap).  39  Several other strengths and limitations are present in this research. In the original study, cases were recruited from the population BC Cancer Registry, while controls came from membership in the population Screening Mammography Program which potentially could introduce selection bias. However, participants for this analysis were restricted to cases (and controls) in the Screening Mammography Program. Although recall bias may apply to some questionnaire data, this information was not used to assess exposure. Objective measures of mammographic density measures were used. Also, it is unlikely that participation would be related to breast density since most participants’ mammographic density values are routinely not provided to screened women in BC. All digitized images were craniocaudal projections, thus avoiding the potential systematic inclusion of subcutaneous adipose tissue as part of the measured non-dense area174. Furthermore, the closeness of the estimated odds ratios in the CBCS for known risk factors to those commonly found in breast cancer case-control and cohort studies, as seen in or previous publication, indicates that no large selection or recall biases are likely present in this study93. Lastly, an additional strength is that the extensive information obtained through questionnaires about a wide variety of factors made possible extensive adjustment of the identified minimal sufficient adjustment set.   2.4 Conclusions  The results of this study provide evidence supporting non-dense area as a factor inversely related with breast cancer risk, independent of dense area and BMI. However, the information captured by the non-dense area did not improve the predictive ability of 40  models of breast cancer risk containing either the dense area or the percent area parameters. 41  2.5 Tables Table 2-1: Inter-rater agreement (ICC) for mammographic density measurements Comparison Mammographic measurement ICC 95% Confidence intervalUniversity of Western Australia Percent mammographic density 91.60% 84.68 – 95.47%Breast area 98.69% 97.53 – 99.30%Dense area 91.21% 84.00 – 95.25%University of Hawaii Percent mammographic density 94.59% 86.87 – 97.82%Breast area 99.61% 99.01 – 99.84%Dense area 89.61% 75.69 – 95.76%University of Toronto Percent mammographic density 98.82% 97.10 – 99.53%Breast area 98.20% 95.52 – 99.28%Dense area 98.25% 95.64 – 99.30% 42  Table 2-2: Characteristics of study population Variables* Cases (N=477)  Mean (SD) / N (%) Controls (N=588) Mean (SD) / N (%) p-value‡Age at study entry (years) 64.0 (7.7) 62.9 (7.9) 0.03Age at mammogram (years) 60.9 (7.7) 63.0 (8.0) <0.001Age at first mammogram (years) 47.7 (7.6) 47.0 (6.8) 0.14Ethnicity European 305 (63.9%) 465 (79.1%) 0.006East Asian 113 (23.7%) 61 (10.3%) Filipino 24 (5.1%) 20 (3.4%) South Asian 22 (4.6%) 23 (3.9%) Mixed/Other 13 (2.7%) 19 (3.3%) Education High school or less 197 (41.3%) 180 (30.6%) <0.001College/trade certificate 132 (27.7%) 169 (28.7%) Undergraduate degree 97 (20.3%) 121 (20.6%) Graduate/professional degree 51 (10.7%) 118 (20.1%) BMI (kg/m2) two years before study entry 26.3 (5.1) 25.1 (4.7) <0.001Family history of breast cancer (%) 117 (24.5%) 90 (15.3%) <0.001Age at menarche (years) 13.0 (1.6) 12.9 (1.5) 0.23Ever been pregnant (yes) 370 (77.6%) 443 (75.3%) 0.40Age at first pregnancy (years)° 26.2 (5.5) 25.8 (4.9) 0.27Parity° 2.3 (1.1) 2.4 (1.0) 0.37Ever breastfed° (%) 367 (99.2%) 439 (99.1%) 0.89Lifetime breastfeeding° (months) 6.3 (5.1) 7.1 (5.0) 0.035Oral contraceptive use (years) Never 239 (50.1%) 249 (42.4%) 0.018<4.5 years 98 (20.5%) 133 (22.6%) 4.5 – 10 years 90 (18.9%) 132 (22.4%) >10 years 50 (10.5%) 74 (12.6%) HRT use (years) Never 286 (60.0%) 343 (58.3%) 0.71<5 years 62 (13.0%) 85 (14.5%) 5 – 12 years 84 (17.6%) 101 (17.2%) >12 years 45 (9.4%) 59 (10.0%) NSAIDs use (years) Never 349 (73.2%) 399 (67.9%) 0.10<2.34 years 43 (9.0%) 70 (11.9%) 2.34 – 8.5 years 46 (9.6%) 56 (9.5%) >8.5 years 39 (8.2%) 63 (10.7%) Smoking (pack/years) 6.7 (13.7) 6.4 (12.4) 0.72Alcohol consumption (drinks/week) 2.8 (5.1) 2.0 (5.0) 0.02‡Unadjusted logistic regression, p-trend. °Among parous women. *The following variables presented missing values: body mass index (0.5% of cases and 0.1% of controls), age at first full 43  term pregnancy (0.8% of cases and 3.3% of controls), lifetime breastfeeding (1.4% of cases and 1.1% of controls), use of oral contraceptives (2.1% of cases and 1.9% of controls), family history of breast cancer (5.6% of cases and 3.1% of controls), HRT (2.3% of cases and 2.5% of controls), lifetime smoking (0.7% of cases and controls), and alcohol consumption (0.7% of cases and 3.3% of controls). Reprinted by permission from Springer Nature: Breast Cancer Research and Treatment, Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study, Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, et al, © 2018, (https://doi.org/10.1007/s10549-018-4737-7. Breast Cancer Res Treat).   44  Table 2-3: Marginal odds ratios (augmented inverse-probability weighting estimators) for mammographic density measures Mammographic density parameter‡ Quartile Cases (n) Controls(n) Odds ratio 95% Confidence interval p-trend Non-dense areaº 1 144 146 1.00 Reference <0.001 2 105 145 0.52 0.35 – 0.76 3 110 144 0.43 0.29 – 0.62 4 114 146 0.39 0.26 – 0.61 Dense area1 1 86 147 1.00 Reference <0.001 2 92 146 0.89 0.56 – 1.22 3 107 147 1.07 0.69 – 1.45 4 192 147 1.81 1.19 – 2.43 Percent mammographic density2 1 75 146 1.00 Reference <0.001 2 78 147 0.94 0.55 – 1.32 3 152 145 2.11 1.30 – 2.92 4 167 147 3.15 1.90 – 4.40 ‡Adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime HRT, lifetime smoking, and alcohol consumption. ºAdjusted for ‡ + dense area. 1Adjusted for ‡ + non-dense area. 2Adjusted for ‡ . Reprinted by permission from Springer Nature: Breast Cancer Research and Treatment, Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study, Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, et al, © 2018, (https://doi.org/10.1007/s10549-018-4737-7. Breast Cancer Res Treat).  45  Table 2-4: Comparison of model accuracy for breast cancer Logistic model‡ AIC AUC (95% Conf. Interval) p-value* Adjusted p-value**PDº 1,331.2 0.74 (0.71 – 0.77) Reference ReferenceNDA1 + DA2 1,350.8 0.73 (0.70 – 0.76) 0.30 0.75DA2 1,351.9 0.73 (0.70 – 0.76) 0.10 0.33NDA1 1,376.4 0.71 (0.68 – 0.74) 0.004 0.015Base (no MD parameters) 1,379.4 0.70 (0.67 – 0.74) <0.001 <0.001ºPercent mammographic density 1Non-dense area 2Dense area *Test of equality of ROC areas vs. model containing percent mammographic density + ‡ . **Šidák correction. ‡All models adjusting for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption, plus the indicated mammographic parameter. Reprinted by permission from Springer Nature: Breast Cancer Research and Treatment, Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study, Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, et al, © 2018, (https://doi.org/10.1007/s10549-018-4737-7. Breast Cancer Res Treat).  46  Table 2-5: Interactions between mammographic density parameters and selected breast cancer risk factors  VARIABLE Non-dense area*º Dense area*1 Percent dense area* Family history  0.75 0.23 0.39 Parity 0.27 0.18 0.45 Hormone replacement therapy 0.23 0.26 0.14 *P-value of interaction term All models adjusting for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption, plus the indicated mammographic parameter. ºModel includes also dense area. 1Model includes also non-dense area. Reprinted by permission from Springer Nature: Breast Cancer Research and Treatment, Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study, Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, et al, © 2018, (https://doi.org/10.1007/s10549-018-4737-7. Breast Cancer Res Treat).  47  Table-2-6: Marginal odds ratios (inverse-probability weighting estimators) for mammographic density measures; no imputation Mammographic density parameter‡ Quartile Cases (n) Controls(n) Odds ratio 95% Confidence interval p-trend Non-dense areaº 1 146 134 1.00 Reference <0.001 2 93 132 0.44 0.30 – 0.65 3 96 131 0.38 0.26 – 0.55 4 96 133 0.32 0.20 – 0.50 Dense area1 1 75 133 1.00 Reference <0.001 2 73 133 0.80 0.48 – 1.11 3 105 134 1.06 0.67 – 1.45 4 181 134 1.81 1.15 – 2.47 Percent mammographic density2 1 62 132 1.00 Reference <0.001 2 63 134 0.92 0.50 – 1.35 3 143 133 2.39 1.39 – 3.39 4 161 134 3.63 2.06 – 5.20 ‡Adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime HRT, lifetime smoking, and alcohol consumption. When calculating marginal estimates, due to propensity scores <0.05, 9 observations (4 cases and 5 controls) were removed when calculating marginal estimates for non-dense area, 2 observations (1 case and 1 control) for dense area, and 8 observations (6 cases and 2 controls) for percent mammographic density. ºAdjusted for ‡ + dense area. 1Adjusted for ‡ + non-dense area. 2Adjusted for ‡ . 48  Table 2-7: Comparison of model accuracy for breast cancer; no imputation Logistic model‡ AIC AUC (95% Conf. Interval) p-value* Adjusted p-value**PDº 1,203.8 0.75 (0.72 – 0.78) Reference ReferenceNDA1 + DA2 1,222.4 0.74 (0.71 – 0.77) 0.37 0.84DA2 1,227.4 0.73 (0.70 – 0.77) 0.10 0.35NDA1 1,251.8 0.72 (0.69 – 0.75) 0.004 0.016Base (no MD parameters) 1,259.5 0.71 (0.67 – 0.74) <0.001 0.002ºPercent mammographic density 1Non-dense area 2Dense area *Test of equality of ROC areas vs. model containing percent mammographic density + ‡ . **Šidák correction. ‡All models adjusting for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption, plus the indicated mammographic parameter. 49  Table 2-8: Marginal odds ratios (inverse-probability weighting estimators) for mammographic density measures, eliminating 92 controls with mammograms generated after study enrollment Mammographic density parameter‡ Quartile Cases (n) Controls(n) Odds ratio 95% Confidence interval p-trend Non-dense areaº 1 148 123 1.00 Reference <0.001 2 101 122 0.50 0.33 – 0.75 3 110 121 0.41 0.28 – 0.62 4 114 124 0.36 0.23 – 0.57 Dense area1 1 86 124 1.00 Reference <0.001 2 101 123 0.97 0.61 – 1.33 3 102 124 1.03 0.65 – 1.42 4 188 124 1.78 1.15 – 2.42 Percent mammographic density2 1 77 123 1.00 Reference <0.001 2 78 124 0.91 0.53 – 1.29 3 151 122 2.00 1.22 – 2.78 4 164 124 3.00 1.77 – 4.22 ‡Adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime HRT, lifetime smoking, and alcohol consumption. When calculating marginal estimates, due to propensity scores <0.05, 10 observations (4 cases and 6 controls) were removed for non-dense area, 1 observation (1 control) for dense area, and 10 observations (7 case and 3 controls) for percent mammographic density. ºAdjusted for ‡ + dense area. 1Adjusted for ‡ + non-dense area. 2Adjusted for ‡ . 50  Table 2-9: Comparison of model accuracy for breast cancer, eliminating 99 controls with mammograms generated after study enrollment Logistic model‡ AIC AUC (95% Conf. Interval) p-value* Adjusted p-value**PDº 1,239.6 0.73 (0.70 – 0.76) Reference ReferenceNDA1 + DA2 1,255.0 0.73 (0.69 – 0.76) 0.30 0.77DA2 1,257.9 0.72 (0.69 – 0.75) 0.07 0.26NDA1 1,275.5 0.71 (0.68 – 0.74) 0.009 0.036Base (no MD parameters) 1,353.7 0.70 (0.67 – 0.73) 0.002 0.006ºPercent mammographic density 1Non-dense area 2Dense area *Test of equality of ROC areas vs. model containing percent mammographic density + ‡ . **Šidák correction. ‡All models adjusting for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption, plus the indicated mammographic parameter.51  2.6 Figures Figure 2-1: Minimally sufficient adjustment set between mammographic density parameters and breast cancer risk in postmenopausal women (causal diagram).  52  Reprinted by permission from Springer Nature: Breast Cancer Research and Treatment, Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study, Velásquez García HA, Sobolev BG, Gotay CC, Wilson CM, et al, © 2018, (https://doi.org/10.1007/s10549-018-4737-7. Breast Cancer Res Treat). 53    3 Associations between mammographic density parameters and breast cancer tumor characteristics  The relative quantities of adipose (radiolucent or non-dense), connective and epithelial tissues (radiopaque or dense) determine mammographic density, and is measured by the heterogeneous levels of x-ray attenuation of these structures visualized on a mammogram; in this way, the presence of larger amounts of dense tissue provide higher degree of mammographic density7. High mammographic density, specified qualitatively or quantitatively both as percent mammographic density as well as absolute dense area, is a very important breast cancer risk factor44-53.  The association between breast cancer and well-known risk factors has been shown to differ according to the characteristics of the tumor175-182. However, for mammographic density, that has not been clearly demonstrated. Some studies have reported no disparity in the magnitude of the association between mammographic density and breast cancer tumor characteristics146,183-192; while other have indicated differences by hormone receptor status51,53,57,193-196, invasiveness51,197, phenotype198,199, tumor size51,195,196,200,201, and stage202. Most studies have limited the assessment of mammographic density qualitatively as defined by the BI-RADS classification, or quantitatively as percent dense area; the potentially different associations between breast tumor characteristics and the other mammographic density parameters, absolute dense area and absolute non-dense area, have seldom been taken into account. 54   It is important to elucidate if mammographic density parameters indicate a predisposition to specific breast cancer subtypes. Such knowledge could help to understand clinical pathways, as well as to identify susceptible groups of women in the general population, providing evidence to improve the formulation of screening protocols and risk-reducing interventions203.  3.1 Methods 3.1.1 Study population As described in chapter 2, the examined data originates from the British Columbia (BC) study subpopulation belonging to the Canadian Breast Cancer Study (CBCS)93, a population-based case-control study from Greater Vancouver, BC, and Kingston, Ontario, finishing gathering data in 2010; the subsequent participation rates were 54% for cases and 57% amid controls. Incident cases ages 40 to 80 that were diagnosed between 2005 and 2009, were recruited from the BC Cancer Registry; controls were cancer-free individuals enrolled from the Screening Mammography Program (SMP), from the same geographic area, and frequency-matched to cases in five-year age groups. The initial sample was composed by 1,003 cases and 1,014 controls. Given that breast cancer is more frequent in postmenopausal women, and etiologically speaking the set of causes of the disease most likely differ in premenopausal versus postmenopausal women, particularly concerning adiposity137,138, this study was 55   restricted to postmenopausal participants, leaving 559 cases and 619 controls suitable for this study. The final study population, determined by the availability of screening film mammograms, was comprised by 477 cases (92.13% of 559) and 588 controls (99.03% of 619). A questionnaire was used to collect information about education, ethnicity, medical history (including family history of cancer), and lifestyle characteristics. Furthermore, to access their medical records for data related to breast health and breast tissue blocks gathered as part of regular care, consent from participants was obtained.   3.1.2 Mammographic density measurement  Described in section 2.1.4; briefly, the most recent normal mammogram preceding recruitment into the study was selected for each subject. It was not possible to locate mammograms prior to study enrollment for 92 controls, so the mammogram closest to that date was chosen (mean time=2.3 years after enrollment, SD=0.7); a sensitivity analysis was conducted excluding these controls. The contralateral breast was selected for cases; for controls, the side was chosen at random. Mammograms were digitized using the same device (iCAD TotalLook Mammo Advantage); the craniocaudal view was used in all instances. Total breast area and dense area were determined with Cumulus software (Imaging Research Program, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Canada), by interactive thresholding method29 by a 56   single reader (HAVG), who was not aware of information related to the assessed mammograms’ participants. The within-set intraclass correlation coefficients (ICC) were above 0.99 for breast area and 0.98 for dense area, and the between-sets ICC were 0.98 and 0.97. To evaluate the reliability of the reader, a random sample of at least 20 images was sent to three experts in mammographic density evaluation via Cumulus; the ensuing between-reader ICC was between 0.98 and 0.99 for breast area, and 0.90 to 0.98 for dense area (table 2-1).  3.1.3 Breast tumor characteristics assessment As it has been previously described93, among cases, information about tumor characteristics such as invasiveness, histology, size, breast cancer stage, estrogen receptor (ER), progesterone receptor (PR), and human epidermal factor receptor 2 (HER2) status, was obtained from the BC Cancer Registry and BC Breast Cancer Outcomes Unit. ER status was defined from immunohistochemistry (IHC) results, classified into one of six categories: negative, weakly positive, moderately positive, strongly positive, receptors tested but not sufficient quantity for interpretation or borderline/equivocal, and not tested. Tumors classified as weakly, moderately or strongly positive, were identified as ER positive. PR status was determined through IHC testing using the same methodology as the ER analysis. HER2 status was evaluated with IHC; scores 0 to 1+ were interpreted as negative, 2+ as borderline, and 3+ as positive. HER2 IHC borderline results were further discriminated through fluorescence 57   in-situ hybridization (FISH); a FISH result of more than 6.0 HER2 gene copies per nucleus was considered positive.   3.1.4 Statistical analysis Exposures of interest (mammographic dense, non-dense, and percent dense areas) were categorized by quartiles according to the distribution in controls. The characteristics of cases and controls were assessed using univariable logistic regression.  Tests for heterogeneity between subtypes for each of the tumor characteristics were assessed by multinomial (polytomous) logistic regression utilizing breast cancer cases only204,205. To model the association between mammographic parameters (dense area, non-dense area, percent density) and the risk of breast cancer, odds ratios were computed for each subtype separately using unconditional logistic regression. Trend tests were conducted by entering the relevant ordinal variable as a continuous variable into the model. Tests for effect modification were explored by inclusion of interaction terms in the models. A causal diagram141,142 (directed acyclic graph, or DAG; figure 2-1) was used to identify minimally sufficient adjustment sets of variables in the path between mammographic density parameters and breast cancer, using DAGgity143. Based on the Akaike 58   information criterion (AIC), the best model was achieved with a minimally sufficient adjustment set represented as BMI (continuous; 2 years before study enrollment), age (continuous), education (high school or less / college or trade certificate / undergraduate degree/ graduate or professional degree), ethnicity (European / East Asian / Filipino / South Asian / Mixed or Other), age at menarche (continuous), age at first full-term pregnancy (never / younger than 20 years/ 20 – 29 years / 30 – 39 years / older than 40 years), parity (yes / no),  lifetime breastfeeding (continuous), use of oral contraceptives (never / <4.5 years / 4.5 – 10 years / >10 years), menopausal status (premenopausal / postmenopausal) , family history of breast cancer (positive / negative), use of hormone replacement therapy (never / <5 years / 5 – 12 years / >12 years), lifetime smoking (continuous), and average alcohol consumption when the mammogram was generated (continuous). Initial analysis also showed an age by BMI interaction (continuous), so this was included in all models.  Values were missing for some variables in 0.5-5.6% of the cases, and in 0.1 to 3.3% of controls (for details, see table 2-2). By using the tools included in the visualization and imputation of missing values (VIM)206 R package (version 4.5), preliminary exploration of the missing data did not find any pattern suggesting that missing information depended on unobserved values; missing values were assigned using predictive mean matching, using the R package mice145. Analyses were also conducted for the complete dataset after eliminating participants with missing values. 59   All statistical tests were two-sided. Analyses were performed using Stata v.14.0 (Stata Corporation, College Station, Texas, USA).  3.2 Results Compared to controls, cases had a greater proportion of women who had been pregnant, were East Asian, were postmenopausal, had a family history of breast cancer among 1st-degree relatives, and also had higher BMI, breastfed for a longer time, and had lower exposure to oral contraceptives (table 2-2). Table 3-1 indicates the distribution of tumor characteristics on cases. 553 (77.45%) of the cases were invasive; with stage 1 being the most common group (n=284, 39.78%). Restricting to the invasive breast cancer cases, most were in the 1.1-2.0 cm size category (n=234, 42.31%), histologically ductal (n=608, 85.15%), ER positive (434, 78.48%), PR positive (n=343, 62.03%), and HER2 negative (n=402, 72.69%). Tumor characteristics evaluated in association with mammographic density were evaluated, for all cases for invasiveness and stage; and restricted to invasive cases only for tumor size, histology, and receptor status. The associations of mammographic density parameters, overall and stratified by breast cancer tumor characteristics, are shown in Table 3-2. Overall, when comparing the highest quartile with the lowest, dense area (OR 2.6, 95%C.I. 1.8-3.8, p-trend <0.001) 60   and percent dense area (OR 3.8, 95%C.I. 2.5-5.9, p-trend <0.001) were found to be directly related to breast cancer after adjustment for confounders (and non-dense area, for dense-area point estimate); non-dense area (OR 0.5, 95%C.I. 0.3-0.8, p-trend 0.025) was observed to be inversely related to breast cancer, controlling for the adjustment set variables (including dense area). Significant heterogeneity was found only for tumor size, when assessed in relation to percent dense area (p-heterogeneity=0.044); risk did not differ by the other assessed tumor subtypes (p-heterogeneity values >0.05), although there was a suggestion of a heterogeneous association between PR and percent dense area (p=0.071). Sensitivity analyses were performed eliminating observations with imputed values or eliminating controls with late mammograms (tables 3-3 and 3-4). The associations from these analyses were similar, but, borderline significant heterogeneity was found when assessing PR status concerning percent dense area when observations with missing values eliminated (p-heterogeneity=0.014).   