Back to Journals » Breast Cancer: Targets and Therapy » Volume 11

Mammographic density parameters and breast cancer tumor characteristics among postmenopausal women

Authors Velásquez García HA , Gotay CC, Wilson CM , Lohrisch CA , Lai AS, Aronson KJ, Spinelli JJ 

Received 30 October 2018

Accepted for publication 18 February 2019

Published 16 August 2019 Volume 2019:11 Pages 261—271

DOI https://doi.org/10.2147/BCTT.S192766

Checked for plagiarism Yes

Review by Single anonymous peer review

Peer reviewer comments 2

Editor who approved publication: Professor Pranela Rameshwar



Héctor A Velásquez García,1,2 Carolyn C Gotay,1 Christine M Wilson,3 Caroline A Lohrisch,4 Agnes S Lai,2 Kristan J Aronson,5 John J Spinelli1,2

1School of Population and Public Health, University of British Columbia, Vancouver, BC, Canada; 2Population Oncology, BC Cancer, Vancouver, BC, Canada; 3Screening Mammography Program, BC Cancer, Vancouver, BC, Canada; 4Department of Medical Oncology, BC Cancer, Vancouver, BC, Canada; 5Department of Public Health Sciences and Division of Cancer Care and Epidemiology, Cancer Research Institute, Queen’s University, Kingston, ON, Canada

Correspondence: Héctor A Velásquez García
Population Oncology, BC Cancer, 8 th floor, 686 West Broadway, Vancouver, BC V5Z 1G1, Canada
Tel +1604 877 6068
Fax +1 604 877 6212
Email [email protected]

Purpose: Mammographic density is an important breast cancer risk factor, although it is not clear whether the association differs across breast cancer tumor subtypes. We examined the association between indicators of mammographic density and breast cancer risk by tumor subtype among postmenopausal women by investigating heterogeneity across tumor characteristics.
Methods: Mammographic density measures were determined for 477 breast cancer cases and 588 controls, all postmenopausal, in Vancouver, British Columbia, using digitized screening mammograms and Cumulus software. Mammographic dense (DA), non-dense (NDA), and percent dense (PDA) areas were treated as continuous covariates and categorized into quartiles according to the distribution in controls. For cases only, tests for heterogeneity between tumor subtypes were assessed by multinomial logistic regression. Associations between mammographic density and breast cancer risk were modeled for each subtype separately through unconditional logistic regression.
Results: Heterogeneity was apparent for the association of PDA with tumor size (p-heterogeneity=0.04). Risk did not differ across the other assessed tumor characteristics (p-heterogeneity values >0.05).
Conclusion: These findings do not provide strong evidence that mammographic density parameters differentially affect specific breast cancer tumor characteristics.

Keywords: mammographic density, breast cancer, tumor characteristics, heterogeneity, multinomial logistic regression

Introduction

Mammographic density is an important breast cancer risk factor.13 The association between breast cancer and many well-established risk factors has been shown to be different according to the characteristics of the tumor.411 However, for mammographic density, this has not been established. Some studies report no heterogeneity in the association between mammographic density and breast cancer tumor characteristics;1222 while others indicate differences by hormone receptor status,3,2328 invasiveness,22,29 phenotype,30,31 tumor size,22,26,28,32,33 and stage.34 Most studies have limited the assessment of mammographic density qualitatively as defined by the BI-RADS classification, or quantitatively as percent dense area (PDA); the other mammographic density parameters, dense area (DA) and non-dense area (NDA) have seldom been taken into account.

It is important to elucidate whether mammographic density parameters are associated differentially across different breast cancer tumor characteristics. Such knowledge could help us understand pathological pathways, as well as identify susceptible groups of women in the general population, providing evidence that would improve the formulation of screening protocols and risk-reducing interventions.35

Materials and methods

Study population

The examined data originate from the British Columbia (BC) study subpopulation belonging to the Canadian Breast Cancer Study (CBCS).36 Incident female breast cancer cases aged 40 to 80 years diagnosed between 2005 and 2009 were recruited from the BC Cancer Registry; controls were enrolled from the Screening Mammography Program, from the same geographic area, and frequency-matched to cases in 5-year age groups. Participation was 54% among cases and 57% amid controls. This study was restricted to postmenopausal participants: 606 cases and 595 controls. The final sample, determined by the availability of screening film mammograms, was comprised of 477 cases and 588 controls. A questionnaire was used to collect information about personal characteristics and medical history.

Mammographic density measurement

Briefly, as it has been previously described,37 the most recent normal mammogram preceding recruitment into the study was selected for each participant. It was not possible to locate mammograms prior to study enrollment for 92 controls, so the mammogram after study enrollment, but closest to that date was chosen (average 2.3 years after enrollment, SD=0.7). 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 DA were determined by using the interactive thresholding method,38 via Cumulus software (Imaging Research Program, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Canada), by a blinded single reader (HAVG).

Breast tumor characteristics assessment

The methodology has been outlined before;35 in summary, 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 in situ hybridization (FISH); a FISH result of more than 6.0 HER2 gene copies per nucleus was considered positive.