3.3 Discussion In this population-based case-control study, a consistent association between mammographic density and breast cancer risk was observed. Overall, the measured mammographic density parameters were found to be important factors for breast cancer in all pathologic subtypes. Dense area and percent dense area were confirmed as 61   independent risk factors directly associated to breast cancer; non-dense area was also found to be an independent factor, inversely related to breast cancer risk. Even though there are some publications that detected differences by hormone receptor status51,53,57,193-196, our observations indicate that these associations do not vary greatly according to tumor characteristics of breast cancer, which is in agreement with various previous reports146,183-192. However, the relatively small sample size of some subgroups (like ER negative or HER2 positive), as well as the inconsistent results regarding PR status heterogeneity in relation to PDA when performing sensitivity analyses, suggest that our study could be underpowered. A strength of this study is that we opted for the DAG approach207 in order to select the covariates for adjustment, minimizing in this way the magnitude of the bias in our estimations. Traditional methods to deal with confounders may increase bias or generate it where it was not present by introducing spurious conditional associations (selection bias)208, which may explain the discrepancies in results of studies assessing associations between tumor characteristics, mammographic density, and breast cancer. Furthermore, the considerable amount of participants’ information gathered in the CBCS made possible to fully adjust by the identified sufficient adjustment set. Other strengths are the objective assessment of mammographic density using the computer-assisted thresholding technique, and the use of craniocaudal views to limit the inclusion of subcutaneous fat in the mammographic density readings174. 62   Some limitations have to be considered. The original CBCS localized cases through the BC Cancer Registry, while controls came from Screening Mammography Program; the different sources for the study population could generate potential selection bias; nonetheless, both population components of the present study were participants in the Screening Mammography Program. Likewise, as mammographic density measurements are not usually revealed to screening participants in BC, it is improbable that breast density influenced enrolment in the study. Some level of recall bias may be present in the information obtained from the questionnaire, used in the models’ adjustment set; however, most likely no important bias is present, given the similarity of the CBCS estimates for known risk factors to those found in other breast cancer studies93. Given that the number of HER2 positive status cases in this study is relatively rather small (n=93), as well as the inconsistent results in regard to heterogeneity related to tumor size and PR status in relation to percent dense area when performing a sensitivity analysis, further studies with larger populations are necessary to better characterize the associations between these tumor characteristics and breast cancer risk.      3.4 Conclusion In conclusion, the findings of the present study indicate that, mammographic density parameters are important factors for breast cancer in all tumor subgroups. Also, overall, said parameters are not associated to breast cancer tumor characteristics. 63   3.5 Tables Table 3-1: Distribution of tumor characteristics on cases Characteristic N (%) Invasiveness In situ 113 (21.94)Invasive 402 (78.06)Breast cancer stage 0 113 (21.94)I 204 (39.61)II 127 (24.66)III 45 (8.74)IV 7 (1.36)Unknown 19 (3.69)Histology* Ductal 341 (84.83)Lobular 26 (6.47)Mixed 11 (2.74)Other 24 (5.97)Tumor size* < 1.1 cm 108 (26.87)1.1 – 2.0 cm 160 (39.80)> 2.0 cm 113 (28.11)Unknown 21 (5.22)ER status* Positive 312 (77.61)Negative 71 (17.66)Unknown 19 (4.73)PR status* Positive 230 (57.21)Negative 153 (38.06)Unknown 19 (4.73)HER2 status* Positive 93 (23.13)Negative 290 (72.14)Unknown 19 (4.73)Phenotype group * (ER|PR+ vs ER&PR-) ER|PR+ 315 (78.36)ER&PR- 68 (16.92)Unknown 19 (4.73)* Invasive cases only.  Table: Copyright ©2019. Dove Medical Press. Velásquez García HA, Gotay CC, et al. Mammographic density parameters and breast cancer tumor characteristics among postmenopausal women. Breast Cancer: Targets and Therapy. In press 2019. 64   Table 3-2: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women  Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Overall              588 477       1  147 86 Reference 144 Reference 76 Reference  2  147 92 1.13 (0.75–1.71) 106 0.76 (0.51–1.12) 78 1.29 (0.84–1.99)  3 147 107 1.34 (0.89–2.01) 112 0.75 (0.49–1.16) 154 3.09 (2.04–4.69)  4 147 192 2.55 (1.74–3.73) 115 0.52 (0.31–0.85) 169 3.84 (2.48–5.95)     p-trend <0.001  p-trend 0.025  p-trend <0.001 Invasiveness In-situ  588 107       1  147 23 Reference 46 Reference 15 Reference  2  147 21 1.02 (0.50–2.07) 26 0.75 (0.40–1.39) 15 1.16 (0.51–2.62)  3 147 26 1.12 (0.56–2.25) 16 0.39 (0.18–0.84) 34 2.81 (1.33–5.97)  4 147 37 1.75 (0.92–3.36) 19 0.41 (0.17–0.98) 43 3.31 (1.51–7.29)     p-trend 0.075  p-trend 0.022  p-trend 0.001 Invasive  588 370       1  147 63 Reference 98 Reference 61 Reference  2  147 71 1.21 (0.77–1.89) 80 0.78 (0.51–1.20) 63 1.38 (0.85–2.19)  3 147 81 1.48 (0.94–2.31) 96 0.91 (0.57–1.46) 120 3.22 (2.05–5.06)  4 147 155 2.84 (1.88–4.29) 96 0.59 (0.34–1.00) 126 4.08 (2.54–6.56)     p-trend <0.001  p-trend 0.148  p-trend <0.001 Invasiveness p-heterogeneity 0.157 0.218 0.689 Histology (Restricted to ductal and lobular invasive subtypes) Ductal  588 310       1  147 56 Reference 85 Reference 54 Reference  2  147 63 1.18 (0.74–1.90) 67 0.76 (0.48–1.20) 53 1.39 (0.85–2.28)  3 147 61 1.27 (0.78–2.05) 78 0.84 (0.51–1.38) 97 2.99 (1.86–4.83)  4 147 130 2.72 (1.76–4.19) 80 0.58 (0.32–1.02) 106 3.87 (2.34–6.38) 65    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)     p-trend <0.001  p-trend 0.146  p-trend <0.001 Lobular  588 26       1  147 3 Reference 7 Reference 4 Reference  2  147 4 1.41 (0.28–7.16) 5 0.89 (0.24–3.27) 2 0.61 (0.10–3.85)  3 147 5 2.04 (0.43–9.67) 8 1.20 (0.32–4.44) 9 3.67 (0.89–15.20)  4 147 14 4.91 (1.24–19.56) 6 0.99 (0.21–4.76) 11 5.08 (1.13–22.80)     p-trend 0.008   p-trend 0.921  p-trend 0.006 Histology p-heterogeneity 0.279 0.590 0.403 Tumor size (missing for 19 invasive cases) < 1.1 cm  588 100       1  147 23 Reference 22 Reference 25 Reference  2  147 20 0.91 (0.45–1.83) 20 1.05 (0.50–2.21) 18 0.73 (0.36-1.52)  3 147 20 0.93 (0.46–1.89) 29 1.49 (0.68–3.29) 31 1.58 (0.81–3.11)  4 147 37 1.65 (0.87–3.12) 29 1.38 (0.57–3.37) 26 1.38 (0.66–2.92)     p-trend 0.094  p-trend 0.336  p-trend 0.179 1.1 – 2.0 cm  588 145       1  147 22 Reference 41 Reference 20 Reference  2  147 26 1.39 (0.71–2.69) 27 0.60 (0.33–1.10) 22 1.77 (0.86–3.63)  3 147 33 1.95 (1.02–3.72) 42 0.93 (0.50–1.72) 47 4.30 (2.20–8.40)  4 147 64 3.46 (1.92–6.22) 35 0.46 (0.22–0.97) 56 6.95 (3.40–14.21)     p-trend <0.001  p-trend 0.139  p-trend <0.001 > 2.0 cm  588 106       1  147 17 Reference 26 Reference 16 Reference  2  147 20 1.61 (0.76–3.44) 26 0.90 (0.47–1.76) 20 2.36 (1.06–5.27)  3 147 21 1.85 (0.86–3.97) 22 0.72 (0.34–1.55) 32 5.28 (2.38–11.71)  4 147 48 4.22 (2.11–8.41) 32 0.61 (0.26–1.43) 38 7.58 (3.30–17.42)     p-trend <0.001  p-trend 0.247  p-trend <0.001 Tumor size p-heterogeneity 0.638 0.379 0.044 66    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Breast cancer stage (missing for 17 cases) Stage 0  588 107       1 147 23 Reference 46 Reference 15 Reference  2 147 21 1.02 (0.50–2.07) 26 0.75 (0.40–1.39) 15 1.16 (0.51–2.62)  3 147 26 1.12 (0.56–2.25) 16 0.39 (0.18–0.84) 34 2.81 (1.33–5.97)  4 147 37 1.75 (0.92–3.36) 19 0.41 (0.17–0.98) 43 3.31 (1.51–7.29)     p-trend 0.075  p-trend 0.022  p-trend 0.001 Stage I  588 189       1 147 38 Reference 47 Reference 37 Reference  2 147 35 0.96 (0.55–1.68) 36 0.74 (0.42–1.29) 30 0.97 (0.54–1.76)  3 147 40 1.23 (0.71–2.14) 54 1.02 (0.57–1.83) 64 2.67 (1.55–4.59)  4 147 76 2.15 (1.30–3.54) 52 0.76 (0.39–1.49) 58 2.74 (1.52–4.94)     p-trend 0.001  p-trend 0.737  p-trend <0.001 Stage II  588 116       1 147 19 Reference 29 Reference 19 Reference  2 147 23 1.55 (0.76–3.16) 31 1.09 (0.57–2.05) 24 2.01 (0.97–4.17)  3 147 22 1.55 (0.74–3.21) 22 0.78 (0.37–1.68) 29 3.25 (1.54–6.86)  4 147 52 3.80 (1.98–7.29) 34 0.66 (0.28–1.51) 44 5.99 (2.79–12.83)     p-trend <0.001  p-trend 0.285  p-trend <0.001 Stage III & IV  588 48       1 147 5 Reference 14 Reference 5 Reference  2 147 9 2.17 (0.64–7.30) 7 0.42 (0.15–1.23) 7 2.96 (0.79–11.03)  3 147 12 3.74 (1.14–12.19) 17 1.01 (0.39–2.65) 17 8.69 (2.53–29.76)  4 147 22 5.14 (1.72 –15.33) 10 0.36 (0.10–1.27) 19 12.64 (3.42–46.64)     p-trend <0.001  p-trend 0.298  p-trend <0.001 Breast ca. stage p-heterogeneity 0.349 0.338 0.516 ER status (missing for 17 invasive cases) Negative  588 66      67    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  1 147 11 Reference 15 Reference 13 Reference  2 147 15 1.60 (0.67–3.82) 12 0.97 (0.29–2.38) 9 0.97 (0.37–2.54)  3 147 11 1.32 (0.52–3.35) 20 1.45 (0.59–3.59) 23 2.99 (1.27–7.03)  4 147 29 3.15 (1.41–7.02) 19 1.11 (0.38–3.22) 21 2.92 (1.15–7.40)     p-trend 0.004  p-trend 0.634  p-trend 0.005 Positive  588 287       1 147 51 Reference 75 Reference 48 Reference  2 147 52 1.13 (0.69–1.84) 62 0.77 (0.49–1.23) 52 1.43 (0.86–2.38)  3 147 63 1.50 (0.93–2.43) 73 0.92 (0.55–1.51) 87 3.10 (1.89–5.08)  4 147 121 2.74 (1.76–4.28) 77 0.60 (0.34–1.07) 100 4.23 (2.52–7.10)     p-trend <0.001  p-trend 0.184  p-trend <0.001 ER status p-heterogeneity 0.639 0.224 0.281 PR status (missing for 17 invasive cases) Negative  588 141       1 147 20 Reference 46 Reference 18 Reference  2 147 23 1.14 (0.57–2.28) 28 0.62 (0.34–1.12) 20 1.74 (0.82–3.70)  3 147 34 1.89 (0.98–3.65) 35 0.75 (0.40–1.42) 46 4.75 (2.35–9.60)  4 147 64 3.34 (1.82–6.11) 32 0.45 (0.20–0.99) 57 6.58 (3.15–13.77)     p-trend <0.001  p-trend 0.098  p-trend <0.001 Positive  588 212       1 147 42 Reference 44 Reference 43 Reference  2 147 44 1.29 (0.76–2.19) 46 0.99 (0.58–1.70) 41 1.31 (0.76–2.25)  3 147 40 1.32 (0.76–2.27) 58 1.23 (0.69–2.17) 64 2.59 (1.52–4.41)  4 147 86 2.64 (1.61–4.30) 64 0.89 (0.47–1.69) 64 3.26 (1.86–5.73)     p-trend <0.001  p-trend 0.938  p-trend <0.001 PR status p-heterogeneity 0.215 0.190 0.071 HER2 status (missing for 17 invasive cases) Negative  588 265       1 147 43 Reference 67 Reference 43 Reference 68    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  2 147 49 1.29 (0.78–2.17) 61 0.82 (0.51–1.32) 43 1.46 (0.85–2.50)  3 147 59 1.71 (1.03–2.85) 61 0.78 (0.46–1.33) 90 3.83 (2.29–6.39)  4 147 114 3.21 (2.01–5.14) 76 0.61 (0.33–1.10) 89 4.88 (2.82–8.44)     p-trend <0.001  p-trend 0.135  p-trend <0.001 Positive  588 88       1 147 19 Reference 23 Reference 18 Reference  2 147 18 0.99 (0.47–2.07) 13 0.66 (0.30–1.44) 18 1.21 (0.58–2.56)  3 147 15 0.91 (0.42–1.98) 32 1.60 (0.76–3.35) 20 1.63 (0.75–3.50)  4 147 36 2.02 (1.05–3.90) 20 0.83 (0.33–2.12) 32 2.63 (1.22–5.65)     p-trend 0.018  p-trend 0.645  p-trend 0.009 HER2 status p-heterogeneity  0.175 0.332 0.112 Phenotype group (ER|PR+ vs ER&PR-) (missing for 17 invasive cases) ER&PR -  588 63       1 147 11 Reference 15 Reference 13 Reference  2 147 13 1.37 (0.55–3.56) 12 0.92 (0.37–2.25) 7 0.73 (0.26–2.05)  3 147 11 1.31 (0.51–3.55) 17 1.12 (0.44–2.85) 22 2.95 (1.24–7.02)  4 147 28 3.07 (1.36–7.60) 19 0.99 (0.34–2.89) 21 3.09 (1.21–7.91)     p-trend 0.004  p-trend 0.881  p-trend 0.002 ER|PR +  588 290       1 147 51 Reference 75 Reference 48 Reference  2 147 54 1.18 (0.72–1.92) 62 0.78 (0.49–1.23) 54 1.51 (0.91–2.49)  3 147 63 1.51 (0.93–2.45) 76 0.96 (0.58–1.59) 88 3.11 (1.90–5.09)  4 147 122 2.77 (1.77–4.32) 77 0.61 (0.34–1.09) 100 4.20 (2.50–7.03)     p-trend <0.001  p-trend 0.221  p-trend <0.001 Phenotype group p-heterogeneity 0.680 0.371  0.420 ‡All models adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption. 1Adjusted for ‡ + dense area. 2Adjusted for ‡ + non-dense area. 69   Adapted table: Copyright ©2019. Dove Medical Press. Velásquez García HA, Gotay CC, et al. Mammographic density parameters and breast cancer tumor characteristics among postmenopausal women. Breast Cancer: Targets and Therapy. In press 2019. 70   Table 3-3: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women (no imputation)   Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Overall            535 435       1  133 75 Reference 146 Reference 63 Reference  2  134 73 1.03 (0.66–1.60) 94 0.66 (0.43–0.99) 63 1.24 (0.77–1.98)  3 134 105 1.38 (0.90–2.12) 97 0.64 (0.41–1.01) 144 3.46 (2.21–5.41)  4 134 182 2.72 (1.82–4.08) 98 0.41 (0.24–0.71) 165 4.43 (2.78–7.06)     p-trend <0.001  p-trend 0.004  p-trend <0.001 Invasiveness In-situ  535 107       1  133 23 Reference 45 Reference 11 Reference  2  134 21 1.22 (0.56–2.68) 24 0.77 (0.40–1.48) 11 1.18 (0.46–3.03)  3 134 26 1.62 (0.77–3.42) 11 0.29 (0.12–0.70) 34 4.51 (1.94–10.48)  4 134 37 2.24 (1.10–4.59) 17 0.39 (0.15–0.98) 41 4.81 (1.99–11.66)     p-trend 0.018  p-trend 0.013  p-trend <0.001 Invasive  535 370       1  133 63 Reference 101 Reference 52 Reference  2  134 71 1.03 (0.63–1.67) 70 0.65 (0.41–1.03) 52 1.27 (0.77–2.11)  3 134 81 1.41 (0.88–2.25) 86 0.80 (0.49–1.30) 110 3.33 (2.06–5.38)  4 134 155 2.90 (1.88–4.48) 81 0.44 (0.25–0.79) 124 4.53 (2.74–7.50)     p-trend <0.001  p-trend 0.033  p-trend <0.001 Invasiveness p-heterogeneity 0.218 0.380 0.888 Histology (Restricted to ductal and lobular invasive subtypes, n=307) Ductal  535 282       1  133 50 Reference 87 Reference 45 Reference  2  134 48 1.02 (0.61–1.69) 61 0.66 (0.41–1.07) 42 1.30 (0.76–2.22)  3 134 61 1.25 (0.76–2.07) 68 0.70 (0.42–1.18) 89 3.19 (1.92–5.30)  4 134 123 2.77 (1.76–4.37) 66 0.43 (0.23–0.79) 106 4.47 (2.63–7.62) 71    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)     p-trend <0.001  p-trend 0.024  p-trend <0.001 Lobular  535 25       1  133 3 Reference 8 Reference 4 Reference  2  134 4 1.35 (0.26–7.10) 4 0.58 (0.15–2.31) 2 0.56 (0.16–3.53)  3 134 5 1.84 (0.38–8.76) 8 0.99 (0.27–3.63) 8 3.19 (0.72–13.64)  4 134 13 4.32 (1.08–17.35) 5 0.56 (0.11–2.89) 11 5.22 (1.12–24.35)     p-trend 0.017  p-trend 0.644  p-trend 0.006 Histology p-heterogeneity 0.554 0.777 0.584 Tumor size (missing for 18 invasive cases) < 1.1 cm  535 101       1  133 20 Reference 23 Reference 22 Reference  2  134 17 0.87 (0.41–1.86) 21 0.98 (0.47–2.05) 17 0.77 (0.37-1.65)  3 134 23 1.14 (0.56–2.33) 27 1.21 (0.54–2.69) 28 1.69 (0.83–3.44)  4 134 34 1.65 (0.84–3.23) 23 0.86 (0.34–2.19) 27 1.61 (0.74–3.49)     p-trend 0.080  p-trend 0.903  p-trend 0.093 1.1 – 2.0 cm  535 149       1  133 20 Reference 42 Reference 18 Reference  2  134 20 1.13 (0.55–2.33) 23 0.49 (0.26–0.93) 17 1.44 (0.66–3.12)  3 134 30 1.79 (0.91–3.52) 38 0.80 (0.42–1.52) 43 4.23 (2.09–8.57)  4 134 64 3.55 (1.93–6.54) 31 0.35 (0.16–0.78) 56 7.48 (3.55–15.76)     p-trend <0.001  p-trend 0.056  p-trend <0.001 > 2.0 cm  535 98       1  133 16 Reference 28 Reference 12 Reference  2  134 13 1.14 (0.48-2.71) 19 0.58 (0.28–1.21) 15 2.42 (0.97–6.06)  3 134 18 1.53 (0.68-3.45) 18 0.54 (0.24–1.23) 29 6.32 (2.60–15.38)  4 134 45 4.24 (2.05-8.79) 27 0.42 (0.17–1.05) 36 10.60 (4.16–26.99)     p-trend <0.001  p-trend 0.087  p-trend <0.001 Tumor size p-heterogeneity 0.577 0.613 0.046 72    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Breast cancer stage (missing for 16 cases) Stage 0  535 97       1 133 18 Reference 45 Reference 11 Reference  2 134 18 1.22 (0.56–2.67) 24 0.77 (0.40–1.48) 11 1.18 (0.46–3.03)  3 134 27 1.62 (0.77–3.41) 11 0.29 (0.12–0.70) 34 4.51 (1.94–10.48)  4 134 34 2.24 (1.10–4.59) 17 0.39 (0.15–0.98) 41 4.82 (1.99–11.66)     p-trend 0.018  p-trend 0.013  p-trend <0.001 Stage I  535 175       1 133 33 Reference 49 Reference 32 Reference  2 134 29 0.89 (0.49–1.63) 34 0.65 (0.37–1.16) 27 0.99 (0.53–1.785)  3 134 41 1.32 (0.75–2.34) 50 0.86 (0.47–1.57) 57 2.79 (1.57–4.96)  4 134 72 2.21 (1.31–3.74) 42 0.50 (0.24–1.02) 59 3.16 (1.70–5.86)     p-trend <0.001  p-trend 0.161  p-trend <0.001 Stage II  535 103       1 133 18 Reference 31 Reference 17 Reference  2 134 15 1.11 (0.50–2.49) 23 0.77 (0.39–1.53) 17 1.53 (0.68–3.46)  3 134 21 1.46 (0.68–3.13) 17 0.61 (0.27–1.37) 27 3.32 (1.50–7.36)  4 134 49 3.77 (1.90–7.50) 32 0.55 (0.23–1.31) 42 6.70 (2.97–15.13)     p-trend <0.001  p-trend 0.193  p-trend <0.001 Stage III & IV  535 44       1 133 5 Reference 14 Reference 3 Reference  2 134 7 1.60 (0.44–5.82) 7 0.43 (0.14–1.25) 6 4.51 (0.94–21.57)  3 134 9 2.48 (0.71–8.58) 16 0.90 (0.33–2.45) 16 13.52 (3.07–59.58)  4 134 23 5.50 (1.82–16.59) 7 0.24 (0.06–0.98) 19 23.47 (4.90–112.5)     p-trend <0.001  p-trend 0.171  p-trend <0.001 Breast ca. stage p-heterogeneity 0.445 0.701 0.505 ER status (missing for 16 invasive cases) Negative  535 58      73    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  1 133 8 Reference 15 Reference 13 Reference  2 134 13 1.80 (0.67–4.84) 11 0.91 (0.35–2.39) 9 1.24 (0.42–3.65)  3 134 9 1.39 (0.48–3.99) 17 1.47 (0.55–3.88) 23 3.83 (1.44–10.21)  4 134 28 4.23 (1.73–10.38) 15 0.90 (0.28–2.91) 21 4.73 (1.66–13.49)     p-trend 0.001  p-trend 0.847  p-trend 0.001 Positive  535 264       1 133 48 Reference 79 Reference 48 Reference  2 134 38 0.87 (0.51–1.49) 53 0.63 (0.39–1.03) 52 1.25 (0.73–2.15)  3 134 62 1.39 (0.84–2.30) 66 0.78 (0.46–1.31) 87 3.05 (1.82–5.13)  4 134 116 2.63 (1.66–4.19) 66 0.45 (0.24–0.83) 100 4.43 (2.58–7.61)     p-trend <0.001  p-trend 0.040  p-trend <0.001 ER status p-heterogeneity 0.823 0.146 0.740 PR status (missing for 16 invasive cases) Negative  535 129       1 133 17 Reference 49 Reference 13 Reference  2 134 17 0.93 (0.43–2.04) 27 0.56 (0.30–1.03) 17 2.09 (0.89–4.87)  3 134 34 2.04 (1.01–4.11) 27 0.52 (0.26–1.03) 40 5.95 (2.69–13.19)  4 134 61 3.61 (1.89–6.91) 26 0.28 (0.12–0.67) 59 10.06 (4.40–23.01)     p-trend <0.001  p-trend 0.007  p-trend <0.001 Positive  535 193       1 133 39 Reference 45 Reference 39 Reference  2 134 34 1.09 (0.61–1.94) 37 0.77 (0.44–1.37) 33 1.11 (0.62–1.98)  3 134 37 1.18 (0.67–2.09) 56 1.14 (0.63–2.06) 60 2.54 (1.45–4.42)  4 134 83 2.63 (1.58–4.38) 55 0.70 (0.35–1.37) 61 3.27 (1.81–5.89)     p-trend <0.001  p-trend 0.608  p-trend <0.001 PR status p-heterogeneity 0.093 0.112 0.014 HER2 status (missing for 16 invasive cases) Negative  535 244       1 133 38 Reference 71 Reference 43 Reference 74    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  2 134 38 1.17 (0.67–2.04) 54 0.65 (0.40–1.08) 43 1.39 (0.78–2.49)  3 134 58 1.70 (1.00–2.89) 53 0.60 (0.35–1.05) 90 3.95 (2.28–6.83)  4 134 110 3.40 (2.07–5.58) 66 0.44 (0.24–0.82) 89 5.73 (3.21–10.23)     p-trend <0.001  p-trend 0.018  p-trend <0.001 Positive  535 78       1 133 18 Reference 23 Reference 18 Reference  2 134 13 0.63 (0.27–1.46) 10 0.73 (0.34–1.54) 18 1.11 (0.51–2.46)  3 134 13 0.73 (0.32–1.68) 30 1.52 (0.74–3.14) 20 1.83 (0.83–4.03)  4 134 34 1.94 (0.98–3.85) 15 0.72 (0.28–1.83) 32 2.76 (1.22–6.22)     p-trend 0.007  p-trend 0.790  p-trend 0.006 HER2 status p-heterogeneity  0.215 0.262 0.132 Phenotype group (ER|PR+ vs ER&PR-) (missing for 16 invasive cases) ER&PR -  535 55       1 133 8 Reference 15 Reference 9 Reference  2 134 11 1.51 (0.54–4.21) 11 0.84 (0.32–2.22) 6 0.90 (0.28–2.87)  3 134 9 1.37 (0.47–3.97) 14 1.08 (0.39–2.97) 18 3.79 (1.39–10.28)  4 134 27 4.12 (1.66–10.23) 15 0.78 (0.24–2.57) 22 5.09 (1.76–14.74)     p-trend 0.001  p-trend 0.871  p-trend <0.001 ER|PR +  535 267       1 133 48 Reference 79 Reference 43 Reference  2 134 40 0.92 (0.54–1.57) 53 0.63 (0.39–1.04) 44 1.32 (0.77–2.27)  3 134 62 1.40 (0.85–2.31) 69 0.82 (0.49–1.38) 82 3.07 (1.83–5.14)  4 134 117 2.66 (1.68–4.23) 66 0.46 (0.25–0.84) 98 4.38 (2.56–7.51)     p-trend <0.001  p-trend 0.053  p-trend <0.001 Phenotype group p-heterogeneity 0.718 0.680  0.975 ‡All models adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption. 1Adjusted for ‡ + dense area. 2Adjusted for ‡ + non-dense area. 75   Table 3-4: Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women (no late controls)   Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Overall            496 477       1  124 86 Reference 148 Reference 78 Reference  2  124 101 1.17 (0.77–1.78) 102 0.71 (0.47–1.07) 78 1.25 (0.80–1.96)  3 124 102 1.24 (0.81–1.89) 111 0.70 (0.45–1.10) 154 2.84 (1.86–4.37)  4 124 188 2.45 (1.65–3.63) 116 0.46 (0.28–0.78) 167 3.59 (2.29–5.62)     p-trend <0.001  p-trend 0.011  p-trend <0.001 Invasiveness In-situ  496 107       1  124 23 Reference 46 Reference 15 Reference  2  124 23 1.02 (0.50–2.07) 26 0.72 (0.38–1.35) 15 1.16 (0.51–2.67)  3 124 25 1.09 (0.54–2.22) 16 0.37 (0.17–0.82) 35 2.80 (1.32–5.97)  4 124 36 1.86 (0.96–3.61) 19 0.38 (0.15–0.91) 42 3.28 (1.49–7.27)     p-trend 0.058  p-trend 0.016  p-trend 0.001 Invasive  496 370       1  124 63 Reference 102 Reference 63 Reference  2  124 78 1.25 (0.79–1.97) 76 0.72 (0.46–1.12) 63 1.31 (0.81–2.11)  3 124 77 1.35 (0.85–2.14) 95 0.84 (0.52–1.36) 119 2.87 (1.81–4.56)  4 124 152 2.67 (1.75–4.13) 97 0.51 (0.30–0.89) 125 3.78 (2.33–6.14)     p-trend <0.001  p-trend 0.069  p-trend <0.001 Invasiveness p-heterogeneity 0.171 0.277 0. 718 Histology (Restricted to ductal and lobular invasive subtypes, n=366) Ductal  496 310       1  124 56 Reference 87 Reference 55 Reference  2  124 69 1.22 (0.76–1.96) 65 0.73 (0.46–1.17) 53 1.32 (0.79–2.19)  3 124 58 1.17 (0.71–1.92) 77 0.79 (0.47–1.32) 97 2.75 (1.69–4.48)  4 124 127 2.56 (1.64–4.01) 81 0.52 (0.29–0.94) 105 3.62 (2.17–6.04) 76    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)     p-trend <0.001  p-trend 0.086  p-trend <0.001 Lobular  496 26       1  124 3 Reference 7 Reference 4 Reference  2  124 4 1.29 (0.25–6.59) 5 0.89 (0.24–3.28) 3 0.90 (0.18–4.65)  3 124 5 1.84 (0.39–8.66) 8 1.11 (0.30–4.08) 8 2.84 (0.69–11.64)  4 124 14 4.67 (1.17–18.53) 6 0.87 (0.18–4.20) 11 4.47 (1.04–19.28)     p-trend 0.009  p-trend 0.958  p-trend 0.013 Histology p-heterogeneity 0.230 0.566 0.512 Tumor size (missing for 19 invasive cases) < 1.1 cm  496 100       1  124 23 Reference 22 Reference 26 Reference  2  124 21 0.94 (0.46–1.92) 20 1.08 (0.50–2.23) 17 0.68 (0.33-1.43)  3 124 19 0.89 (0.43–1.85) 28 1.42 (0.63–3.17) 31 1.43 (0.72–2.83)  4 124 37 1.72 (0.89–3.29) 30 1.37 (0.56–3.37) 26 1.37 (0.65–2.88)     p-trend 0.089  p-trend 0.344  p-trend 0.208 1.1 – 2.0 cm  496 145       1  124 22 Reference 44 Reference 20 Reference  2  124 31 1.55 (0.80–3.00) 24 0.49 (0.26–0.91) 23 1.79 (0.87–3.69)  3 124 31 1.72 (0.89–3.34) 42 0.81 (0.43–1.52) 46 3.87 (1.96–7.62)  4 124 61 3.13 (1.71–5.71) 35 0.37 (0.17–0.80) 56 6.61 (3.22–13.56)     p-trend <0.001  p-trend 0.052  p-trend <0.001 > 2.0 cm  496 106       1  124 17 Reference 27 Reference 17 Reference  2  124 21 1.59 (0.73–3.42) 25 0.85 (0.43–1.67) 20 2.06 (0.93–4.58)  3 124 20 1.65 (0.76–3.60) 22 0.67 (0.31–1.45) 32 4.39 (2.00–9.62)  4 124 48 4.13 (2.04–8.36) 32 0.53 (0.22–1.24) 37 6.22 (2.74–14.12)     p-trend <0.001  p-trend 0.147  p-trend <0.001 Tumor size p-heterogeneity 0.647 0.230 0.046 77    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.) Breast cancer stage (missing for 17 cases) Stage 0  496 113       1 124 23 Reference 46 Reference 15 Reference  2 124 23 1.02 (0.50–2.07) 26 0.72 (0.38–1.35) 15 1.16 (0.51–2.67)  3 124 25 1.09 (0.54–2.22) 16 0.37 (0.17–0.82) 35 2.80 (1.32–5.97)  4 124 36 1.86 (0.96–3.61) 19 0.38 (0.15–0.91) 42 3.29 (1.49–7.27)     p-trend 0.058  p-trend 0.016  p-trend <0.001 Stage I  496 204       1 124 38 Reference 50 Reference 38 Reference  2 124 41 1.09 (0.62–1.90) 33 0.64 (0.36–1.13) 29 0.93 (0.51–1.69)  3 124 37 1.11 (0.62–1.97) 54 0.92 (0.51–1.67) 64 2.46 (1.42–4.25)  4 124 73 1.99 (1.20–3.33) 52 0.63 (0.32–1.24) 58 2.65 (1.46–4.80)     p-trend 0.005  p-trend 0.419  p-trend <0.001 Stage II  496 127       1 124 19 Reference 30 Reference 20 Reference  2 124 24 1.51 (0.73–3.12) 30 1.04 (0.55–2.00) 23 1.74 (0.83–3.65)  3 124 21 1.37 (0.65–2.90) 22 0.75 (0.35–1.62) 29 2.77 (1.32–5.83)  4 124 52 3.41 (1.91–7.24) 34 0.58 (0.25–1.36) 44 5.21 (2.45–11.09)     p-trend <0.001  p-trend 0.207  p-trend <0.001 Stage III & IV  496 52       1 124 5 Reference 14 Reference 5 Reference  2 124 9 2.18 (0.63–7.57) 7 0.42 (0.14–1.20) 9 3.51 (0.97–12.72)  3 124 12 3.77 (1.14–12.54) 16 0.86 (0.32–2.29) 16 7.59 (2.17–26.58)  4 124 22 5.21 (1.71–15.90) 11 0.35 (0.10–1.21) 18 11.04 (2.96–41.19)     p-trend 0.001  p-trend 0.217  p-trend <0.001 Breast ca. stage p-heterogeneity 0.223 0.435 0.657 ER status (missing for 17 invasive cases) Negative  496 66      78    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  1 124 11 Reference 15 Reference 13 Reference  2 124 15 1.59 (0.65–3.86) 12 0.99 (0.40–2.48) 9 1.09 (0.42–2.81)  3 124 12 1.41 (0.56–3.57) 20 1.42 (0.56–3.59) 23 2.89 (1.21–6.88)  4 124 28 2.98 (1.32–6.75) 19 0.97 (0.32–2.86) 21 2.87 (1.12–7.32)     p-trend 0.007  p-trend 0.846  p-trend 0.008 Positive  496 287       1 124 51 Reference 79 Reference 48 Reference  2 124 59 1.20 (0.73–1.97) 58 0.69 (0.43–1.12) 52 1.30 (0.77–2.19)  3 124 58 1.33 (0.81–2.20) 72 0.83 (0.50–1.40) 87 2.69 (1.63–4.44)  4 124 119 2.61 (1.64–4.14) 78 0.52 (0.29–0.94) 100 3.84 (2.27–6.47)     p-trend <0.001  p-trend 0.090  p-trend <0.001 ER status p-heterogeneity 0.689 0.199 0.237 PR status (missing for 17 invasive cases) Negative  496 141       1 124 20 Reference 48 Reference 18 Reference  2 124 26 1.23 (0.61–2.48) 26 0.56 (0.30–1.04) 21 1.83 (0.86–3.90)  3 124 34 1.85 (0.95–3.61) 35 0.69 (0.36–1.33) 46 4.49 (2.20–9.19)  4 124 61 3.06 (1.65–5.68) 32 0.36 (0.16–0.81) 56 6.36 (3.02–13.39)     p-trend <0.001  p-trend 0.036  p-trend <0.001 Positive  496 212       1 124 42 Reference 46 Reference 45 Reference  2 124 48 1.33 (0.78–2.28) 44 0.90 (0.52–1.56) 40 1.17 (0.68–2.04)  3 124 36 1.12 (0.63–1.97) 57 1.11 (0.62–1.99) 63 2.20 (1.29–3.76)  4 124 86 2.58 (1.56–4.26) 65 0.79 (0.42–1.52) 64 2.93 (1.67–5.17)     p-trend <0.001  p-trend 0.730  p-trend <0.001 PR status p-heterogeneity 0.294 0.148 0.075 HER2 status (missing for 17 invasive cases) Negative  496 265       1 124 43 Reference 70 Reference 45 Reference 79    Quartile Dense Area1 Non-dense Area2 Percent Dense Area  Controls Cases OR (95% C.I.) Cases OR (95% C.I.) Cases OR (95% C.I.)  2 124 56 1.40 (0.84–2.35) 58 0.74 (0.46–1.22) 42 1.33 (0.77–2.31)  3 124 54 1.49 (0.88–2.52) 60 0.71 (0.41–1.22) 89 3.29 (1.96–5.52)  4 124 112 3.07 (1.90–4.98) 77 0.53 (0.29–0.98) 89 4.42 (2.55–7.67)     p-trend <0.001  p-trend 0.069  p-trend <0.001 Positive  496 88       1 124 19 Reference 24 Reference 18 Reference  2 124 18 0.92 (0.43–1.96) 12 0.59 (0.26–1.31) 19 1.26 (0.60–2.67)  3 124 16 0.97 (0.45–2.10) 32 1.47 (0.70–3.10) 20 1.56 (0.72–3.38)  4 124 35 1.91 (0.98–3.74) 20 0.69 (0.27–1.77) 31 2.49 (1.16–5.38)     p-trend 0.024  p-trend 0.905  p-trend 0.016 HER2 status p-heterogeneity  0.247 0.343 0.099 Phenotype group (ER|PR+ vs ER&PR-) (missing for 17 invasive cases) ER&PR -  496 63       1 124 11 Reference 15 Reference 13 Reference  2 124 13 1.33 (0.53–3.34) 12 0.94 (0.37–2.35) 8 0.85 (0.31–2.31)  3 124 12 1.41 (0.55–3.60) 17 1.10 (0.42–2.86) 22 2.85 (1.18–6.88)  4 124 27 2.90 (1.27–6.61) 19 0.85 (0.28–2.56) 20 2.93 (1.13–7.55)     p-trend 0.007  p-trend 0.895  p-trend 0.005 ER|PR +  496 290       1 124 51 Reference 79 Reference 50 Reference  2 124 61 1.26 (0.77–2.06) 58 0.70 (0.43–1.13) 53 1.38 (0.82–2.30)  3 124 58 1.34 (0.81–2.21) 75 0.87 (0.52–1.46) 87 2.70 (1.64–4.45)  4 124 120 2.64 (1.66–4.18) 78 0.53 (0.30–0.96) 100 3.83 (2.27–6.44)     p-trend <0.001  p-trend 0.111  p-trend <0.001 Phenotype group p-heterogeneity 0.740 0.624  0.349 ‡All models adjusted for BMI, age, BMI by age interaction, education, ethnicity, age at menarche, parity, age at first full term pregnancy, lifetime breastfeeding, lifetime use of oral contraceptives, family history of breast cancer, lifetime use of hormone replacement therapy, lifetime smoking, and alcohol consumption. 1Adjusted for ‡ + dense area. 2Adjusted for ‡ + non-dense area. 80  4 Genetic variation associated with breast cancer risk and mammographic density parameters  The degree of x-ray attenuation rendered in a mammogram characterizes the mammographic density, as a result of the physical properties of the different types of tissues (epithelial, connective, adipose) found in the breast7. High mammographic density is one of the most important known breast cancer risk factors: for instance, it has been shown that women with more than 75% of the breast area occupied by dense tissue have four to five times the risk of breast cancer compared to women with a mammographic content of dense tissue less than 5%47. A recent meta-analysis by Bond-Smith and Stone (2018) estimated that a change in 10 units of percent mammographic density area increases the odds of breast cancer by 13%209. However, the etiologic determinants of the increment in the risk of breast cancer due to high mammographic density remain mostly unknown126,210,211. Evidence of the important genetic influence on mammographic density has been provided by studies conducted in twin and family populations212,122,213. It has been estimated that more than 60% of the mammographic density variance is determined by a heritable component122,123,213,214. Similarly, compared to the general population, the risk of breast cancer is about two times greater in women with first–degree relatives with breast cancer215,114, pointing towards a familial component216. Around 20% of familial breast cancer cases can be explained by mutations in high risk genes such as BRCA1/2, TP53, PTEN, and PALB2217,218. 81  Growing evidence of shared genetic origins affecting both mammographic density and breast cancer has been described219-221,124,126,127,222-224. It has been conjectured that the dense area is the only mammographic density parameter pertinent in breast cancer etiology225, as most breast tumors have been demonstrated to originate principally from the radiopaque (dense) tissue226. Most studies looking at associations between mammographic density and breast cancer susceptibility loci (also known as single nucleotide polymorphisms or SNPs) have limited their mammographic density assessments to percent mammographic density and/or dense area124,127,218,220. Only a few studies have considered non-dense area126,219,221,223, reporting SNPs associated to both non-dense area and breast cancer risk 126,223. During the past decade, genome-wide association studies (GWAS) have been able to identify more than 300 low-penetrance alleles associated with breast cancer, accounting for approximately 21% of the familial breast cancer risk for ER-positive cases and about 12% for ER-negative tumors227-233. The effect sizes associated with each of these common variants is modest; however, a combined summary expressed as a polygenic risk score (PRS) can be used to explain a substantial fraction of breast cancer risk variation in the general population. Polygenic risk scores can be used to attain a level of risk discrimination in prediction models useful for stratification in population-based programs of breast cancer early detection234,235. Evidence of shared genetic origins affecting both mammographic density and breast cancer has been described in large GWAS124. 82  The purpose of this study is to assess the relationships between the combined genetic effect summarized in a polygenic risk score (three types: overall and weighted by estrogen-receptor subtype) of 313 SNPs previously associated with in GWAS to breast cancer risk and mammographic density measurements (dense area, non-dense area, percent mammographic density). By focusing on the genetics of heritable risk factors of a complex disease like breast cancer, the understanding of the genetic origins of disease can be enhanced236. Consequently, a better understanding of the mammographic density biology and the manner by which it modifies the risk of breast cancer may be gained from studying associations between mammographic density and genetic variants linked to breast cancer237.  4.1 Methods 4.1.1 Study population This study is based on the Canadian Breast Cancer Study (CBCS)93, a population-based case-control study with participants from Greater Vancouver, BC and Kingston, Ontario, with data collection completed in 2010. The participation rates were 54% for cases and 57% for controls. In BC, incident cases were recruited from the BC Cancer Registry, with ages 40 to 80 and diagnosis between 2005 and 2009; cancer-free women were recruited from the Screening Mammography Program, frequency matched to geographic area and five-year age groups. The original BC sample consisted of 1,003 cases and 1,014 controls. For the evaluation of the associations between polygenic risk 83  scores and mammographic density parameters, analyses were restricted to controls only; cases were not utilized because the goal was to examine mammographic density parameters in healthy women, prior to cancer diagnosis.  Analyses included only European descent controls (n=714), since polygenic risk scores were not available for non-Europeans. Given that the biological relevance of menopausal status in the associations that we are examining is unknown, in contrast to the studies described on Chapters 2 and 3 of this thesis, premenopausal and postmenopausal controls were included.     Information about education, ethnicity, medical history, and lifestyle characteristics, was collected through a CBCS questionnaire. Permission was also obtained from participants to access their medical records for data related to breast health and breast tissue blocks gathered as part of their regular care.  4.1.2 Mammographic density measurement  This step is detailed in section 2.1.4; in short, for each subject the most recent mammogram previous to study enrolment was selected. For 140 women, it was not possible to localize mammographic films prior to recruitment into the study; for these individuals, the mammogram closest to study enrolment was selected (mean time=2.3 years after enrollment, SD=0.7). The craniocaudal view of a randomly selected breast was chosen. Mammographic films were digitized using a single scanner (iCAD TotalLook Mammo Advantage). Cumulus software (Imaging Research Program, 84  Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Canada), was used by a single reader (HAVG) to measure total breast area and dense area, by interactive thresholding method29. Within-set intraclass correlation coefficients (ICC) were above 0.99 for total breast area and 0.98 for dense area; between-sets ICC were respectively 0.98 and 0.97. Reliability was assessed by submitting a random sample of no less than 20 images to three experts in mammographic density evaluation via Cumulus; the resultant between-reader ICC oscillated between 0.98 and 0.99 for total breast area, and 0.90 to 0.98 for dense area (table 2.1).  4.1.3 Genotyping Each participant was genotyped using the OncoArray238 as part of a pooled analysis Conducted through the Breast Cancer Association Consortium (BCAC)239, an international multidisciplinary conglomerate of researchers focused on the inherited risk of breast cancer, which includes data from multiple epidemiological breast cancer studies (CBCS is one of them). The genotyping calling, quality control and imputation for Oncoarray, has been previously described229,231. Overall and estrogen-receptor (ER) specific polygenic risk scores were produced by Mavaddat et al233, using subjects of European ancestry composed by 94,094 breast cancer cases and 75,017 controls, belonging to 69 BCAC-affiliated studies, genotyped with either iCOGS229 or Oncoarray231. SNPs with MAF > 0.01 and imputation r2 > 0.9 (OncoArray), and r2 > 0·3 (iCOGS) were utilized for the purpose of evaluating the best SNPs to generate an overall breast cancer polygenic risk score (313 SNPs), as well as ER status specific 85  polygenic risk scores with SNP weightings chosen to maximize prediction of ER specific status. 4.1.4 Statistical analysis  Associations between the three mammographic density parameters (dependent variables) and the three polygenic risk scores (independent variables) were evaluated through multiple linear regression, using age as the sole adjustment covariate, which was entered into all the models as a continuous variable. In order to reduce heteroskedasticity, the mammographic density parameters were log-transformed. Interactions with menopausal status and family history were explored by inclusion of interaction terms in the models. Statistical analyses were computed with R v.3.5.1 (R Foundation for Statistical Computing, Vienna, Austria) and Stata v.14.0 (StataCorp, College Station, Texas, USA).  4.2 Results Table 4-1 shows the characteristics of the study subpopulations used for polygenic risk scores analyses. The associations between polygenic risk scores and the log-transformed mammographic density parameters are illustrated in table 4-2. The ER-negative-specific polygenic risk score was observed to be associated with dense area (for a 10 unit increase in polygenic risk score, β=1.03, 95%CI 0.02–2.03, p=0.046); all the other polygenic risk scores assessments were not statistically significant (p>0.05).  86  No interaction between the three types of polygenic risk scores and menopausal status (table 4-3) or family history of breast cancer was observed in any of the evaluations (table 4-4). The significant association previously found was not present in the stratified analyses, except for the negative family history subgroup (for a 10 unit increase in polygenic risk score β=1.17, 95%CI 0.06 – 2.27, p=0.038).  4.3 Discussion Recently published three varieties of 313-SNPs polygenic risk scores were utilized in the assessments of potential relationships between breast cancer-related genetic variation and mammographic density parameters, in a European descent subpopulation. The findings show evidence of some common genetic determinants for breast cancer risk and mammographic density, specifically when looking at the association between the ER-negative-specific polygenic risk scores and the mammographic dense area. This indicates that the set of SNPs associated with breast cancer risk that are included in the polygenic risk scores (or part of it), influences the characterization of the dense mammary tissue. Unfortunately, polygenic risk scores cannot provide further information to formulate precise hypothesis about the biological mechanism responsible for the observed association.  Further, considering that percent mammographic density is a synthesis measurement of the other two mammographic density parameters, it is 87  possible that the association is not present when assessing the relationship with percent mammographic density due to “dilution” of the signal.  To our knowledge, this is the first time that breast cancer polygenic risk scores have been examined as a potential predictor of mammographic density parameters. A polygenic risk score is, of course, a composite estimate of the total effect of the SNPs involved in its computation, so, it is not possible at this point to state which SNPs in particular are responsible for the relationship seen here, nonetheless, a polymorphism associated with both ER-negative tumors and mammographic dense area, has been described at least in one instance240. Strengths of the present analyses are the comprehensive assessments taking into account the three mammographic density parameters that can be obtained from screening mammograms, the use of novel breast cancer specific PRS, the utilization of the computer-assisted threshold method to assess mammographic density, and the population-based study sample. The main limitation of these analyses is the likely lack of statistical power to detect effect modification in the evaluated associations with regard to family history of breast cancer and menopausal status; for instance, as seen in table 4-3, the stratified analyses by menopausal status indicates no association at all, while the analyses by family history strata shows that only in the larger group (negative family history), the association between polygenic risk scores and mammographic dense area remains significant. Also, since this study was limited to women of European descent, generalization to other ethnic groups is not possible. For these 88  reasons, additional investigation in larger populations is necessary to eventually specify the causal gene(s). 4.4 Conclusions Using a recently developed 313-SNPs polygenic risk score for breast cancer risk, and weighted polygenic risk scores for ER+ and ER- breast cancer risk, this study provides new evidence of shared genetic factors between breast cancer risk and mammographic density, providing insights into the pathways determining breast cancer risk, which could potentially help to formulate prevention strategies, and to improve risk assessment tools. Further research is required using larger samples, for validation and identification of causal genetic variants. 89  4.5 Tables Table 4-1: Characteristics of study population.  Variables Analysis group, Mean (SD) / N (%) Ethnicity European (N=714) Menopausal status (yes) 446 (62.5)Age at study entry (years) 57.5 (10.1)BMI (kg/m2) 25.2 (4.8)Family history of breast cancer (%)* 115 (16.1)Non-dense area (cm2) 115.5 (64.9)Dense area (cm2) 19.2 (15.0)Percent mammographic density (continuous) 17.4 (14.0)PRS□ – Overall (continuous) -0.48 (0.63)PRS□ – ER positive (continuous) -0.48 (0.67)PRS□ – ER negative (continuous) -0.34 (0.62)* Family history of breast cancer was unknown for 24 women. □ PRS: Polygenic risk score         90  Table 4-2: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population  PRS type Parameter β 95% Confidence interval P-value Overall Dense area 0.82 -0.16 – 1.81 0.100 Non-dense area 0.24 -0.39 – 0.87 0.451 Percent mammographic density 0.53 -0.60 – 1.67 0.356 ER positive Dense area 0.66 -0.27 – 1.58 0.163 Non-dense area 0.24 -0.35 – 0.83 0.423 Percent mammographic density 0.38 -0.69 – 1.45 0.484 ER negative Dense area 1.03 0.02 – 2.03 0.046 Non-dense area 0.19 -0.45 – 0.84 0.560 Percent mammographic density 0.78 -0.39 – 1.94 0.192 β – Linear regression coefficient per 10 PRS units.   91  Table 4-3: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population, stratified by menopausal status (premenopausal N=268, postmenopausal N=446)  PRS type Parameter Menopausal statusPre Post P-het β 95% Conf. interval P-value β 95% Conf. interval P-value Overall Dense area 0.97 -0.54 – 2.49 0.206 0.79 -0.50 – 2.07 0.229 0.815 Non-dense area 0.02 -1.11 – 1.15 0.969 0.37 -0.38 – 1.11 0.332 0.549 Percent mammographic density 0.78 -1.02 – 2.59 0.392 0.43 -1.03 – 1.89 0.567 0.711 ER positive Dense area 0.79 -0.64 – 2.23 0.275 0.62 -0.58 – 1.82 0.312 0.814 Non-dense area 0.009 -1.06 – 1.08 0.987 0.38 -0.32 – 1.07 0.288 0.498 Percent mammographic density 0.64 -1.06 – 2.35 0.460 0.26 -1.11 – 1.63 0.709 0.679 ER negative Dense area 1.18 -0.35 – 2.72 0.131 0.98 -0.34 – 2.30 0.146 0.837 Non-dense area 0.03 -1.11 – 1.18 0.118 0.26 -0.50 – 1.03 0.103 0.716 Percent mammographic density 0.98 -0.85 – 2.81 0.295 0.71 -0.80 – 2.22 0.354 0.813 β – Linear regression coefficient per 10 PRS units.   92  Table 4-4: Associations between polygenic risk scores (PRS) and mammographic density parameters in European control population, stratified by family history of breast cancer (negative N=578, positive N=112)*  PRS type Parameter Family history of breast cancerNegative Positive P-het β 95% Conf. interval P-value β 95% Conf. interval P-value Overall Dense area 0.87 -0.17 – 1.92 0.101 0.62 -2.20 – 3.42 0.666 0.325 Non-dense area 0.38 -0.31 – 1.07 0.278 -0.81 -2.36 – 0.74 0.305 0.217 Percent mammographic density 0.43 -0.80 – 1.65 0.495 1.42 -1.66 – 4.49 0.363 0.993 ER positive Dense area 0.70 -0.28 – 1.68 0.161 0.50 -2.23 – 3.23 0.716 0.367 Non-dense area 0.38 -0.27 – 1.03 0.248 -0.88 -2.38 – 0.62 0.246 0.206 Percent mammographic density 0.27 -0.88 – 1.42 0.646 1.38 -1.60 – 4.36 0.362 0.940 ER negative Dense area 1.17 0.06 – 2.27 0.038 0.42 -2.06 – 2.90 0.739 0.837 Non-dense area 0.22 -0.50 – 0.95 0.541 -0.02 -1.39 – 1.35 0.975 0.560 Percent mammographic density 0.85 -0.45 – 2.14 0.199 0.51 -2.20 – 3.23 0.710 0.548 β – Linear regression coefficient per 10 PRS units. * Family history of breast cancer was unknown for 24 participants.  93  5 Conclusion This thesis assesses the risk of breast cancer in relation to mammographic density parameters with emphasis on the absolute non-dense area, as well as the associations between these parameters and breast cancer tumor characteristics.  One of the innovative aspects of this research is the utilization of causal inference methods in a population-based case-control study in order to estimate the causal effect of mammographic density on breast cancer. It also assesses potential common genetic determinants for both mammographic density and breast cancer risk.   5.1 Summary of Study Findings 5.1.1 Estimation of the causal effect of mammographic density parameters over breast cancer in postmenopausal women Chapter 2 estimated the causal exposure effect of mammographic dense, non-dense, and percent dense areas on the risk of breast cancer. It is well-documented that elevated mammographic density, defined qualitatively (e.g., BI-RADS) or quantified (e.g., thresholding) as either percent density or absolute dense area, is an important breast cancer risk factor. However, the role of absolute non-dense area in breast cancer risk is not as clear since the few studies that assessed this relationship have observed conflicting results. These discrepancies can be attributed to potential bias resulting from non-randomly selected samples, as well as potential residual confounding. For this 94  reason, the potential outcomes framework was used as analytical alternative. Specifically, to evaluate objective 1 of this thesis, to determine the association between non-dense area and breast cancer, as well as confirming the association between the other mammographic density parameters and breast cancer, a minimally sufficient set of potential confounders was selected with the help of a direct acyclic diagram (DAG), and computation of marginal (causal) odds ratios was performed with the help of augmented inverse-probability weighted treatment-effect estimators. The results allowed us to conclude that the mammographic non-dense area is a risk factor for breast cancer, independent of dense area and BMI, with a marginal effect inversely related to breast cancer. Also, in agreement with the published literature, we observed that both dense area and percent density are directly related risk factors for breast cancer. Objective 2, to assess the discriminating power provided by the information captured by the non-dense area parameter on models forecasting breast cancer, was addressed by estimating the predictive accuracy of models containing mammographic density parameters. The results indicate that even though, as concluded previously, the evidence favors the notion of non-dense area is an independent factor inversely related with breast cancer risk, the information captured by this mammographic density parameter does not improve the performance of breast cancer risk predictive models containing either the dense area or the percent area parameters.  95  5.1.2 Associations between mammographic density parameters and breast cancer tumor characteristics among postmenopausal women Chapter 3 assessed the relationships between the various mammographic density parameters and the characteristics of breast cancer tumors, the goal of objective 3. Some studies have reported homogeneity in the association between mammographic density and breast cancer tumor characteristics, while others have indicated differences. Most published studies have limited the evaluation of mammographic density qualitatively, or quantitatively in terms of percent dense area. The possibility of differential associations between breast tumor characteristics and the other quantitative mammographic density parameters, absolute dense area, and absolute non-dense area have rarely been considered. If mammographic parameters were associated differentially with breast cancer subtypes or tumor characteristics, such information could potentially help us understand the mechanisms by which mammographic density affects breast cancer. By means of multinomial logistic regression analyses on breast cancer cases only, the presence of heterogeneity was evaluated. Also, utilizing the same minimally sufficient adjustment set identified with the DAG as described on chapter 2, unconditional logistic regression was used to model associations between mammographic density parameters and the risk of breast cancer.  96  Associations between mammographic density parameters and breast cancer risk observed in this population-based case-control study did not vary greatly across pathologic subtypes. In general, the measured mammographic density parameters were important factors for breast cancer for all pathologic subtypes. Dense and percent dense areas were confirmed as independent risk factors directly associated with breast cancer, while the non-dense area was an independent factor, inversely related to breast cancer risk. The observations of this study indicate that these associations do not vary to a large extent according to breast cancer tumor characteristics. These findings are generally congruent with the literature.  5.1.3 Genetic variation associated with breast cancer risk and mammographic density parameters Chapter 4 examines objective 4: to evaluate the relationship between polygenic risk scores developed to predict genetic risk of breast cancer with mammographic density parameters. In doing so, this analysis explores the possibility of common origins at the genetic level for both breast cancer risk and mammographic density. About 5-10% of breast cancer cases are the result of genetic factors, and it is established that the risk of breast cancer is about two times higher in women with positive family history of breast cancer in a first-degree relative. Also, at least 60% of the mammographic density variance is determined by a heritable component. Given all these facts, it makes sense to look for a genetic common ground affecting simultaneously the risk of breast cancer and mammographic density. The majority of studies exploring associations between 97  mammographic density and breast cancer genetic determinants have limited their assessments to percent dense area and/or dense area parameters, with the non-dense area seldom being considered. Three novel 313-SNPs polygenic risk scores (overall and weighted by estrogen-receptor subtype) were used as the exposure of interest in age-adjusted linear regression models of mammographic density parameters. In this way, the combined contributions of a group of low-penetrance alleles associated with breast cancer to mammographic density parameters, were evaluated. To our knowledge, this methodological approach has not been attempted before in this context. Further, potential interactions of the assessed associations with both menopausal status and family history of breast cancer were evaluated. This study found an association between the ER-negative-specific polygenic risk score and the mammographic dense area, adding evidence to the literature of common genetic variants shared by breast cancer risk and mammographic density. No other association was observed. No evidence of interaction with menopausal status or family history of breast cancer was found.  5.2  Strengths and Limitations 98  Most of the strengths and limitations of the research conducted for this doctoral thesis can be found summarized in each Chapter (2 ‒ 4). However, a more thorough description of these topics with additional information follows. Generally speaking, randomized controlled trials are the preferred method to assess the causal effect of an exposure of interest on an outcome. However, this method can seldom be used to examine risk factors for a disease due to technical, economical, and ethical considerations. The potential outcomes framework allows the analysis of epidemiological data in a way that resembles a randomized trial. One of the main assets of this work, as explained in Chapter 2, is the fact that marginal (causal) effect estimates for mammographic density parameters over breast cancer risk were computed. A conditional treatment effect, such as what is obtained by traditional regression adjustment, represents the average effect of the exposure of interest (or treatment) on an individual. In contrast, a marginal exposure effect characterizes the average effect of the exposure in the entire the population241. Marginal estimates are of particular importance in public health since, once an association is ascertained as causal, a potential intervention is highlighted242. From a more practical perspective, the risk estimates calculated using causal inference methodology in Chapter 2, are what would be seen in the population if it was possible to expose all subjects to the same mammographic density parameter level. Causal inference methods in epidemiological studies are not without caveats. No technique, no matter how advanced, can be presumed to be a magic bullet to estimate 99  causality. The assumptions required to estimate causal effects cannot be fully presumed to be present. For instance, the assumption of no unmeasured confounding is often necessary for causal inference in observational studies243. By using a DAG to represent causal relationships, combined with sufficiently rich data, this assumption is plausible. Since the matter of the amount of collected information has been mentioned, it is precisely this factor coming from the original CBCS that made it possible to identify a complex minimally sufficient adjustment set of variables. Conversely, the directed acyclic graph (DAG) used in this research is limited, of course, to what is currently known about breast cancer risk and mammographic density. Also, this particular graph, as is the case with most causal diagrams explaining the relationships between exposure and outcome, is the product of expert opinion. For this reason, some degree of subjectivity cannot be excluded. All these circumstances cast some level of doubt on the validity of the causal inference assumptions. However, as it was mentioned before, the use of an augmented IPW estimator in the analyses described on Chapter 2, makes it possible to estimate causal effects in a reliable, robust manner, as long as at least one of the models (either exposure or outcome) is correctly specified, which provides more certainty about the results of the study. It is not uncommon to find missing data in observational studies. The examination of such data, when limited to complete-case analysis or substituting the missing data with the group mean, is often inefficient (reduction of statistical power), and if the missingness does not depend on randomness, resulting estimates can be severely biased. Although there was not a large amount of missing data for any variable (all 100  variables had <6% missing values), we decided to use a robust form of multiple imputation, the estimation approach introduced by Rubin244. The use of this method constitutes another strength of this research.  By virtue of the multiple imputation by chained equations (MICE) method, implemented in the R mice145 package, missing data was reconstructed by specifying different imputation models, on a variable by variable basis. With this process, the missing values are iteratively imputed depending on the fitted conditional models specified for each variable, until a stopping condition is mollified (convergence). More specifically, mice makes use of predictive mean matching for continuous variables, logistic regression for binary values, Bayesian polytomous regression for categorical variables with more than two levels, and proportional odds models for ordered categorical values with multiple (three or more) levels. MICE assumes that the unknown data are missing at random (MAR); this implies that the probability of missingness for a given value depends on observed values that can be used to predict it: in another way to see it, “any remaining missingness is completely at random”245 after controlling for the available information (i.e., the variables present in the imputation model). Assessment of the missingness mechanism assumption required for MICE is conducted via exploration of missingness patterns, which can be accomplished by using both mice and VIM (Visualization and Imputation of Missing Values) R packages. Advantages of MICE over other imputation methods are mainly that there is no need for distributional assumptions, and the prevention of impossible data assignation. The major disadvantage of this method is the fact that, in contrast to other imputation techniques, MICE does not have a theoretical justification. The way it samples from conditional distributions in an iterative manner is not unlike a Monte Carlo 101  Markov Chain process; however its properties have not been demonstrated246, so the arguments for its use are limited to empirical studies247. Nevertheless, sensitivity analyses excluding imputed values were also performed in the research, and similar estimates were obtained in all instances. In summary, MICE, like other multiple imputation methods used in biomedical investigation, provides improved validity over ad hoc missing data approaches, allows the utilization of all the available data, and conserves statistical power248. An important characteristic of this doctoral thesis is the way the exposures of interest were measured. As explained on Chapter 2, a single reader quantified the mammographic density parameters from screening mammography films. Given that one of the most important source of variability in the assessment of mammographic density is the subjectivity of readers involved in such a task249,250, inter-observer variance is eliminated by having one reader. Further, high reliability was observed in terms of intra-rater and inter-rater ICC for the single reader blinded to the participant’s conditions (as explained on section 2.1.4).  Recent evidence suggests that the quantitative assessment of mammographic density should be expressed in terms of continuous measurements instead of categorical estimates (quantiles), because the precision and the strength of the mammographic density as a breast cancer risk factor can be reduced251,252. While categorical scales are easier to interpret and report, categorization of a continuous measurement results in loss of information with consequential decrease in statistical power253. Also, since the 102  thresholding method used in our studies involves some subjectivity that comes from the reader’s input (the computer facilitates the task, but the final call comes from the reader), some level of measurement error of the continuous measurements is possible, which may generate some degree of misclassification when the continuous parameters are categorized. Although we are aware of the previously described limitations that categorical mammographic density values can potentially cause, historically, categorization has been used since the first publications linking mammographic density to the risk of breast cancer, as well as in the seminal work in 2006 by McCormack and dos Santos Silva47. Moreover, the report of categorical mammographic density measurements resembles classification schemes used in clinical settings, which could potentially facilitate knowledge translation. A limitation of using a minimally sufficient adjustment set containing a relatively large number of variables given the sample size is the decreased statistical power to detect effects, especially interactions. This is evident in the in the lack of statistically significant interactions between mammographic density parameters and breast cancer risk factors in Chapter 2 and  the lack of statistically significant associations in the subtype analyses in Chapter 3. Even though power was low, it was important to look at these given the biological plausibility and potential impact. Reduced statistical power as a consequence of the implementation of a minimally sufficient set of confounders can be seen as the tradeoff for estimation with reduced bias, as well as the possibility of calculation of marginal risk estimates. Data-driven methods could specify a smaller set of confounders with subsequent better statistical power; but, technically such approaches cannot really 103  be considered in causal inference, and it is prone to generate biased estimation of the effect of the exposure of interest254.  5.3  Implications and Future Research Despite the general acknowledgement of mammographic density as an important breast cancer risk factor, this information is not regularly being used in public health or in the clinical arena such as in screening strategies203,255. Nevertheless, this may change in the near future, as policy makers are initiating the notification of breast density to women in North America. For instance, the BC Cancer Breast Screening Program is now providing the mammographic density to health care providers and participants of the program (the screening policy for women with dense breasts remains unchanged)256. Similarly, the US Food and Drug Administration recently published a proposal to update the Mammography Quality Standards Act of 1992; if approved, screened women in the US will be required to receive, together with mammogram results, a letter providing information about their breast density257. The availability of breast density information could shape the way mammographic information is used in public and personalized health.  For example, there is growing evidence of improvements in risk stratification258,259 and risk prediction260 when mammographic density data is added to standard risk assessment algorithms. This process could evolve to clinical research aiming to identify thresholds of mammographic density values characterizing different risk-levels groups 104  in the population, as well as enhanced screening programs consistent with such information. Moreover, there is evidence indicating that mammographic density can be modified by chemotherapy; for example Tamoxifen has been consistently associated to diminution in breast density261-264, and furthermore, these reductions have been linked to lower risk of breast cancer and of recurrence265-268. Lifestyle factors have also been associated to reductions in breast density, such as diet and physical activity269, and its effect could very well be translated in a similar way as in the case with chemotherapy.  Ultimately, more research is required in the areas previously mentioned to make strong, well-founded clinical recommendations consequent to breast density findings. But, even if the biological and molecular pathways from breast density to elevated breast cancer risk are not clear211, the fact is that subpopulation groups with high mammographic density can be distinguished with relative ease in screening programs, and therefore, primary prevention could be potentially focused on these strata of the population, as well as new research efforts delving into the pathogenesis of breast cancer. Our findings confirm that mammographic density parameters are important risk factors for breast cancer, and that their relevance does not vary by the populations or tumor characteristics we studied. Even though lack of statistical power cannot be eliminated as a possibility from the evaluations of effect modification, the consistency of our associations suggests that the performance of breast cancer risk prediction algorithms that include mammographic density parameters most likely will remain unaffected by 105  individual or tumor characteristics, such as family history of breast cancer or estrogen receptor status.  When our research was in its early stages nearly 6 years ago, we hypothesized that the non-dense area component would offer pertinent risk information, in addition to the inclusion of dense area or percent mammographic density. The results of our research indicate that, even though the non-dense area is an independent factor inversely associated with the risk of breast cancer, in prediction/discrimination models in our data, it does not provide additional information to predict breast cancer risk beyond what is provided by either dense area or percent mammographic density. For this reason, further research making use of mammographic density parameters as part of algorithms determining screening guidelines, or as part of breast cancer treatment assessment schemes, should likely focus on the two latter parameters.  Polygenic risk scores were originally conceived as a way to estimate the susceptibility of individuals to a given disease with known multiple genetic loci related to that disease. The observed direct association between genetic variation linked to breast cancer risk and mammographic density independently points towards some common genomic origins shared by both factors. With a composite measure such as a polygenic risk score, it is unfortunately not possible to determine which SNPs are specifically responsible for the associations seen. To determine more precise insights into the mechanisms and genetic determinants shaping the path between mammographic 106  density and breast cancer, future research involving larger sample sizes will be necessary. This thesis used a semi-automated method (Cumulus) for breast density assessment on film-based mammograms. This method would likely be impractical for population-level screening270, given the time involved in the process as well as training required for the reader. However, newer fully automated methods are available for digital mammography which is fully incorporated into most current population screening programs.  5.4  Concluding Remarks The results of these studies provide further evidence supporting importance of mammographic density as a breast cancer risk factor. To examine the marginal (causal) effect of mammographic density parameters (with emphasis on the non-dense area) on breast cancer risk among postmenopausal women, causal inference methods were implemented in a population-based case-control study. Analyses allow us to conclude that the non-dense area is an independent risk factor after adjustment for dense area and other covariates, inversely related with the risk of breast cancer. We can also confirm that that both dense area and percent mammographic density are directly related risk factors for breast cancer. These evaluations also indicate that the non-dense area parameter does not improve discrimination in prediction models over that offered by percent mammographic density or dense area alone. Our studies also show 107  that these associations remain relatively consistent across tumor characteristics. Our assessments using breast cancer risk polygenic risk scores increase the evidence for common genetic determinants affecting mammographic density and breast cancer.  The findings described in this thesis add to the knowledge of the effect of mammographic density parameters on the risk of breast cancer, as well as to probable pathways involving mammographic density and breast cancer. This information contributes to a better etiologic understanding of the disease and in turn to public health efforts such as screening strategies. 108  Bibliography 1. Johnson WD, Koch GG. Intraclass correlation coefficient. In Encyclopedia of Statistical Sciences. New York: John Wiley & Sons. 2004:685-687. 2. Freemantle N, Marston L, Walters K, Wood J, Reynolds MR, Petersen I. Making inferences on treatment effects from real world data: Propensity scores, confounding by indication, and other perils for the unwary in observational research. BMJ. 2013;347. 3. Robins JM, Hernán MÁ, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11(5):550–60. 4. Streiner DL, Norman GR. Health measurement scales. A practical guide to their development and use. Third edition. New York:Oxford University Press; 2003:139-141. 5. Ingleby H, Gerson-Cohen J. Comparative anatomy, pathology and roentgenology of the breast. Philadelphia: University of Philadelphia Press; 1960. 6. Boyd NF, Lockwood GA, Byng JW, Tritchler DL, Yaffe MJ. Mammographic densities and breast cancer risk. Cancer Epidemiol Biomarkers Prev. 1998 Dec;7(12):1133–44. 7. Yaffe MJ. Mammographic density. Measurement of mammographic density. Breast Cancer Res. 2008;10(3):209. 8. Wolfe JN. Breast patterns as an index of risk for developing breast cancer. Am J Roentgenol. 1976;126(6):1130–9. 109  9. Wolfe JN. Risk for breast cancer development determined by mammographic parenchymal pattern. Cancer. 1976;37(5):2486–92. 10. Wolfe JN, Saftlas AF, Salane M. Mammographic parenchymal patterns and quantitative evaluation of mammographic densities: A case-control study. Am J Roentgenol. 1987;148(6):1087–92. 11. Saftlas AF, Szklo M. Mammographic parenchymal patterns and breast cancer risk. Epidemiol Rev. 1987;15(1):146–74. 12. Goodwin P, Boyd NF. Mammographic parenchymal pattern and breast cancer risk: A critical appraisal of the evidence. Am J Epidemiol. 1988;127:1097–108. 13. Oza AM, Boyd NF. Mammographic parenchymal patterns: a marker of breast cancer risk. Epidemiol Rev. 1993;15(1):196–208. 14. Boyd N, Byng J, Jong R, Fishell E, Little L, Miller A, et al. Quantitative classification of mammographic densities and breast cancer risk: results from the Canadian National Breast Screening Study. J Natl Cancer Inst. 1995;87(9):670–5. 15. Lee-Han H, Cooke G, Boyd NF. Quantitative evaluation of mammographic densities: a comparison of methods of assessment. Eur J Cancer Prev. 1995;4(4):285–92. 16. Gram IT, Funkhouser E, Tabár L. The Tabar classification of mammographic parenchymal patterns. Eur J Radiol. 1997;24(2):131–6. 17. Tabár L, Tot T, Dean PB. Breast Cancer: The Art and Science of Early Detection with Mammography. Thieme, editor. New York; 2005. 110  18. Winkel RR, von Euler-Chelpin M, Nielsen M, Diao P, Nielsen MB, Uldall WY, et al. Inter-observer agreement according to three methods of evaluating mammographic density and parenchymal pattern in a case control study: impact on relative risk of breast cancer. BMC Cancer. 2015;15(1):274. 19. Jamal N, Ng K-H, Looi L-M, McLean D, Zulfiqar a, Tan S-P, et al. Quantitative assessment of breast density from digitized mammograms into Tabar’s patterns. Phys Med Biol. 2006;51(22):5843–57. 20. American College of Radiology: Breast Imaging Reporting and Data System (BIRADS). Reston, VA: American College of Radiology; 1993. 21. Carney PA, Miglioretti DL, Yankaskas BC, Kerlikowske K, Rosenberg R, Rutter CM, et al. Individual and combined effects of age, breast density, and hormone replacement therapy use on the accuracy of screening mammography. Ann Intern Med. 2003;138(3):168–75. 22. Buist DSM, Porter PL, Lehman C, Taplin SH, White E. Factors contributing to mammography failure in women aged 40-49 years. J Natl Cancer Inst. 2004;96(19):1432–40. 23. D'Orsi CJ, Sickles EA, Mendelson EB, Morris EA, et al. ACR BI-RADS® Atlas, Breast Imaging Reporting and Data System. Reston, VA: American College of Radiology; 2013. 24. Lam PB, Vacek PM, Geller BM, Muss HB. The association of increased weight, body mass index, and tissue density with the risk of breast carcinoma in Vermont. Cancer. 2000;89(2):369–75. 111  25. Berg WA, Campassi C, Langenberg P, Sexton MJ. Breast Imaging Reporting and Data System: inter- and intraobserver variability in feature analysis and final assessment. AJR Am J Roentgenol. 2000;174(6):1769–77. 26. Kerlikowske K, Grady D, Barclay J, Frankel SD, Ominsky SH, Sickles EA, et al. Variability and accuracy in mammographic interpretation using the American College of Radiology Breast Imaging Reporting and Data System. J Natl Cancer Inst. 1998;90(23):1801–9. 27. Saftlas A, Hoover R, Brinton L, Szklo M, Olson D, Salane M, et al. Mammographic densities and risk of breast cancer. Cancer. 1991;67(11):2833–8. 28. Kato I, Beinart C, Bleich A, Su S, Kim M, Toniolo PG. A nested case-control study of mammographic patterns, breast volume, and breast cancer (New York City, NY, United States). Cancer Causes Control. 1995;6(5):431–8. 29. Byng JW, Boyd NF, Fishell E, Jong RA, Yaffe MJ. The quantitative analysis of mammographic densities. Phys Med Biol. 1994 Oct;39(10):1629–38. 30. Byng JW, Yaffe MJ, Jong RA, Shumak RS, Lockwood GA, Tritchler DL, et al. Analysis of mammographic density and breast cancer risk from digitized mammograms. Radiographics. 1998;18(6):1587–98. 31. Ursin G, Hovanessian-Larsen L, Parisky YR, Pike MC, Wu AH. Greatly increased occurrence of breast cancers in areas of mammographically dense tissue. Breast Cancer Res. 2005;7(5):R605-8. 32. Lokate M, Peeters PHM, Peelen LM, Haars G, Veldhuis WB, van Gils CH. Mammographic density and breast cancer risk: the role of the fat surrounding the fibroglandular tissue. Breast Cancer Res. 2011;13(5):R103. 112  33. Pettersson A, Hankinson SE, Willett WC, Lagiou P, Trichopoulos D, Tamimi RM. Nondense mammographic area and risk of breast cancer. Breast Cancer Res. 2011;13(5):R100. 34.  Pettersson A, Graff RE, Ursin G, Dos Santos Silva I, McCormack V, Baglietto L, et al. Mammographic density phenotypes and risk of breast cancer: A meta-analysis. Vol. 106, Journal of the National Cancer Institute. 2014. 35. Byng JW, Boyd NF, Little L, Lockwood G, Fishell E, Jong RA, et al. Symmetry of projection in the quantitative analysis of mammographic images. Vol. 5, Eur J Cancer Prev. 1996. p. 319–27. 36. Pisano ED, Gatsonis C, Hendrick E, Yaffe M, Baum JK, Acharyya S, et al. Diagnostic Performance of Digital versus Film Mammography for Breast-Cancer Screening. N Engl J Med. Massachusetts Medical Society ; 2005 Oct 27;353(17):1773–83. 37. Kerlikowske K, Hubbard RA, Miglioretti DL, Geller BM, Yankaskas BC, Lehman CD, et al. Comparative effectiveness of digital versus film-screen mammography in community practice in the United States: a cohort study. Ann Intern Med. NIH Public Access; 2011 Oct 18;155(8):493–502. 38. Astley SM, Harkness EF, Sergeant JC, Warwick J, Stavrinos P, Warren R, et al. A comparison of five methods of measuring mammographic density: a case-control study. Breast Cancer Res. BioMed Central; 2018 Dec 5;20(1):10. 39. Canadian Partnership Against Cancer. Breast Cancer Screening in Canada: Environmental Scan. Toronto, ON: Canadian Partnership Against Cancer; 2018. Available at https://s22457.pcdn.co/wp-content/uploads/2019/04/Breast-Cancer-Screening-Environmental-Scan_EN_2018_final.pdf. Retrieved on April 9th, 2019. 113  40. Alonzo-Proulx O, Mawdsley GE, Patrie JT, Yaffe MJ, Harvey JA. Reliability of Automated Breast Density Measurements. Radiology. 2015 May;275(2):366–76. 41. Inc Hologic. Understanding Quantra 2.0 User Manual - MAN-02004 Rev 002. Bedford, Mass: Hologic, 2012. 42. Volpara Solutions from Matakina Technology . Volpara Density User Manual Version 1.5.11. Wellington, New Zealand: Volpara Solutions, 2013. 43. Eng A, Gallant Z, Shepherd J, McCormack V, Li J, Dowsett M, et al. Digital mammographic density and breast cancer risk: a case–control study of six alternative density assessment methods. Breast Cancer Res. BioMed Central; 2014 Oct 20;16(5):439. 44. Byrne C, Schairer C, Wolfe J, Parekh N, Salane M, Brinton L a, et al. Mammographic features and breast cancer risk: effects with time, age, and menopause status. J Natl Cancer Inst. 1995;87(21):1622–9. 45. Maskarinec G, Meng L. A case-control study of mammographic densities in Hawaii. Breast Cancer Res Treat. 2000;63(2):153–61. 46. Tice J, Cummings S, Smith-Bindman R, Ichikawa L, Barlow W, Kerlikowske K. Using clinical factors and mammographic breast density to estimate breast cancer risk: Development and validation of a new predictive model. Ann Intern Med. 2008;148(5):337–47. 47. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev. 2006;15(6):1159–69. 114  48. Boyd NF, Guo H, Martin LJ, Sun L, Stone J, Fishell E, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med. 2007;356(3):227–36. 49. Chiu SYH, Duffy S, Yen AMF, Tabár L, Smith RA, Chen HH. Effect of baseline breast density on breast cancer incidence, stage, mortality, and screening parameters: 25-Year follow-up of a Swedish mammographic screening. Cancer Epidemiol Biomarkers Prev. 2010;19(5):1219–28. 50. Heusinger K, Loehberg CR, Haeberle L, Jud SM, Klingsiek P, Hein A, et al. Mammographic density as a risk factor for breast cancer in a German case–control study. Eur J Cancer Prev. 2011 Jan;20(1):1–8. 51. Yaghjyan L, Colditz GA, Collins LC, Schnitt SJ, Rosner B, Vachon C, et al. Mammographic Breast Density and Subsequent Risk of Breast Cancer in Postmenopausal Women According to Tumor Characteristics. JNCI J Natl Cancer Inst. 2011 Aug 3;103(15):1179–89. 52. Baglietto L, Krishnan K, Stone J, Apicella C, Southey MC, English DR, et al. Associations of mammographic dense and nondense areas and body mass index with risk of breast cancer. Am J Epidemiol. 2014;179(4):475–83. 53. Bertrand KA, Tamimi RM, Scott CG, Jensen MR, Pankratz VS, Visscher D, et al. Mammographic density and risk of breast cancer by age and tumor characteristics. Breast Cancer Res. 2013;15(6):R104. 54. Bond-Smith D, Stone J. Methodological Challenges and Updated Findings from a Meta-analysis of the Association between Mammographic Density and Breast Cancer. Cancer Epidemiol Biomarkers Prev. 2018 Sep 11; [Epub ahead of print] 115  55. Torres-Mejia G, De Stavola B, Allen DS, Pérez-Gavilán JJ, Ferreira JM, Fentiman IS, et al. Mammographic Features and Subsequent Risk of Breast Cancer: A Comparison of Qualitative and Quantitative Evaluations in the Guernsey Prospective Studies. Cancer Epidemiol Biomarkers Prev. 2005 May 1;14(5):1052–9. 56. Stone J, Ding J, Warren RM, Duffy SW, Hopper JL. Using mammographic density to predict breast cancer risk: dense area or percentage dense area. Breast Cancer Res. 2010;12(6):R97. 57. Bertrand KA, Scott CG, Tamimi RM, Jensen MR, Pankratz VS, Norman AD, et al. Dense and nondense Mammographic area and risk of breast cancer by age and tumor characteristics. Cancer Epidemiol Biomarkers Prev. 2015;24(5):798–809. 58. Canadian Cancer Society's Advisory Committee on Cancer Statistics. (2017). Canadian Cancer Statistics 2017. Toronto, ON: Canadian Cancer Society. 59. Ward EM, DeSantis CE, Lin CC, Kramer JL, Jemal A, Kohler B, et al. Cancer statistics: Breast cancer in situ. CA Cancer J Clin. 2015;65(6):481–95. 60. Schottenfeld D, Fraumeni JF. Cancer Epidemiology and Prevention: Oxford University Press, USA, 2006. 61. Longnecker MP, Bernstein L, Paganini-Hill A, Enger SM, Ross RK. Risk factors for in situ breast cancer. Cancer Epidemiol Biomarkers Prev. 1996;5(12):961–5. 62. Trentham-Dietz A, Newcomb PA, Storer BE, Remington PL. Risk factors for carcinoma in situ of the breast. Cancer Epidemiol Biomarkers Prev. 2000;9(7):697–703. 116  63. Rakha EA, Ellis IO. Lobular breast carcinoma and its variants. Semin Diagn Pathol. 2010;27(1):49–61. 64. American Cancer Society. Breast Cancer Facts & Figures 2015-2016. Atlanta, GA:American Cancer Society, Inc. 2015. 65. Ernster VL, Ballard-Barbash R, Barlow WE, Zheng Y, Weaver DL, Cutter G, et al. Detection of ductal carcinoma in situ in women undergoing screening mammography. J Natl Cancer Inst. 2002;94(20):1546–54. 66. Public Health Agency of Canada. Organized Breast Cancer Screening Programs in Canada - Report on Program Performance in 2005 and 2006. Ottawa, Ontario, 2011 67. Hayes, D. Atlas of Breast Cancer. Mosby, 2000. 68. Onitilo AA, Engel JM, Greenlee RT, Mukesh BN. Breast cancer subtypes based on ER/PR and Her2 expression: comparison of clinicopathologic features and survival. Clin Med Res. Marshfield Clinic; 2009 Jun;7(1–2):4–13. 69. Yang XR, Sherman ME, Rimm DL, Lissowska J, Brinton LA, Peplonska B, et al. Differences in risk factors for breast cancer molecular subtypes in a population-based study. Cancer Epidemiol Biomarkers Prev. 2007;16(3):439–43. 70. Anderson WF, Rosenberg PS, Prat A, Perou CM, Sherman ME. How many etiological subtypes of breast cancer: Two, three, four, or more? Vol. 106, Journal of the National Cancer Institute. 2014. 71. Britt K, Ashworth A, Smalley M. Pregnancy and the risk of breast cancer. Vol. 14, Endocrine-Related Cancer. 2007. p. 907–33. 117  72. Singletary SE. Rating the Risk Factors for Breast Cancer. Ann Surg. 2003;237(4):474–82. 73. McPherson K, Steel CM, Dixon JM. Breast cancer — epidemiology , risk factors, and genetics. Br Med J. 2000;321(9):624–8. 74. Beral V, Bull D, Doll R, Peto R, Reeves G. Breast cancer and breastfeeding: Collaborative reanalysis of individual data from 47 epidemiological studies in 30 countries, including 50 302 women with breast cancer and 96 973 women without the disease. Lancet. 2002;360(9328):187–95. 75. Ma H, Bernstein L, Pike MC, Ursin G. Reproductive factors and breast cancer risk according to joint estrogen and progesterone receptor status: a meta-analysis of epidemiological studies. Breast Cancer Res. 2006;8(4):R43. 76. Collaborative Group on Hormonal Factors in Breast Cancer CG on HF in B. Menarche, menopause, and breast cancer risk: individual participant meta-analysis, including 118 964 women with breast cancer from 117 epidemiological studies. Lancet Oncol. 2012;13(11):1141–51. 77. Lambertini M, Santoro L, Del Mastro L, Nguyen B, Livraghi L, Ugolini D, et al. Reproductive behaviors and risk of developing breast cancer according to tumor subtype: A systematic review and meta-analysis of epidemiological studies. Vol. 49, Cancer Treatment Reviews. 2016. p. 65–76. 78. Cheraghi Z, Poorolajal J, Hashem T, Esmailnasab N, Doosti Irani A. Effect of Body Mass Index on Breast Cancer during Premenopausal and Postmenopausal Periods: A Meta-Analysis. Vol. 7, PLoS ONE. 2012. 118  79. Boyd NF, Stone J, Vogt KN, Connelly BS, Martin LJ, Minkin S. Dietary fat and breast cancer risk revisited: a meta-analysis of the published literature. Br J Cancer. 2003;89(9):1672–85. 80. Zhou Y, Zhao H, Peng C. Association of sedentary behavior with the risk of breast cancer in women: Update meta-analysis of observational studies. Vol. 25, Annals of Epidemiology. 2015. p. 687–97. 81. Hamajima N, Hirose K, Tajima K, Rohan T, Calle EE, Heath CW, et al. Alcohol, tobacco and breast cancer--collaborative reanalysis of individual data from 53 epidemiological studies, including 58,515 women with breast cancer and 95,067 women without the disease. Br J Cancer. 2002;87(11):1234–45. 82. Macacu A, Autier P, Boniol M, Boyle P. Active and passive smoking and risk of breast cancer: a meta-analysis. Vol. 154, Breast Cancer Research and Treatment. 2015. p. 213–24. 83. Gaudet MM, Carter BD, Brinton LA, Falk RT, Gram IT, Luo J, et al. Pooled analysis of active cigarette smoking and invasive breast cancer risk in 14 cohort studies. Int J Epidemiol. 2016;(September):dyw288. 84. Clegg LX, Reichman ME, Miller BA, Hankey BF, Singh GK, Lin YD, et al. Impact of socioeconomic status on cancer incidence and stage at diagnosis: selected findings from the surveillance, epidemiology, and end results: National Longitudinal Mortality Study. Cancer Causes Control. Springer Netherlands; 2009 May 12;20(4):417–35. 85. Heck KE, Pamuk ER. Explaining the relation between education and postmenopausal breast cancer. Am J Epidemiol. 1997; 119  86. Larsen SB, Olsen A, Lynch J, Christensen J, Overvad K, Tjønneland A, et al. Socioeconomic position and lifestyle in relation to breast cancer incidence among postmenopausal women: A prospective cohort study, Denmark, 1993-2006. Cancer Epidemiol. 2011; 87. National Institute for Occupational Safety and Health (NIOSH). Polychlorinated Biphenyls (PCB's): Current Intelligence Bulletin 45 (https://www.cdc.gov/niosh/docs/86-111/default.html). Accessed June 2017 88. International Agency for Research on Cancer. List of Classifications by cancer sites with sufficient or limited evidence in humans, Volumes 1 to 119 (http://monographs.iarc.fr/ENG/Classification/index.php). Accessed July 2017 89. National Institute for Occupational Safety and Health (NIOSH). Dieldrin. IOSH Pocket Guide to Chemical Hazards (https://www.cdc.gov/niosh/npg/npgd0206.html). Accessed June 2017 90. Mane SS, Purnell DM, Hsu IC. Genotoxic effects of five polycyclic aromatic hydrocarbons in human and rat mammary epithelial cells. Env Mol Mutagen. 1990;15(2):78–82. 91. Larsen MC, Angus WGR, Brake PB, Eltom SE, Sukow KA, Jefcoate CR. Characterization of CYP1B1 and CYP1A1 Expression in Human Mammary Epithelial Cells: Role of the Aryl Hydrocarbon Receptor in Polycyclic Aromatic Hydrocarbon Metabolism. Cancer Res. 1998;58(11):2366–74. 92. Rundle A, Tang DL, Hibshoosh H, Estabrook A, Schnabel F, Cao WF, et al. The relationship between genetic damage from polycyclic aromatic hydrocarbons in breast tissue and breast cancer. Carcinogenesis. 2000;21(7):1281–9. 120  93. Grundy A, Richardson H, Burstyn I, Lohrisch C, SenGupta SK, Lai AS, et al. Increased risk of breast cancer associated with long-term shift work in Canada. Occup Environ Med. 2013;70(12):831–8. 94. Lin X, Chen W, Wei F, Ying M, Wei W, Xie X. Night-shift work increases morbidity of breast cancer and all-cause mortality: A meta-analysis of 16 prospective cohort studies. Vol. 16, Sleep Medicine. 2015. p. 1381–7. 95. Järup L, Åkesson A. Current status of cadmium as an environmental health problem. Vol. 238, Toxicology and Applied Pharmacology. 2009. p. 201–8. 96. Byrne C, Divekar SD, Storchan GB, Parodi DA, Martin MB. Cadmium - A metallohormone? Vol. 238, Toxicology and Applied Pharmacology. 2009. p. 266–71. 97. McElroy JA, Shafer MM, Trentham-Dietz A, Hampton JM, Newcomb PA. Cadmium exposure and breast cancer risk. J Natl Cancer Inst. 2006;98(12):869–73. 98. Adams S V., Newcomb PA, Shafer MM, Atkinson C, Bowles EJA, Newton KM, et al. Urinary cadmium and mammographic density in premenopausal women. Breast Cancer Res Treat. 2011;128(3):837–44. 99. Adams S V, Hampton JM, Trentham-Dietz A, Gangnon RE, Shafer MM, Newcomb PA. Urinary Cadmium and Mammographic Density. Epidemiology. 2017;28(1). 100. Yaghjyan L, Arao R, Brokamp C, O’Meara ES, Sprague BL, Ghita G, et al. Association between air pollution and mammographic breast density in the Breast Cancer Surveilance Consortium. Breast Cancer Res. 2017;19(1):36. 121  101. Collaboratative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormonal contraceptives: collaborative reanalysis of individual data on 53,297 women with breast cancer and 100,239 women without breast cancer from 54 epidemiological studies. Lancet. 1996;347(9017):1713–27. 102. Gierisch JM, Coeytaux RR, Urrutia RP, Havrilesky LJ, Moorman PG, Lowery WJ, et al. Oral contraceptive use and risk of breast, cervical, colorectal, and endometrial cancers: a systematic review. Cancer Epidemiol Biomarkers Prev. 2013;22(11):1931–43. 103. Moorman PG, Havrilesky LJ, Gierisch JM, Coeytaux RR, Lowery WJ, Urrutia RP, et al. Oral contraceptives and risk of ovarian cancer and breast cancer among high-risk women: A systematic review and meta-analysis. J Clin Oncol. 2013;31(33):4188–98. 104. Beral V. Breast cancer and hormone-replacement therapy in the Million Women Study. Lancet. 2003;362(9382):419–27. 105. Chlebowski RT, Manson JE, Anderson GL, Cauley JA, Aragaki AK, Stefanick ML, et al. Estrogen plus progestin and breast cancer incidence and mortality in the women’s health initiative observational study. J Natl Cancer Inst. 2013;105(8):526–35. 106. Ross RK, Paganini-Hill A, Wan PC, Pike MC. Effect of hormone replacement therapy on breast cancer risk: estrogen versus estrogen plus progestin. J Natl Cancer Inst. 2000;92(4):328–32. 107. Reeves GK, Beral V, Green J, Gathani T, Bull D. Hormonal therapy for menopause and breast-cancer risk by histological type: a cohort study and meta-analysis. Lancet Oncol. 2006;7(11):910–8. 122  108. Collaboratative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormone replacement therapy: collaborative reanalysis of data from 51 epidemiological studies of 52 705 women with breast cancer and 108 411 women without breast cancer. Lancet. 1997;350(9084):1047–59. 109. Clemons M, Goss P. Estrogen and the risk of breast cancer. N Engl J Med. 2001;344(4):276–85. 110. Henderson BE, Feigelson HS. Hormonal carcinogenesis. Carcinogenesis. 2000;21(3):427–33. 111. El-Bastawissi a Y, White E, Mandelson MT, Taplin SH. Reproductive and hormonal factors associated with mammographic breast density by age (United States). Cancer Causes Control. 2000;11(10):955–63. 112. Boyd NF, Melnichouk O, Martin LJ, Hislop G, Chiarelli AM, Yaffe MJ, et al. Mammographic density, response to hormones, and breast cancer risk. J Clin Oncol. 2011;29(22):2985–92. 113. Pharoah PD, Day NE, Duffy S, Easton DF, Ponder BA. Family history and the risk of breast cancer: a systematic review and meta-analysis. Int J cancer. 1997;71(5):800–9. 114. Collaborative Group on Hormonal Factors in Breast Cancer. Familial breast cancer: collaborative reanalysis of individual data from 52 epidemiological studies including 58,209 women with breast cancer and 101,986 women without the disease. Lancet. 2001 Oct 27;358(9291):1389–99. 115. Schottenfeld D, Fraumeni JF. Cancer Epidemiology and Prevention. Oxford University Press: USA; 2006 123  116. Zöller B, Li X, Sundquist J, Sundquist K. Familial transmission of prostate, breast and colorectal cancer in adoptees is related to cancer in biological but not in adoptive parents: A nationwide family study. Eur J Cancer. 2014;50(13):2319–27. 117. Lee M, Czene K, Rebora P, Reilly M. Patterns of changing cancer risks with time since diagnosis of a sibling. Int J Cancer. 2015;136(8):1948–56. 118. Hall JM, Lee MK, Newman B, Morrow JE, Anderson LA, Huey B, et al. Linkage of Early-Onset Familial Breast Cancer to Chromosome 17q21. Science (80- ). 1990;250:1684–9. 119. Antoniou A, Pharoah PDP, Narod S, Risch HA, Eyfjord JE, Hopper JL, et al. Average risks of breast and ovarian cancer associated with BRCA1 or BRCA2 mutations detected in case Series unselected for family history: a combined analysis of 22 studies. Am J Hum Genet. 2003;72(5):1117–30. 120. Easton D, Ford D, Peto J. Inherited susceptibility to breast cancer. Cancer Surv. 1993;18:95–113. 121. Bennett IC, Gattas M, Teh BT. The genetic basis of breast cancer and its clinical implications. Aust N Z J Surg. 1999;69(2):95–105. 122. Boyd NF, Dite GS, Stone J, Gunasekara A, English DR, McCredie MRE, et al. Heritability of mammographic density, a risk factor for breast cancer. N Engl J Med. 2002;347(12):886–94. 123. Stone J, Dite GS, Gunasekara A, English DR, McCredie MRE, Giles GG, et al. The heritability of mammographically dense and nondense breast tissue. Cancer Epidemiol Biomarkers Prev. 2006;15(4):612–7. 124  124. Lindström S, Vachon CM, Li J, Varghese J, Thompson D, Warren R, et al. Common variants in ZNF365 are associated with both mammographic density and breast cancer risk. Nat Genet. 2011;43(3):185–7. 125. Stevens KN, Lindstrom S, Scott CG, Thompson D, Sellers TA, Wang X, et al. Identification of a novel percent mammographic density locus at 12q24. Hum Mol Genet. 2012;21(14):3299–305. 126. Vachon CM, Scott CG, Fasching PA, Hall P, Tamimi RM, Li J, et al. Common breast cancer susceptibility variants in LSP1 and RAD51L1 are associated with mammographic density measures that predict breast cancer risk. Cancer Epidemiol Biomarkers Prev. 2012;21(7):1156–66. 127. Lindström S, Thompson DJ, Paterson AD, Li J, Gierach GL, Scott C, et al. Genome-wide association study identifies multiple loci associated with both mammographic density and breast cancer risk. Nat Commun. 2014;5:5303. 128. Zhang B, Beeghly-Fadiel A, Long J, Zheng W. Genetic variants associated with breast-cancer risk: Comprehensive research synopsis, meta-analysis, and epidemiological evidence. Lancet Oncol. 2011;12(5):477–88. 129. Mitrunen K, Hirvonen A. Molecular epidemiology of sporadic breast cancer. The role of polymorphic genes involved in oestrogen biosynthesis and metabolism. Mutat Res. 2003;544(1):9–41. 130. The International SNP Map Working Group. A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature. 2001;409(6822):928–33. 125  131. Marias K, Petroudi S, English R, Adams R, Brady M. Subjective and computer-based characterization of mammographic patterns. In: Digital Mammography. Berlin, Heidelberg: Springer Berlin Heidelberg; 2003. p. 552–6. 132. BC Cancer. BC Cancer Screening: Breast Density - How is breast density determined? Available at http://www.bccancer.bc.ca/screening/breast/breast-health/breast-density. Retrieved on January 14th, 2019. 133. Heckman, JJ.Sample Selection Bias as a Specification Error. Econometrica. 1979; 47, 151–161. 134. Greenland S, Morgenstern H. Confounding in health research. Ann Rev Public Health. 2001;22:189–212. 135. Holland PW. Statistics and Causal Inference. J Am Stat Assoc. 1986 Dec;81(396):945. 136. Pearl J. Causality. New York: Springer; 2000. 137. Ursin G, Longnecker MP, Haile RW, Greenland S. A meta-analysis of body mass index and risk of premenopausal breast cancer. Epidemiology. 1995;6(2):137–41. 138. Key TJ, Appleby PN, Reeves GK, Roddam A, Dorgan JF, Longcope C, et al. Body mass index, serum sex hormones, and breast cancer risk in postmenopausal women. J Natl Cancer Inst. 2003;95(16):1218–26. 139. Friedenreich CM, Courneya KS, Bryant HE. Influence of physical activity in different age and life periods on the risk of breast cancer. Epidemiology. 2001;12(6):604–12. 126  140. Imaging Research Program, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Canada 141. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10(1):37–48. 142. Pearl J. Causality: Models, Reasoning, and Inference. Cambridge: Cambridge University Press; 2000. 143. Textor J, Hardt J, Knüppel S. DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology. 2011 Sep;22(5):745. 144. Didelez V, Kreiner S, Keiding N. Graphical Models for Inference Under Outcome-Dependent Sampling. Stat Sci. 2010;25(3):368–87. 145. Buuren S van, Groothuis-Oudshoorn K. mice : Multivariate Imputation by Chained Equations in R. J Stat Softw. 2011;45(3):1–67. 146. Maskarinec G, Dartois L, Delaloge S, Hopper J, Clavel-Chapelon F, Baglietto L. Tumor characteristics and family history in relation to mammographic density and breast cancer: The French E3N cohort. Cancer Epidemiol. 2017;49:156–60. 147. Rice MS, Bertrand KA, VanderWeele TJ, Rosner BA, Liao X, Adami H-O, et al. Mammographic density and breast cancer risk: a mediation analysis. Breast Cancer Res. 2016;18(1):94. 148. Boyd NF, Melnichouk O, Martin LJ, Hislop G, Chiarelli AM, Yaffe MJ, et al. Mammographic density, response to hormones, and breast cancer risk. J Clin Oncol. 2011;29(22):2985–92. 149. Vittinghoff E, Glidden DV, Shiboski SC, McCulloch CE. Chapter 9: Strengthening Causal Inference. In Regression Methods in Biostatistics: Linear, Logistic, 127  Survival, and Repeated Measures Models. 2nd ed. New York, NY: Springer; 2012. p. 331-394. 150. StataCorp. Stata treatment-effects reference manual: potential outcomes/counterfactual outcomes. Release 14; 2015. Available at http://www.stata.com/manuals14/te.pdf; 2013. Retrieved on December 20th, 2016. 151. Busso M, DiNardo J, McCrary J. New Evidence on the Finite Sample Properties of Propensity Score Reweighting and Matching Estimators. Rev Econ Stat. The MIT Press ; 2014 Dec;96(5):885–97. 152. StataCorp. Stata treatment-effects reference manual: teffects overlap. Release 14; 2015. Available at http://www.stata.com/manuals14/teteffectsoverlap.pdf. Retrieved on December 20th, 2016. 153. Vittinghoff E, Glidden DV, Shiboski SC, McCulloch CE. Chapter 9: Strengthening Causal Inference. In Regression Methods in Biostatistics: Linear, Logistic, Survival, and Repeated Measures Models. 2nd ed. New York, NY: Springer; 2012. p. 359-360. 154. Oehlert GW. A note on the delta method. Am Stat. 1992;46(1):27–9. 155. R Foundation for Statistical Computing, Vienna, Austria. 156. Stata Corporation, College Station, Texas, USA. 157. Ching S, Kashinkunti S, Niehaus MD, Zinser GM. Mammary adipocytes bioactivate 25-hydroxyvitamin D3 and signal via vitamin D3 receptor, modulating mammary epithelial cell growth. J Cell Biochem. 2011 Nov;112(11):3393–405. 128  158. Ghosh K, Hartmann LC, Reynolds C, Visscher DW, Brandt KR, Vierkant RA, et al. Association Between Mammographic Density and Age-Related Lobular Involution of the Breast. J Clin Oncol. 2010 May 1;28(13):2207–12. 159. Sellers T, Vachon C, Pankratz V, Janney C, Fredericksen Z, Brandt K, et al. Association of Childhood and Adolescent Anthropometric Factors, Physical Activity, and Diet with Adult Mammographic Breast Density. Am J Epidemiol. 2007 Jun 14;166(4):456–64. 160. Samimi G, Colditz GA, Baer HJ, Tamimi RM. Measures of energy balance and mammographic density in the Nurses’ Health Study. Breast Cancer Res Treat. 2008 May 26;109(1):113–22. 161. Fuemmeler BF, Pendzich MK, Tercyak KP. Weight, Dietary Behavior, and Physical Activity in Childhood and Adolescence: Implications for Adult Cancer Risk. Obes Facts. 2009;2(3):179–86. 162. Lope V, Pérez-Gómez B, Moreno MP, Vidal C, Salas-Trejo D, Ascunce N, et al. Childhood factors associated with mammographic density in adult women. Breast Cancer Res Treat. 2011 Dec 12;130(3):965–74. 163. Fagherazzi G, Guillas G, Boutron-Ruault M-C, Clavel-Chapelon F, Mesrine S. Body shape throughout life and the risk for breast cancer at adulthood in the French E3N cohort. Eur J Cancer Prev. 2013 Jan;22(1):29–37. 164. Baglietto L, English DR, Hopper JL, Morris HA, Tilley WD, Giles GG. Circulating Insulin-Like Growth Factor-I and Binding Protein-3 and the Risk of Breast Cancer. Cancer Epidemiol Biomarkers Prev. 2007 Apr 1;16(4):763–8. 165. Poole EM, Tworoger SS, Hankinson SE, Schernhammer ES, Pollak MN, Baer HJ. Body Size in Early Life and Adult Levels of Insulin-like Growth Factor 1 and 129  Insulin-like Growth Factor Binding Protein 3. Am J Epidemiol. 2011 Sep 15;174(6):642–51 166. Simpson ER, Clyne C, Rubin G, Boon WC, Robertson K, Britt K, et al. Aromatase—A Brief Overview. Annu Rev Physiol. 2002 Mar;64(1):93–127. 167. McInnes KJ, Brown KA, Knower KC, Chand AL, Clyne CD, Simpson ER. Characterisation of aromatase expression in the human adipocyte cell line SGBS. Breast Cancer Res Treat. 2008 Dec 5;112(3):429–35. 168. Cole SR, Hernan MA. Constructing Inverse Probability Weights for Marginal Structural Models. Am J Epidemiol. 2008 Jul 15;168(6):656–64. 169. Cattaneo MD. Efficient semiparametric estimation of multi-valued treatment effects under ignorability. J Econom. North-Holland; 2010;155(2):138–54. 170. Austin PC. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Stat Med. 2014 Mar 30;33(7):1242–58. 171. Rubin DB. Using Multivariate Matched Sampling and Regression Adjustment to Control Bias in Observational Studies. J Am Stat Assoc. 1979;74(366a):318–28. 172. Gutman R, Rubin DB. Estimation of causal effects of binary treatments in unconfounded studies. Stat Med. 2015;34(26):3381-3398. 173. Lunceford JK, Davidian M. Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study. Stat Med. 2004 Oct 15;23(19):2937–60. 174. Shepherd JA, Kerlikowske K. Do fatty breasts increase or decrease breast cancer risk? Breast Cancer Res. 2012;14(1):102. 130  175. Potter JD, Cerhan JR, Sellers TA, McGovern PG, Drinkard C, Kushi LR, et al. Progesterone and estrogen receptors and mammary neoplasia in the Iowa Women’s Health Study: how many kinds of breast cancer are there? Cancer Epidemiol Biomarkers Prev. 1995 Jun;4(4):319–26. 176. Huang WY, Newman B, Millikan RC, Schell MJ, Hulka BS, Moorman PG. Hormone-related factors and risk of breast cancer in relation to estrogen receptor and progesterone receptor status. Am J Epidemiol. 2000 Apr 1;151(7):703–14. 177. Colditz GA, Rosner BA, Chen WY, Holmes MD, Hankinson SE. Risk factors for breast cancer according to estrogen and progesterone receptor status. J Natl Cancer Inst. 2004 Feb 4;96(3):218–28. 178. Rusiecki JA, Holford TR, Zahm SH, Zheng T. Breast cancer risk factors according to joint estrogen receptor and progesterone receptor status. Cancer Detect Prev. 2005 Jan;29(5):419–26. 179. Ma H, Bernstein L, Ross RK, Ursin G. Hormone-related risk factors for breast cancer in women under age 50 years by estrogen and progesterone receptor status: results from a case–control and a case–case comparison. Breast Cancer Res. 2006 Aug 17;8(4):R39. 180. Ma H, Bernstein L, Pike MC, Ursin G. Reproductive factors and breast cancer risk according to joint estrogen and progesterone receptor status: a meta-analysis of epidemiological studies. Breast Cancer Res. 2006 Aug 19;8(4):R43. 181. Gaudet MM, Press MF, Haile RW, Lynch CF, Glaser SL, Schildkraut J, et al. Risk factors by molecular subtypes of breast cancer across a population-based study of women 56 years or younger. Breast Cancer Res Treat. 2011 Nov 11;130(2):587–97. 131  182. Turkoz FP, Solak M, Petekkaya I, Keskin O, Kertmen N, Sarici F, et al. Association between common risk factors and molecular subtypes in breast cancer patients. The Breast. 2013 Jun;22(3):344–50. 183. Ziv E, Tice J, Smith-Bindman R, Shepherd J, Cummings S, Kerlikowske K. Mammographic density and estrogen receptor status of breast cancer. Cancer Epidemiol Biomarkers Prev. 2004 Dec;13(12):2090–5. 184. Gill JK, Maskarinec G, Pagano I, Kolonel LN. The association of mammographic density with ductal carcinoma in situ of the breast: the Multiethnic Cohort. Breast Cancer Res. 2006 Jun 23;8(3):R30. 185. Yang W-T, Dryden M, Broglio K, Gilcrease M, Dawood S, Dempsey PJ, et al. Mammographic features of triple receptor-negative primary breast cancers in young premenopausal women. Breast Cancer Res Treat. 2008 Oct 17;111(3):405–10. 186. Ma H, Luo J, Press MF, Wang Y, Bernstein L, Ursin G. Is there a difference in the association between percent mammographic density and subtypes of breast cancer? Luminal A and triple-negative breast cancer. Cancer Epidemiol Biomarkers Prev. 2009 Feb 3;18(2):479–85. 187. Phipps AI, Li CI, Kerlikowske K, Barlow WE, Buist DSM. Risk factors for ductal, lobular, and mixed ductal-lobular breast cancer in a screening population. Cancer Epidemiol Biomarkers Prev. 2010 Jun 1;19(6):1643–54. 188. Phipps AI, Buist DSM, Malone KE, Barlow WE, Porter PL, Kerlikowske K, et al. Breast density, body mass index, and risk of tumor marker-defined subtypes of breast cancer. Ann Epidemiol. 2012 May;22(5):340–8. 132  189. Eriksson L, Hall P, Czene K, dos Santos Silva I, McCormack V, Bergh J, et al. Mammographic density and molecular subtypes of breast cancer. Br J Cancer. 2012 Jun 26;107(1):18–23. 190. Antoni S, Sasco AJ, Dos Santos Silva I, McCormack V. Is mammographic density differentially associated with breast cancer according to receptor status? A meta-analysis. Breast Cancer Res Treat. 2013;137(2):337–47. 191. Pollán M, Ascunce N, Ederra M, Murillo A, Erdozáin N, Alés-Martínez J, et al. Mammographic density and risk of breast cancer according to tumor characteristics and mode of detection: a Spanish population-based case-control study. Breast Cancer Res. 2013 Jan 29;15(1):R9. 192. Krishnan K, Baglietto L, Stone J, McLean C, Southey MC, English DR, et al. Mammographic density and risk of breast cancer by tumor characteristics: a case-control study. BMC Cancer. BioMed Central; 2017 Dec 16;17(1):859. 193. Ding J, Warren R, Girling A, Thompson D, Easton D. Mammographic Density, Estrogen Receptor Status and Other Breast Cancer Tumor Characteristics. Breast J. 2010 Apr 12;16(3):279–89. 194. Conroy SM, Pagano I, Kolonel LN, Maskarinec G. Mammographic density and hormone receptor expression in breast cancer: The Multiethnic Cohort Study. Cancer Epidemiol. 2011 Oct;35(5):448–52. 195. Heusinger K, Jud SM, Häberle L, Hack CC, Adamietz BR, Meier-Meitinger M, et al. Association of mammographic density with hormone receptors in invasive breast cancers: results from a case-only study. Int J cancer. 2012 Dec 1;131(11):2643–9. 133  196. Kerlikowske K, Gard CC, Tice JA, Ziv E, Cummings SR, Miglioretti DL, et al. Risk Factors That Increase Risk of Estrogen Receptor–Positive and –Negative Breast Cancer. J Natl Cancer Inst. 2017 May 31;109(5):djw276. 197. Aiello EJ, Buist DSM, White E, Porter PL. Association between Mammographic Breast Density and Breast Cancer Tumor Characteristics. Cancer Epidemiol Biomarkers Prev. 2005 Mar 1;14(3):662–8. 198. Razzaghi H, Troester MA, Gierach GL, Olshan AF, Yankaskas BC, Millikan RC. Association between mammographic density and basal-like and luminal A breast cancer subtypes. Breast Cancer Res. 2013 Oct 4;15(5):R76. 199. Baré M, Torà N, Salas D, Sentís M, Ferrer J, Ibáñez J, et al. Mammographic and clinical characteristics of different phenotypes of screen-detected and interval breast cancers in a nationwide screening program. Breast Cancer Res Treat. 2015 Nov 3;154(2):403–15. 200. Sala E, Solomon L, Warren R, McCann J, Duffy S, Luben R, et al. Size, node status and grade of breast tumours: association with mammographic parenchymal patterns. Eur Radiol. 2000 Jan 10;10(1):157–61. 201. Porter GJR, Evans AJ, Cornford EJ, Burrell HC, James JJ, Lee AHS, et al. Influence of Mammographic Parenchymal Pattern in Screening-Detected and Interval Invasive Breast Cancers on Pathologic Features, Mammographic Features, and Patient Survival. Am J Roentgenol. 2007 Mar;188(3):676–83. 202. Kerlikowske K, Cook AJ, Buist DSM, Cummings SR, Vachon C, Vacek P, et al. Breast Cancer Risk by Breast Density, Menopause, and Postmenopausal Hormone Therapy Use. J Clin Oncol. 2010 Aug 20;28(24):3830–7. 134  203. Evans DGR, Warwick J, Astley SM, Stavrinos P, Sahin S, Ingham S, et al. Assessing individual breast cancer risk within the U.K. National Health Service Breast Screening Program: a new paradigm for cancer prevention. Cancer Prev Res. 2012 Jul 1;5(7):943–51. 204. Dubin N, Pasternack BS. Risk assessment for case-control subgroups by polychotomous logistic regression. Am J Epidemiol. 1986;123(6):1101–17. 205. Begg CB, Zhang ZF. Statistical analysis of molecular epidemiology studies employing case-series. Cancer Epidemiol Biomarkers Prev. 1994;3(2):173–5. 206. Kowarik A, Templ M. Imputation with the R Package VIM. Journal of Statistical Software. 2016; 74(7), 1-16. 207. Shrier I, Platt RW. Reducing bias through directed acyclic graphs. BMC Med Res Methodol. BioMed Central Ltd; 2008 Jan 30;8(1):70. 208. Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15(5):615–25. 209. Bond-Smith D, Stone J. Methodological Challenges and Updated Findings from a Meta-analysis of the Association between Mammographic Density and Breast Cancer. Cancer Epidemiol Biomarkers Prev. 2018 Sep 11; [Epub ahead of print] 210. Kelemen LE, Sellers TA, Vachon CM. Can genes for mammographic density inform cancer aetiology? Vol. 8, Nature Reviews Cancer. 2008. p. 812–23. 211. Sherratt MJ, McConnell JC, Streuli CH. Raised mammographic density: causative mechanisms and biological consequences. Breast Cancer Res. BioMed Central; 2016 Dec 3;18(1):45. 135  212. Pankow JS, Vachon CM, Kuni CC, King RA, Arnett DK, Grabrick DM, et al. Genetic Analysis of Mammographic Breast Density in Adult Women: Evidence of a Gene Effect. JNCI J Natl Cancer Inst. Narnia; 1997 Apr 16;89(8):549–56. 213. Stone J, Gurrin LC, Byrnes GB, Schroen CJ, Treloar SA, Padilla EJD, et al. Mammographic density and candidate gene variants: A twins and sisters study. Cancer Epidemiol Biomarkers Prev. 2007;16(7):1479–84. 214. Brand JS, Humphreys K, Thompson DJ, Li J, Eriksson M, Hall P, et al. Volumetric Mammographic Density: Heritability and Association With Breast Cancer Susceptibility Loci. JNCI J Natl Cancer Inst. Oxford University Press; 2014 Nov 5;106(12):dju334-dju334. 215. Newman B, Austin M a, Lee M, King MC. Inheritance of human breast cancer: evidence for autosomal dominant transmission in high-risk families. Proc Natl Acad Sci U S A. 1988;85(9):3044–8. 216. Peto J, Mack TM. High constant incidence in twins and other relatives of women with breast cancer. Nat Genet. 2000;26(4):411–4. 217. Antoniou AC, Easton DF. Models of genetic susceptibility to breast cancer. Vol. 25, Oncogene. 2006. p. 5898–905. 218. Eccles SA, Aboagye EO, Ali S, Anderson AS, Armes J, Berditchevski F, et al. Critical research gaps and translational priorities for the successful prevention and treatment of breast cancer. Breast Cancer Res. 2013;15(5). 219. Woolcott C, Maskarinec G, Haiman C, Verheus M, Pagano I, Le Marchand L, et al. Association between breast cancer susceptibility loci and mammographic density: the Multiethnic Cohort. Breast Cancer Res. 2009;11(1):R10. 136  220. Odefrey F, Stone J, Gurrin LC, Byrnes GB, Apicella C, Dite GS, et al. Common genetic variants associated with breast cancer and mammographic density measures that predict disease. Cancer Res. 2010;70(4):1449–58. 221. Varghese JS, Thompson DJ, Michailidou K, Lindström S, Turnbull C, Brown J, et al. Mammographic breast density and breast cancer: Evidence of a shared genetic basis. Cancer Res. 2012;72(6):1478–84. 222. Fernandez-Navarro P, Pita G, Santamariña C, Moreno MP, Vidal C, Miranda-García J, et al. Association analysis between breast cancer genetic variants and mammographic density in a large population-based study (Determinants of Density in Mammographies in Spain) identifies susceptibility loci in TOX3 gene. Eur J Cancer. 2013;49(2):474–81. 223. Keller BM, McCarthy AM ari., Chen J, Armstrong K, Conant EF, Domchek SM, et al. Associations between breast density and a panel of single nucleotide polymorphisms linked to breast cancer risk: a cohort study with digital mammography. BMC Cancer. 2015;15:143. 224. Stone J, Thompson DJ, Dos Santos Silva I, Scott C, Tamimi RM, Lindstrom S, et al. Novel associations between common breast cancer susceptibility variants and risk-predicting mammographic density measures. Cancer Res. 2015;75(12):2457–67. 225. Haars G, Van Noord PAH, Van Gils CH, Grobbee DE, Peeters PHM. Measurements of breast density: No ratio for a ratio. Cancer Epidemiol Biomarkers Prev. 2005;14(11 I):2634–40. 226. Pinto Pereira SM, McCormack VA, Hipwell JH, Record C, Wilkinson LS, Moss SM, et al. Localized fibroglandular tissue as a predictor of future tumor location within the breast. Cancer Epidemiol Biomarkers Prev. 2011;20(8):1718–25. 137  227. Easton DF, Pooley KA, Dunning AM, Pharoah PDP, Thompson D, Ballinger DG, et al. Genome-wide association study identifies novel breast cancer susceptibility loci. Nature. 2007;447(7148):1087–93. 228. Turnbull C, Ahmed S, Morrison J, Pernet D, Renwick A, Maranian M, et al. Genome-wide association study identifies five new breast cancer susceptibility loci. Nat Genet. 2010;42(6):504–7. 229. Michailidou K, Hall P, Gonzalez-Neira A, Ghoussaini M, Dennis J, Milne RL, et al. Large-scale genotyping identifies 41 new loci associated with breast cancer risk. Nat Genet. 2013;45(4):353–61. 230. French JD, Ghoussaini M, Edwards SL, Meyer KB, Michailidou K, Ahmed S, et al. Functional variants at the 11q13 risk locus for breast cancer regulate cyclin D1 expression through long-range enhancers. Am J Hum Genet. 2013;92(4):489–503. 231. Michailidou K, Lindstrom S, Dennis J, Beesley J, Hui S, Kar S, et al. Association analysis identifies 65 new breast cancer risk loci. Nature. 2017;551(7678):92–4. 232. Milne RL, Kuchenbaecker KB, Michailidou K, Beesley J, Kar S, Lindström S, et al. Identification of ten variants associated with risk of estrogen-receptor-negative breast cancer. Nat Genet. 2017;49(12):1767–78. 233. Mavaddat N, Michailidou K, Dennis J, Lush M, Fachal L, Lee A, et al. Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. Am J Hum Genet. 2018 Dec 5;[Epub ahead of print]. 234. Burton H, Chowdhury S, Dent T, Hall A, Pashayan N, Pharoah P. Public health implications from COGS and potential for risk stratification and screening. Vol. 45, Nature Genetics. 2013. p. 349–51. 138  235. Mavaddat N, Pharoah PDP, Michailidou K, Tyrer J, Brook MN, Bolla MK, et al. Prediction of breast cancer risk based on profiling with common genetic variants. J Natl Cancer Inst. 2015;107(5):1–15. 236. Vachon CM, Sellers TA, Carlson EE, Cunningham JM, Hilker CA, Smalley RL, et al. Strong evidence of a genetic determinant for mammographic density, a major risk factor for breast cancer. Cancer Res. American Association for Cancer Research; 2007 Sep 1;67(17):8412–8. 237. Tamimi RM, Cox D, Kraft P, Colditz GA, Hankinson SE, Hunter DJ. Breast cancer susceptibility loci and mammographic density. Breast Cancer Res. BioMed Central; 2008;10(4):R66. 238. Amos CI, Dennis J, Wang Z, Byun J, Schumacher FR, Gayther SA, et al. The oncoarray consortium: A network for understanding the genetic architecture of common cancers. Cancer Epidemiol Biomarkers Prev. 2017;26(1):126–35. 239. BCAC – The Breast Cancer Association Consortium.Centre for Cancer Genetic Epidemiology, University of Cambridge. (http://bcac.ccge.medschl.cam.ac.uk). Accessed July 2018. 240. Brand JS, Humphreys K, Li J, Karlsson R, Hall P, Czene K. Common genetic variation and novel loci associated with volumetric mammographic density. Breast Cancer Res. BioMed Central; 2018 Dec 17;20(1):30. 241.  Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behav Res. 2011 May 31;46(3):399–424. 242.  Glass TA, Goodman SN, Hernán MA, Samet JM. Causal inference in public health. Annu Rev Public Health. 2013/01/07. 2013;34:61–75. 139  243.  Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41–55. 244. Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York, NY: John Wiley & Sons, Inc.; 1987 245.  Graham JW. Missing Data Analysis: Making It Work in the Real World. Annu Rev Psychol. 2009 Jan;60(1):549–76. 246. Kenward MG, Carpenter J. Multiple imputation: current perspectives. Stat Methods Med Res. 2007 Jun 2;16(3):199–218. 247. Van Buuren S, Brand JPL, Groothuis-Oudshoorn CGM, Rubin DB. Fully conditional specification in multivariate imputation. J Stat Comput Simul. Taylor & Francis ; 2006 Dec;76(12):1049–64. 248.  McCleary L. Using multiple imputation for analysis of incomplete data in clinical research. Nurs Res. 2002;51(5):339–43. 249.  Keller BM, Nathan DL, Gavenonis SC, Chen J, Conant EF, Kontos D. Reader variability in breast density estimation from full-field digital mammograms: the effect of image postprocessing on relative and absolute measures. Acad Radiol. NIH Public Access; 2013 May;20(5):560–8. 250.  Sprague BL, Conant EF, Onega T, Garcia MP, Beaber EF, Herschorn SD, et al. Variation in Mammographic Breast Density Assessments Among Radiologists in Clinical Practice: A Multicenter Observational Study. Ann Intern Med. NIH Public Access; 2016 Oct 4;165(7):457–64. 140  251. Abdolell M, Tsuruda KM, Brown P, Caines JS, Iles SE. Breast density scales: the metric matters. Br J Radiol. British Institute of Radiology; 2017 Oct;90(1078):20170307. 252. Bond-Smith D, Stone J. Methodological Challenges and Updated Findings from a Meta-analysis of the Association between Mammographic Density and Breast Cancer. Cancer Epidemiol Biomarkers Prev. 2018 Sep 11; [Epub ahead of print] 253. Zhao LP, Kolonel LN: Efficiency loss from categorizing quantitative exposures into qualitative exposures in case-control studies. Am J Epidemiol 1992;136:464–474 254. Vansteelandt S, Bekaert M, Claeskens G: On model selection and model misspecification in causal inference. Stat Methods Med Res. 2012, 21: 7-30. 255. Brentnall AR, Harkness EF, Astley SM, Donnelly LS, Stavrinos P, Sampson S, et al. Mammographic density adds accuracy to both the Tyrer-Cuzick and Gail breast cancer risk models in a prospective UK screening cohort. Breast Cancer Res. BioMed Central; 2015 Dec 1;17(1):147. 256. BC Cancer. Breast Density Provider Guidance. Version: October 2018. Available at http://www.bccancer.bc.ca/screening/Documents/Breast_20181004_BreastDensityProviderGuidance.pdf. Retrieved on April 5th, 2019. 257. US Food and Drug Administration. Mammography Quality Standards Act update proposal (March 28th, 2019). Available at https://s3.amazonaws.com/public-inspection.federalregister.gov/2019-05803.pdf. Retrieved on April 5th, 2019. 141  258. WarwickJ, BirkeH, StoneJ,et al.Mammographic breast density refines Tyrer-Cuzick estimates of breast cancer risk in high-risk women.Breast Cancer Res. 2014;16(5):451. 259. BrentnallAR, HarknessEF, AstleySM,et al.Mammographic density adds accuracy to both the Tyrer-Cuzick and Gail breast cancer risk models in a prospective UK screening cohort.Breast Cancer Res. 2015;17(1):147. 260. Brentnall AR, Cuzick J, Buist DSM, Bowles EJA. Long-term Accuracy of Breast Cancer Risk Assessment Combining Classic Risk Factors and Breast Density. JAMA Oncol. 2018;4(9):e180174. 261. Atkinson C, Warren R, Bingham SA, Day NE. Mammographic patterns as a predictive biomarker of breast cancer risk: effect of tamoxifen. Cancer Epidemiol Biomarkers Prev. 1999;8:863–6. 262. Son HJ, Oh KK. Significance of follow-up mammography in estimating the effect of tamoxifen in breast cancer patients who have undergone surgery. AJR Am J Roentgenol. 1999;173:905–9. 263. Brisson J, Brisson B, Cote G, Maunsell E, Berube S, Robert J. Tamoxifen and mammographic breast densities. Cancer Epidemiol Biomarkers Prev. 2000;9:911–5  264. Cuzick J, Warwick J, Pinney E, Warren RM, Duffy SW. Tamoxifen and breast density in women at increased risk of breast cancer. J Natl Cancer Inst. 2004;96:621–8. 265. Cuzick J, Warwick J, Pinney E, Duffy SW, Cawthorn S, Howell A, et al. Tamoxifen-induced reduction in mammographic density and breast cancer risk reduction: a nested case-control study. J Natl Cancer Inst 2011;103:744–52 142  266. Mullooly M, Pfeiffer RM, Nyante SJ, et al. . Mammographic density as a biosensor of tamoxifen effectiveness in adjuvant endocrine treatment of breast cancer: opportunities and implications. J Clin Oncol . 2016;34(18):2093–2097. 267. Shawky MS, Martin H, Hugo HJ, et al. . Mammographic density: a potential monitoring biomarker for adjuvant and preventative breast cancer endocrine therapies. Oncotarget . 2017;8(3):5578–5591. 268. Eriksson L, He W, Eriksson K, Humphreys K, Bergh J, Hall P, Czene K. Adjuvant Therapy and Mammographic Density Changes in Women With Breast Cancer. JNCI Cancer Spectr. 2018;00(0):pky071. 269. Masala G, Assedi M, Sera F, Ermini I, Occhini D, Castaldo M, et al. Can Dietary and Physical Activity Modifications Reduce Breast Density in Postmenopausal Women? The DAMA Study, a Randomized Intervention Trial in Italy. Cancer Epidemiol Biomarkers &amp;amp; Prev. 2019 Jan 1;28(1):41 LP-50. 270.  Astley SM, Harkness EF, Sergeant JC, Warwick J, Stavrinos P, Warren R, et al. A comparison of five methods of measuring mammographic density: a case-control study. Breast Cancer Res. BioMed Central; 2018;20(1):10. 271. Armstrong K, Eisen A, Weber B. Assessing the Risk of Breast Cancer. N Engl J Med. 2000 Feb 24;342(8):564–71. 272. Wang Y, Beydoun MA. The Obesity Epidemic in the United States Gender, Age, Socioeconomic, Racial/Ethnic, and Geographic Characteristics: A Systematic Review and Meta-Regression Analysis. Epidemiol Rev. 2007 May 2;29(1):6–28. 273. Jackson AS, Stanforth PR, Gagnon J, Rankinen T, Leon AS, Rao DC, et al. The effect of sex, age and race on estimating percentage body fat from body mass index: The Heritage Family Study. Int J Obes. 2002 May 30;26(6):789–96. 143  274. Heo M, Faith MS, Pietrobelli A, Heymsfield SB. Percentage of body fat cutoffs by sex, age, and race-ethnicity in the US adult population from NHANES 1999-2004. Am J Clin Nutr. 2012 Mar 1;95(3):594–602. 275. Li CI, Malone KE, Daling JR, Potter JD, Bernstein L, Marchbanks PA, et al. Timing of Menarche and First Full-Term Birth in Relation to Breast Cancer Risk. Am J Epidemiol. 2007 Oct 17;167(2):230–9. 276. MacMahon B, Cole P, Lin TM, Lowe CR, Mirra AP, Ravnihar B, et al. Age at first birth and breast cancer risk. Bull World Health Organ. 1970;43(2):209–21. 277. Trapido EJ. Age at first birth, parity, and breast cancer risk. Cancer. 1983 Mar 1;51(5):946–8. 278. El-Bastawissi AY, White E, Mandelson MT, Taplin SH. Reproductive and hormonal factors associated with mammographic breast density by age (United States). Cancer Causes Control. 2000 Dec;11(10):955–63. 279. Bergkvist L, Tabàr L, Bergström R, Adami HO. Epidemiologic determinants of the mammographic parenchymal pattern. A population-based study within a mammographic screening program. Am J Epidemiol. 1987;126(6):1075–81. 280. Bernstein L, Ross RK. Endogenous hormones and breast cancer risk. Epidemiol Rev. 1993;15(1):48–65. 281. Henson DE, Tarone RE. Involution and the etiology of breast cancer. Cancer. 1994 Jul 1;74(1 Suppl):424–9. 282. Kelemen LE, Couch FJ, Ahmed S, Dunning AM, Pharoah PD, Easton DF, et al. Genetic variation in stromal proteins decorin and lumican with breast cancer: investigations in two case--control studies. Breast Cancer Res. 2008;10. 144  283. Ghosh K, Hartmann LC, Reynolds C, Visscher DW, Brandt KR, Vierkant RA, et al. Association Between Mammographic Density and Age-Related Lobular Involution of the Breast. J Clin Oncol. 2010 May 1;28(13):2207–12. 284. Hamajima N, Hirose K, Tajima K, Rohan T, Calle EE, Heath CW, et al. Alcohol, tobacco and breast cancer – collaborative reanalysis of individual data from 53 epidemiological studies, including 58 515 women with breast cancer and 95 067 women without the disease. Br J Cancer. 2002 Nov 18;87(11):1234–45. 285. Key TJ, Appleby PN, Reeves GK, Roddam AW, Helzlsouer KJ, Alberg AJ, et al. Circulating sex hormones and breast cancer risk factors in postmenopausal women: reanalysis of 13 studies. Br J Cancer. 2011 Aug 23;105(5):709–22. 286. Maskarinec G, Takata Y, Pagano I, Lurie G, Wilkens LR, Kolonel LN. Alcohol consumption and mammographic density in a multiethnic population. Int J Cancer. 2006 May 15;118(10):2579–83. 287. Anderson SE, Dallal GE, Must A. Relative weight and race influence average age at menarche: results from two nationally representative surveys of US girls studied 25 years apart. Pediatrics. 2003 Apr;111(4 Pt 1):844–50. 288. World Cancer Research Fund/American Institute for Cancer Research. Food, nutrition, physical activity, and the prevention of cancer: a global perspective. Washington DC: AICR; 2007.  289. World Cancer Research Fund/American Institute for Cancer Research. The associations between food, nutrition and physical activity and the risk of breast cancer WCRF/AICR systematic literature review, continuous update report. Washington DC: AICR; 2008. 145  290. Lipworth L, Bailey LR, Trichopoulos D. History of breast-feeding in relation to breast cancer risk: a review of the epidemiologic literature. J Natl Cancer Inst. 2000 Feb 16;92(4):302–12. 291. Lope V, Pérez-Gómez B, Sánchez-Contador C, Santamariña MC, Moreo P, Vidal C, et al. Obstetric history and mammographic density: a population-based cross-sectional study in Spain (DDM-Spain). Breast Cancer Res Treat. 2012 Apr 4;132(3):1137–46. 292. Sung J, Song Y-M, Stone J, Lee K, Lee D. Reproductive factors associated with mammographic density: a Korean co-twin control study. Breast Cancer Res Treat. 2011 Jul 26;128(2):567–72. 293. Braithwaite D, Moore DH, Lustig RH, Epel ES, Ong KK, Rehkopf DH, et al. Socioeconomic status in relation to early menarche among black and white girls. Cancer Causes Control. 2009 Jul 25;20(5):713–20. 294. Rindfuss RR, Morgan SP, Offutt K. Education and the changing age pattern of American fertility: 1963-1989. Demography. 1996 Aug;33(3):277–90. 295. Anthony JC, Warner LA, Kessler RC. Comparative epidemiology of dependence on tobacco, alcohol, controlled substances, and inhalants: Basic findings from the National Comorbidity Survey. Exp Clin Psychopharmacol. 1994;2(3):244–68. 296. Gilman SE, Breslau J, Conron KJ, Koenen KC, Subramanian S V, Zaslavsky AM. Education and race-ethnicity differences in the lifetime risk of alcohol dependence. J Epidemiol Community Heal. 2008 Mar 1;62(3):224–30. 297. Dettwyler KA. When to wean: biological versus cultural perspectives. Clin Obstet Gynecol. 2004 Sep;47(3):712–23. 146  298. Wang Y, Beydoun MA. The obesity epidemic in the United States - Gender, age, socioeconomic, racial/ethnic, and geographic characteristics: A systematic review and meta-regression analysis. Vol. 29, Epidemiologic Reviews. 2007. p. 6–28. 299. Marks NF, Shinberg DS. Socioeconomic status differences in hormone therapy. Am J Epidemiol. 1998 Sep 15;148(6):581–93. 300. Mosher WD, Martinez GM, Chandra A, Abma JC, Willson SJ. Use of contraception and use of family planning services in the United States: 1982-2002. Adv Data. 2004 Dec 10;(350):1–36. 301. Foster DG, Bley J, Mikanda J, Induni M, Arons A, Baumrind N, et al. Contraceptive use and risk of unintended pregnancy in California. Contraception. 2004 Jul;70(1):31–9. 302. Krings KM, Matteson KA, Allsworth JE, Mathias E, Peipert JF. Contraceptive choice: how do oral contraceptive users differ from condom users and women who use no contraception? Am J Obstet Gynecol. NIH Public Access; 2008 May;198(5):e46-7. 303. Ham SA, Ainsworth BE. Disparities in Data on Healthy People 2010 Physical Activity Objectives Collected by Accelerometry and Self-Report. Am J Public Health. 2010 Apr 1;100(S1):S263–8. 304. Aitken Z, Walker K, Stegeman BH, Wark PA, Moss SM, McCormack VA, et al. Mammographic density and markers of socioeconomic status: a cross-sectional study. BMC Cancer. BioMed Central; 2010 Feb 9;10:35. 305. Kravdal Ø. Education and fertility in sub-Saharan Africa: individual and community effects. Demography. 2002 May;39(2):233–50. 147  306. Graff M, Yount KM, Ramakrishnan U, Martorell R, Stein AD. Childhood nutrition and later fertility: pathways through education and pre-pregnant nutritional status. Demography. 2010 Feb;47(1):125–44. 307. Wagenknecht LE, Perkins LL, Cutter GR, Sidney S, Burke GL, Manolio TA, et al. Cigarette smoking behavior is strongly related to educational status: the CARDIA study. Prev Med (Baltim). 1990 Mar;19(2):158–69. 308. Zhu BP, Giovino GA, Mowery PD, Eriksen MP. The relationship between cigarette smoking and education revisited: implications for categorizing persons’ educational status. Am J Public Health. American Public Health Association; 1996 Nov;86(11):1582–9. 309. Escobedo LG, Anda RF, Smith PF, Remington PL, Mast EE. Sociodemographic characteristics of cigarette smoking initiation in the United States. Implications for smoking prevention policy. JAMA. 1990 Sep 26;264(12):1550–5. 310. Aronson KJ, Miller AB, Woolcott CG, Sterns EE, McCready DR, Lickley LA, et al. Breast adipose tissue concentrations of polychlorinated biphenyls and other organochlorines and breast cancer risk. Cancer Epidemiol Biomarkers Prev. 2000 Jan;9(1):55–63. 311. Brody JG, Moysich KB, Humblet O, Attfield KR, Beehler GP, Rudel RA. Environmental pollutants and breast cancer. Cancer. 2007 Jun 15;109(S12):2667–711. 312. Nagata C, Nagao Y, Nakamura K, Wada K, Tamai Y, Tsuji M, et al. Cadmium exposure and the risk of breast cancer in Japanese women. Breast Cancer Res Treat. 2013 Feb;138(1):235–9. 148  313. Apter D, Reinilä M, Vihko R. Some endocrine characteristics of early menarche, a risk factor for breast cancer, are preserved into adulthood. Int J cancer. 1989 Nov 15;44(5):783–7. 314. Williams DR, Mohammed SA, Leavell J, Collins C. Race, socioeconomic status, and health: Complexities, ongoing challenges, and research opportunities. Ann N Y Acad Sci. 2010 Feb;1186(1):69–101. 315. Windham GC, Pinney SM, Sjodin A, Lum R, Jones RS, Needham LL, et al. Body burdens of brominated flame retardants and other persistent organo-halogenated compounds and their descriptors in US girls. Environ Res. 2010 Apr;110(3):251–7. 316. Setiawan VW, Haiman CA, Stanczyk FZ, Le Marchand L, Henderson BE. Racial/Ethnic Differences in Postmenopausal Endogenous Hormones: The Multiethnic Cohort Study. Cancer Epidemiol Biomarkers Prev. 2006 Oct 1;15(10):1849–55. 317. Raskin L, Pinchev M, Arad C, Lejbkowicz F, Tamir A, Rennert HS, et al. FGFR2 is a breast cancer susceptibility gene in Jewish and Arab Israeli populations. Cancer Epidemiol Biomarkers Prev. 2008 May 7;17(5):1060–5. 318. Rebbeck TR, DeMichele A, Tran T V., Panossian S, Bunin GR, Troxel AB, et al. Hormone-dependent effects of FGFR2 and MAP3K1 in breast cancer susceptibility in a population-based sample of post-menopausal African-American and European-American women. Carcinogenesis. 2008 Oct 8;30(2):269–74. 319. Park HL, Ziogas A, Chang J, Desai B, Bessonova L, Garner C, et al. Novel polymorphisms in caspase-8 are associated with breast cancer risk in the California Teachers Study. BMC Cancer. 2016 Jan 12;16(1):14. 149  320. Handa VL, Landerman R, Hanlon JT, Harris T, Cohen HJ. Do older women use estrogen replacement? Data from the Duke Established Populations for Epidemiologic Studies of the Elderly (EPESE). J Am Geriatr Soc. 1996;44(1):1–6. 321. Brown AF, Pérez-Stable EJ, Whitaker EE, Posner SF, Alexander M, Gathe J, et al. Ethnic differences in hormone replacement prescribing patterns. J Gen Intern Med. 1999 Nov;14(11):663–9. 322. Habel LA, Capra AM, Oestreicher N, Greendale GA, Cauley JA, Bromberger J, et al. Mammographic density in a multiethnic cohort. Menopause. 2007 Sep;14(5):891–9. 323. McCormack VA, Perry N, Vinnicombe SJ, Silva IDS. Ethnic variations in mammographic density: A British multiethnic longitudinal study. Am J Epidemiol. 2008;168(4):412–21. 324. Mosher WD, Martinez GM, Chandra A, Abma JC, Willson SJ. Use of contraception and use of family planning services in the United States: 1982-2002. Adv Data. 2004 Dec 10;(350):1–36. 325. Frost JJ, Darroch JE. Factors Associated with Contraceptive Choice and Inconsistent Method Use, United States, 2004. Perspect Sex Reprod Health. 2008 Jun;40(2):94–104. 326. Wiebe E. Contraceptive practices and attitudes among immigrant and nonimmigrant women in Canada. Can Fam Physician. 2013 Oct;59(10):e451-5. 327. Aliyu MH, Salihu HM, Keith LG, Ehiri JE, Islam MA, Jolly PE. Trends in birth across high-parity groups by race/ethnicity and maternal age. J Natl Med Assoc. 2005 Jun;97(6):799–804. 150  328. Ward KD, Vander Weg MW, Kovach KW, Klesges RC, DeBon MW, Haddock CK, et al. Ethnic and gender differences in smoking and smoking cessation in a population of young adult air force recruits. Am J Health Promot. 2002;16(5):259–66. 329. Karlsen S, Millward D, Sandford A. Investigating ethnic differences in current cigarette smoking over time using the health surveys for England. Eur J Public Health. 2012 Apr 1;22(2):254–6. 330. Garcia-Closas M, Hall P, Nevanlinna H, Pooley K, Morrison J, Richesson DA, et al. Heterogeneity of breast cancer associations with five susceptibility loci by clinical and pathological characteristics. PLoS Genet. 2008;4(4). 331. Rebbeck TR, Demichele A, Tran T V., Panossian S, Bunin GR, Troxel AB, et al. Hormone-dependent effects of FGFR2 and MAP3K1 in breast cancer susceptibility in a population-based sample of post-menopausal African-American and European-American women. Carcinogenesis. 2009;30(2):269–74. 332. Zhang J, Qiu L-X, Wang Z-H, Leaw S-J, Wang B-Y, Wang J-L, et al. Current evidence on the relationship between three polymorphisms in the FGFR2 gene and breast cancer risk: a meta-analysis. Breast Cancer Res Treat. 2010;124(2):419–24. 333. Nelson HD, Zakher B, Cantor A, Fu R, Griffin J, O ’meara ES, et al. Risk factors for breast cancer for women aged 40 to 49 years: a systematic review and meta-analysis. Ann Intern Med. 2012;156(9):635–48. 334. PDQ® Cancer Genetics Editorial Board. PDQ Genetics of Breast and Gynecologic Cancers. Bethesda, MD: National Cancer Institute. Updated <11/30/2017>. Available at: https://www.cancer.gov/types/breast/hp/breast-ovarian-genetics-pdq. Accessed <12/01/2017> 151  335. Ursin G, Lillie EO, Lee E, Cockburn M, Schork NJ, Cozen W, et al. The Relative Importance of Genetics and Environment on Mammographic Density. Cancer Epidemiol Biomarkers Prev. 2009 Jan 1;18(1):102–12. 336. Martin LJ, Melnichouk O, Guo H, Chiarelli AM, Hislop TG, Yaffe MJ, et al. Family History, Mammographic Density, and Risk of Breast Cancer. Cancer Epidemiol Biomarkers Prev. 2010 Feb 1;19(2):456–63. 337. Colditz GA, Hankinson SE, Hunter DJ, Willett WC, Manson JE, Stampfer MJ, et al. The Use of Estrogens and Progestins and the Risk of Breast Cancer in Postmenopausal Women. N Engl J Med. 1995 Jun 15;332(24):1589–93. 338. Rossouw JE, Anderson GL, Prentice RL, LaCroix AZ, Kooperberg C, Stefanick ML, et al. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results From the Women’s Health Initiative randomized controlled trial. JAMA. 2002 Jul 17;288(3):321–33. 339. Rutter CM, Mandelson MT, Laya MB, Seger DJ, Taplin S. Changes in breast density associated with initiation, discontinuation, and continuing use of hormone replacement therapy. JAMA. 2001 Jan 10;285(2):171–6. 340. McTiernan A, Martin CF, Peck JD, Aragaki AK, Chlebowski RT, Pisano ED, et al. Estrogen-Plus-Progestin Use and Mammographic Density in Postmenopausal Women: Women’s Health Initiative Randomized Trial. JNCI J Natl Cancer Inst. 2005 Sep 21;97(18):1366–76. 341. Vachon CM, Kuni CC, Anderson K, Anderson VE, Sellers TA. Association of mammographically defined percent breast density with epidemiologic risk factors for breast cancer (United States). Cancer Causes Control. 2000 Aug;11(7):653–62. 152  342. Verheus M, Peeters PHM, Kaaks R, van Noord PAH, Grobbee DE, van Gils CH. Premenopausal Insulin-Like Growth Factor-I Serum Levels and Changes in Breast Density over Menopause. Cancer Epidemiol Biomarkers Prev. 2007 Mar 1;16(3):451–7. 343. Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormonal contraceptives: collaborative reanalysis of individual data on 53 297 women with breast cancer and 100 239 women without breast cancer from 54 epidemiological studies. Lancet. 1996;347(9017):1713–27. 344. Canadian consensus on hormone replacement therapy. Estrogen and progestin use in postmenopausal women. Can Fam Physician. 2003 Feb;49(2):188–91. 345. Willett WC, Tamimi RM, Hankinson SE, Hunter DJ, Colditz GA. Chapter 20: Nongenetic Factors in the Causation of Breast Cancer, in Harris JR, Lippman ME, Morrow M, Osborne CK. Diseases of the Breast, 4th edition, Lippincott Williams & Wilkins, 2010. 346. Li T, Sun L, Miller N, Nicklee T, Woo J, Hulse-Smith L, et al. The association of measured breast tissue characteristics with mammographic density and other risk factors for breast cancer. Cancer Epidemiol Biomarkers Prev. 2005;14(2):343–9. 347. Dite GS, Gurrin LC, Byrnes GB, Stone J, Gunasekara A, McCredie MRE, et al. Predictors of Mammographic Density: Insights Gained from a Novel Regression Analysis of a Twin Study. Cancer Epidemiol Biomarkers Prev. 2008 Dec 1;17(12):3474–81.. 348. Kaufman Z, Garstin WI, Hayes R, Michell MJ, Baum M. The mammographic parenchymal patterns of nulliparous women and women with a family history of breast cancer. Clin Radiol. 1991 Jun;43(6):385–8. 153  349. Grove JS, Goodman MJ, Gilbert FI, Mi MP. Factors associated with mammographic pattern. Br J Radiol. 1985;58(685):21–5. 350. de Stavola BL, Gravelle IH, Wang DY, Allen DS, Bulbrook RD, Fentiman IS, et al. Relationship of mammographic parenchymal patterns with breast cancer risk factors and risk of breast cancer in a prospective study. Int J Epidemiol. 1990 Jun;19(2):247–54. 351. Weinsier RL, Hunter GR, Heini AF, Goran MI, Sell SM. The etiology of obesity: relative contribution of metabolic factors, diet, and physical activity. Am J Med. 1998 Aug;105(2):145–50. 352. Eisenmann JC, Bartee RT, Wang MQ. Physical Activity, TV Viewing, and Weight in U.S. Youth: 1999 Youth Risk Behavior Survey. Obes Res. 2002 May;10(5):379–85. 353. Ladabaum U, Mannalithara A, Myer PA, Singh G. Obesity, abdominal obesity, physical activity, and caloric intake in US adults: 1988 to 2010. Am J Med. 2014 Aug;127(8):717–727.e12. 354. Hiatt RA, Fireman BH. Smoking, menopause, and breast cancer. J Natl Cancer Inst. 1986 May;76(5):833–8. 355. Ambrosone CB, Kropp S, Yang J, Yao S, Shields PG, Chang-Claude J. Cigarette Smoking, N-Acetyltransferase 2 Genotypes, and Breast Cancer Risk: Pooled Analysis and Meta-analysis. Cancer Epidemiol Biomarkers Prev. 2008 Jan 9;17(1):15–26. 356. Gaudet MM, Carter BD, Brinton LA, Falk RT, Gram IT, Luo J, et al. Pooled analysis of active cigarette smoking and invasive breast cancer risk in 14 cohort studies. Int J Epidemiol. 2016;(September):dyw288. 154  357. Windham GC, Mitchell P, Anderson M, Lasley BL. Cigarette smoking and effects on hormone function in premenopausal women. Environ Health Perspect. 2005 Oct;113(10):1285–90. 358. Butler LM, Gold EB, Conroy SM, Crandall CJ, Greendale GA, Oestreicher N, et al. Active, but not passive cigarette smoking was inversely associated with mammographic density. Cancer Causes Control. 2010 Feb 14;21(2):301–11.  155  Appendices Appendix A:  Causal relationships between mammographic density parameters and breast cancer This section summarizes the evidence supporting the causal relationships rendered in the DAG model utilized in chapter 2 (figure 2-1), stating nodes and their relationships as follows:  Age ← Sampling → Breast cancer. The data comes from a case-control study; participants were selected based on their age and breast cancer status (this node represents the study's sampling scheme)144.  Age → Breast cancer. Older age is a strong risk factor for breast cancer (Armstrong 2000)271.  Age → BMI. The authors conclude that “the associations of obesity with gender, age, ethnicity, and socioeconomic status are complex and dynamic” (Wang 2007)245. The results published in the literature show that BMI is not independent of age and gender (Jackson 2002273, Heo 2012274)  Age at menarche → Breast cancer. Evidence indicates that late age at menarche 156  decreases the risk of breast cancer (Armstrong 2000271, Li 2007275).  Age at first full-term birth → Breast cancer. Reduced risk of breast cancer was observed in women having their first birth at an early age (MacMahon 1970)276. Women who had their first birth at an older age were found to have increased incidence rates of breast cancer; the risk of breast cancer fell with increasing parity (Trapido 1983)277.  Age at first full-term birth → Mammographic density (both: dense and non-dense areas). Nulliparity is associated with high density. Older age at first birth has been associated with higher density; the association has been found to be stronger in women older than 55 than among younger women (El-Bastawissi 2000)278. Older age at first birth has been associated to greater odds of having a high-risk Tabár pattern (Bergkvist 1987)279.  Age at first full-term birth → Endogenous hormones. Lower age at first birth and higher parity is related to decreased levels of endogenous hormones (Bernstein 1993)280.   Age → Mammographic density (both: dense and non-dense areas). Mammographic density is known to decrease with age; this is the result of the well-known physiological process of lobular involution: with age, breast 157  epithelium and stroma are progressively replaced by adipose tissue (Henson 1994)281.  On average, premenopausal women have higher mammographic density as compared with postmenopausal women (Kelemen 2008)282. Evidence of an inverse association between age-related lobular involution of the breast and mammographic density has been reported: when assessed by components, a strong positive association of involution with non-dense area was observed; no association was seen between involution and dense area (Ghosh 2010)283.  Alcohol → Breast cancer. Pooled analysis of data (53 studies) concludes that the relative risk of breast cancer is increased by about 7% for each alcoholic drink consumed per day (Hamajima 2002)284.  Alcohol → Endogenous hormones. Women who drink alcohol have estrogen levels higher than in abstainers (cross-sectional analyses of 13 prospective studies; Key 2011)285.  Alcohol → Mammographic density (both: dense and non-dense areas). Percent density was found to be greater in alcohol consumers than in abstainers (Maskarinec 2006)286.   BMI → Age at menarche. Findings suggest higher BMI is related to earlier age at menarche (Anderson 2003)287. 158   BMI → Breast cancer. The 2007 American Institute for Cancer Research (AICR) expert panel declared that “the evidence that greater body fatness is a cause of postmenopausal breast cancer is convincing” 288. The conclusion of the AICR 2008 systematic review supports the concept of lower BMI associated to decreased breast cancer incidence289.  Breastfeeding → Breast cancer. The evidence suggests that prolonged duration of breastfeeding is associated with lower incidence of breast cancer (Lipworth 2000)290.  Breastfeeding → Mammographic density (dense area). Higher mammographic density (percent dense area) has been found to be more prevalent in women who have breastfed for longer periods (Lope 2012)291. Absolute dense area has been positively associated with the duration of lactation (Sung 2011)292.   Education (SES) → Age at menarche. The timing of menarche differs according to race and SES (Braithwaite 2009)293.  Education → Age at first full term birth. Lower education is associated to earlier first full term pregnancy (Rindfuss 1996)294.  159   Education → Alcohol. Lower education increases the likelihood of alcohol dependence (Anthony 1994)295. Lower SES is related to higher risk for alcohol dependence; the magnitude of this association is modified by ethnicity (Gilman 2008)296.   Education → Breastfeeding. Length of breastfeeding is associated to ethnicity and SES/education (Dettwyler 2004)297.  Education → BMI. The authors conclude that low SES highly increases the likelihood of obesity (Wang 2007)298.   Education → HRT. Positive association between hormone use and educational level (Marks 1998)299.   Education → Oral contraceptives (OC). OC use is more likely in women with higher levels of education (Mosher 2002300, Foster 2004301, Krings 2008302).  Education → Physical activity. Lower education was found associated with decreased physical activity (Ham 2010)303.  Education (SES) → Mammographic density (non-dense area). Strong negative 160  associations between these SES variables and lucent area, which were attenuated upon adjustment for body mass index (BMI); SES variables were not associated with dense area (Aitken 2010)304.  Education → Parity. Higher maternal educational levels have been observed to be related to reduction of fertility (Rindfuss 1996294, Kravdal 2002305, Graff 2010306).  Education → Smoking. A strong inverse relationship was found between education and smoking (Wagenknecht 1990)307. The likelihood of smoking decreases with education level (Zhu 1996)308. Smoking is associated to education and ethnicity (Escobedo 1990)309.  Endocrine disruptors → Breast cancer. Associations between PCBs measured in breast adipose tissue and breast cancer (Aronson 2000)310. The evidence supports the presence of gene–environment interactions leading to  breast cancer, specifically polycyclic aromatic hydrocarbons (PAHs) and polychlorinated biphenyls (PCBs) coupled with certain genetic polymorphisms related steroid hormone metabolism and activation of carcinogens (Brody 2007)311. Increasing levels of urine cadmium (which may have estrogenic activity in rats and epithelial breast line cells) were found related to related to greater odds for breast cancer (Nagata 2013)312. 161   Endogenous hormones → Age at menarche. The presence of higher levels of endogenous hormones (serum estradiol) is associated to earlier menarche (Apter 1989)313.   Endogenous hormones → Menopause. The presence of lower levels of endogenous hormones is associated to earlier menopause (Key 2011)285.  Endogenous hormones → BMI. Levels of estrogens were found to be higher in women with higher BMI; androstenedione and testosterone were also found to be increased with higher BMI (Key 2011)285.  Endogenous hormones → Breast cancer. Higher estrogen levels increase the incidence of breast cancer; women with higher levels of estradiol have twice the risk of breast cancer, compared to women with lower levels (Key 2011)285.  Ethnicity → Alcohol. Lower SES is related to higher risk for alcohol dependence; the magnitude of this association is modified by ethnicity (Gilman 2008)296.  Ethnicity → Age at menarche. The timing of menarche differs according to race and SES (Braithwaite 2009)293. 162   Ethnicity → BMI. There is evidence indicating that BMI varies according to ethnicity (Wang 2007)272.  Ethnicity → Breastfeeding. Length of breastfeeding is associated to ethnicity and SES/education (Dettwyler 2004)297.   Ethnicity → Education. An important contributor to racial disparities in health is the presence of differences in socioeconomic status across racial groups (Williams 2010)314.  Ethnicity → Endocrine disruptors. Differences in association between ethnic groups and exposure to hormonally active agents (such as polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), and organo-chlorinated pesticides (OCPs)) have been observed (Windham 2010)315.   Ethnicity → Endogenous hormones. Differences in endogenous postmenopausal hormonal milieu across five racial/ethnic groups were reported (Native Hawaiians, Japanese Americans, Whites, African Americans, and Latinas), consistent with the known differences in breast cancer incidence across the studied ethnic groups (Setiawan 2006)316. 163   Ethnicity → Family history/Genotype. A considerable fraction of breast cancer in Arab, Ashkenazi, and Sephardi Jewish populations may be explained by genetic variation in FGFR2, SNP rs1219648 (Raskin 2008)317. The evidence found by Rebbeck et al. (2008)318 confirm that FGFR2 and MAP3K1 participate in breast cancer susceptibility and confer their effects primarily in ER+ and PR+ tumors; also, a significant interaction between MAP3K1 rs889312 and FGFR2 rs2981582 (P = 0.022) was observed in African-American but not European-American women. Genotype CASP8 SNP, rs2293554, was found to be significantly associated with HER2-positive breast cancer risk in non-Hispanic white women (Park 2016)319.   Ethnicity → HRT. White women are more likely to be prescribed hormone replacement therapy (Handa 1996)320, (Brown 1999)321.  Ethnicity → Mammographic density (both: dense and non-dense areas). Mammographic density has been shown to vary by race; African Americans were found to have the highest fully adjusted mean mammographic density, while Japanese women had the lowest (Habel 2007)322. Age-adjusted percent mammographic densities in Afro-Caribbeans and South Asians were found to be lower than in Caucasians (McCormack 2008)323.   Ethnicity → Oral contraceptives (OC). White women are more likely to use oral 164  contraceptives than Black or Hispanic (Mosher 2004324, Frost 2008325). Also, there is evidence indicating that OC use is less likely in immigrant women in Canada (Wiebe 2013)326.  Ethnicity → Parity. It has been observed (U.S. setting) that after maternal age standardization, black and Hispanic women are more likely to have higher parity (defined as 5 or more births) as compared to whites (Aliyu 2005)327.   Ethnicity → Smoking. Differences in smoking patterns according to ethnic groups have been identified (Ward 2002328; Karlsen 2002329).  Smoking is associated to education and ethnicity (Escobedo 1990)309.  Family history/Genotype → Breast cancer. A considerable fraction of breast cancer in Arab, Ashkenazi, and Sephardi Jewish populations may be explained by genetic variation in FGFR2, SNP rs1219648 (Raskin 2008)317. There is evidence of common genetic variants (SNPs in five loci: FGFR2, TNRC9, MAP3K1, 8q24, and LSP1) influencing the pathological subtype of breast cancer, providing support for the hypothesis that ER-positive and ER-negative disease are biologically distinct (Garcia-Closas 2008)330. The evidence found by Rebbeck et al. (2008)331 confirm that FGFR2 and MAP3K1 participate in breast cancer susceptibility and confer their effects primarily in ER+ and PR+ tumors; also, a significant interaction between MAP3K1 rs889312 and FGFR2 rs2981582 was 165  observed in African-American but not European-American women. The 2010 meta-analysis by Zhang et al.332 suggests that FGFR2 polymorphisms are important determinants of breast cancer susceptibility. Genotype. CASP8 SNP, rs2293554, was found to be significantly associated with HER2-positive breast cancer risk in non-Hispanic white women (Park 2016)319. The risk of breast cancer was found to be increased by at least two times in women aged 40-49 years, with family history of breast cancer (first-degree relatives) and high mammographic density (Nelson 2012)333. The risk of breast cancer is increased in women with a first-degree relative with breast cancer, or with first-degree female relative with ovarian cancer (Pharoah 1997113, Collaborative Group on Hormonal Factors in Breast Cancer 2001114, PDQ® Cancer Genetics Editorial Board 2017334). Several inheritable gene mutations that increase breast cancer risk have been identified; the risk from mutations can vary greatly. A few greatly increase breast cancer risk (such as BRCA1/2, ATM, or PTEN); most others are responsible for modest risk increments. It is uncommon to find inherited mutations that are known to increase the risk of breast cancer in the general population; mutations account for 5 to 10% of breast cancers diagnosed in the U.S. (American Cancer Society 201564, PDQ® Cancer Genetics Editorial Board 2017334)  Family history/Genotype → Mammographic density (both: dense and non-dense areas). It has been observed that population variation in the percentage of dense 166  tissue on mammography at a given age has high heritability (Boyd 2002)122. It is likely that dense breasts found in families are the result of genetic factors (Stone 2006)123. There is evidence confirming that the genome is an important determinant of mammographic density (Ursin 2009)335. Specific genes that might be involved in mammographic density are currently being studied (Martin 2010)336. Associations between several specific SNPs and mammographic density have been observed (Stone 2015)224.  HRT (hormone replacement therapy) → Breast cancer. The evidence indicates that postmenopausal hormone use increases the risk of breast cancer (Colditz 1997337, Collaborative Group on Hormonal Factors in Breast Cancer 1997108, Ross 2000106, Rossouw 2002338).  HRT (hormone replacement therapy) → Mammographic density (both: dense and non-dense areas). Women who use postmenopausal hormones (estrogen plus progestin) have higher breast density than women who do not use these hormones; their breast density decreases once they stop using hormones (Rutter 2001339, McTiernan 2005340).  Mammographic density (dense area → percent dense area) → Breast cancer. The association between high mammographic density, expressed qualitatively, or quantitatively, both as percent mammographic density, as well as absolute dense 167  area, and elevated breast cancer risk is well-established (Byrne 199544, Maskarinec 200045, McCormack 200647, Tice 200846, Boyd 200748, Chiu 201049, Heusinger 201150, Yaghjyan 201151, Baglietto 201452, Bertrand 201553).  Mammographic density (non-dense area) → Mammographic density (percent dense area). This path represents one of the two components of percent dense area (mammary adipose tissue). [Percent dense area = (dense area / (dense area + non-dense area)) *100]  Mammographic density (dense area) → Mammographic density (percent dense area). Complementing the relationship described above, this path represents the other component of percent dense area (mammary glandular tissue plus stroma).  Menopause → Breast cancer. Younger age at menopause is associated with a modest decrease in breast cancer risk (Armstrong 2000)271.  Menopause → Mammographic density (both: dense and non-dense areas). Younger age at menopause is associated to a mild reduction in mammographic density (Vachon 2000)341. It has been reported that median dense area decreased after menopause; non-dense area slightly increased over the same process (Verheus 2007)342. 168   Oral contraceptives (OC) → Breast cancer. Current and recent use of OC has been associated to 20-30 percent higher risk of breast cancer than in never-users (Collaborative Group on Hormonal Factors in Breast Cancer 1996343, Gierish 2013102). A mild increment in the risk of breast cancer was found in women who are current or recent user of oral contraceptives (Nelson 2012)333. The risk is progressively diminished in women who stop OC; after 10 years the risk of breast cancer returns to the same levels as in never-users (Collaborative Group on Hormonal Factors in Breast Cancer 1996)343.  Menopause → HRT. In general, symptomatic postmenopausal women are likely to be offered/prescribed HRT (“The best treatment for distressing menopausal symptoms remains HRT”, Canadian consensus on hormone replacement therapy 2003)344.  Parity → Breast cancer. Women who had their first birth at an older age were found to have increased incidence rates of breast cancer; the risk of breast cancer fell with increasing parity (Trapido 1983)277. It has been seen that the more births a woman has, the lower her breast cancer risk is (Willet 2010)345.  Parity → Endogenous hormones. Lower age at first birth and higher parity is related to decreased levels of endogenous hormones (Bernstein 1993)280. 169   Parity → Mammographic density (both: dense and non-dense areas). Parity and number of births have been shown to be associated inversely only with percent collagen, a component of dense area (Li 2005)346. Several studies have seen a negative association between number of live births and dense area, as well as a positive association with non-dense area, and therefore, a negative association with percent dense area (Dite 2008)347. Similarly, it has been seen that nulliparity is associated to increased percent density (Vachon 2000)341 and high-risk Tabár patterns (Kaufman 1991)348. Greater parity has been seen to be inversely related to the likelihood of presenting high-risk Tabár patterns (Grove 1985349, Gram 199516, de Stavola 1990350).  Physical activity → Breast cancer. The expert panels (World Cancer Research Fund/American Institute for Cancer Research 2007288, 2008289) concluded that physical activity is associated to decreased breast cancer incidence.  Physical activity → BMI. Higher levels of physical activity are related to leaner BMI (Weinsier 1998351, Eisenmann 2002352). Trends in BMI and waist circumference were directly associated with physical activity level but not caloric intake (Ladabaum 2014353).  Smoking → Breast cancer. Moderate smoking has been associated with earlier menopause, as well as with reduction of breast cancer risk (Hiatt 1986354, 170  Ambrosone 2008355). However, also a modest direct association between early smoking initiation (before first birth and the risk of ER positive breast cancer has been seen (Gaudet 2016)356.  Smoking → Endogenous hormones. The evidence shows that tobacco use decreases the levels of endogenous hormones (possible anti-estrogenic effect; Windham 2005)357.   Smoking → Mammographic density (both: dense and non-dense areas). Active smoking was found to decrease mammographic density by 7.2%, compared to never-smokers. This evidence, reported by Butler et al.358, supports the hypothesis that smoking exerts an antiestrogenic effect on breast tissue (but antagonizes the accepted direct association between breast cancer risk and smoking prior to first full-term birth).  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0378180/manifest

Comment

Related Items