Statistical analysis

Mammographic density parameters were analyzed as continuous covariates (DA and NDA expressed in terms of cm2, the percentage for PDA) and categorized into quartiles according to the distribution in controls. Since data-driven methods for the selection of confounders are susceptible to generate biased estimation of the effect of the exposure of interest,39 a direct acyclic diagram (DAG) was used to identify minimally sufficient adjustment sets of variables in the path between mammographic density parameters and breast cancer,40,41 through DAGgity42 (details can be found at Velásquez García et al).37 Even though the resulting number of the adjustment variables is relatively large, which results in diminished statistical power, the implementation of a minimally sufficient adjustment set in the models provides the best trade between statistical power loss and estimation with reduced bias. The Akaike information criterion was used to find the best characterization of the adjustment set variables in the models, as follows: body mass index (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 (hormone replacement therapy: never, <5 years, 5–12 years, >12 years), lifetime smoking (continuous), and alcohol consumption (continuous). In addition, an age by BMI interaction term (continuous) was incorporated in all models, to allow the associations of breast cancer risk and BMI to be subject to age, as suggested by Baglietto et al.2

Tests for heterogeneity between subtypes for each of the tumor characteristic were assessed by multinomial logistic regression utilizing breast cancer cases only.43,44 Adjusted odds ratios (aOR) and 95% CI were computed to estimate the associations between mammographic density parameters and breast cancer risk for each subtype separately using unconditional logistic regression, adjusted for the previously described variables. Trend tests were conducted by entering the relevant ordinal variable as a continuous variable into the model. Values were missing for some variables in 0.5–5.6% of the cases, and in 0.1–3.3% of controls;37 missing values were imputed via multiple imputations by chained equations (five iterations), present in the mice R package.45 Evaluations were also conducted after eliminating observations with missing values. Analyses were performed using Stata v.14.0 (Stata Corporation, College Station, TX, USA). All statistical tests were two-sided; the critical level of significance was set at 5%.

Results

Table 1 shows the characteristics of the study participants according to case or control status. Table 2 indicates the distribution of tumor characteristics for cases: over 75% were invasive cancers, with most in the 1.1–2.0 cm size category (n=145, 39.2%), and stage I (n=189, 39.6%). As expected in a population-based study, over 80% of tumors were histologically ductal (n=310, 83.8%), ER positive (n=287, 77.6%), PR positive (n=212, 57.3%), and HER2 negative (n=265, 71.6%). Tumor characteristics evaluated in association with mammographic density were invasiveness and stage and, for invasive cases only, tumor size, histology, and receptor status were also considered.

Table 1 Characteristics of study population

Table 2 Distribution of tumor characteristics on cases

Overall, when comparing the highest quartile with the lowest, DA (aOR=2.6, 95% CI 1.8–3.8, p-trend<0.001) and PDA (aOR=3.8, 95% CI 2.5–5.9, p-trend <0.001) were found directly associated to breast cancer in fully adjusted models; NDA (aOR=0.5, 95% CI 0.3–0.8, p-trend=0.025) was inversely related to breast cancer, controlling for the adjustment set variables. Similar results in terms of directions of the associations were obtained when using continuous values in the models (estimates for a 10-unit change in mammographic parameter value: DA, aOR=1.4, 95% CI 1.3–1.5, p-trend<0.001; PDA, aOR=1.4, 95% CI 1.3–1.6, p-trend<0.001; NDA, aOR=0.94, 95% CI 0.91–0.97, p-trend<0.001).

The results of the tests of heterogeneity among cases only, as well as the estimates of the associations between mammographic density parameters and breast cancer risk stratified by tumor characteristics, are shown in Table 3. Heterogeneity was found in the analyses by quartiles only for the association of PDA with tumor size (p-heterogeneity=0.04), and risk did not differ across the other assessed tumor characteristics (p-heterogeneity values >0.05). Sensitivity analyses eliminating observations with imputed values, as well as excluding the controls with breast density measured from mammograms taken after study enrollment, produced similar results (not shown). However, heterogeneity was found when assessing the association between PR status and PDA when observations with missing values eliminated (p-heterogeneity=0.01), as well as when using continuous values for mammographic density parameters (p-heterogeneity=0.016) in the main analyses with imputed values.

Table 3 Associations of mammographic density parameters stratified by breast cancer tumor characteristics in postmenopausal women

Discussion

In this population-based case–control study, a consistent association between mammographic density and breast cancer risk was observed. The measured mammographic density parameters were found to be important risk factors for breast cancer in all tumor types. DA and PDA were confirmed as independent risk factors directly associated with breast cancer; NDA was also found to be an independent factor, inversely associated with breast cancer risk. Our observations indicate that these associations do not vary according to breast cancer tumor characteristics, which is in agreement with various previous reports.1220 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, suggests that our study could be underpowered.

In this study, the purpose was not to evaluate absolute breast cancer subtype risk; instead, we estimated the relative risk (aOR) of cancer subtypes according to the value for breast density. In this way, OR can be calculated from a case–control study without knowledge of the exposure prevalence.

A strength of this study is that we opted for the DAG approach to select the covariates for adjustment, minimizing in this way the magnitude of the bias in our estimations.46,47 Furthermore, the considerable amount of participants’ information gathered in the CBCS made it possible to adjust for the identified minimally sufficient set. Another strength is the inclusion of in situ cases which enables the examination of previously reported differences in the association between mammographic density and invasiveness.22,28 Other strengths are the objective assessment of mammographic density via computer-assisted thresholding, and the use of craniocaudal views to limit the inclusion of subcutaneous fat in the mammographic density readings.48

Another limitation to be considered is the fact that, given the participation rates of the original study, potential response bias could be present in the information gathered through the questionnaire, used in the models’ adjustment set. However, CBSC estimates for known breast cancer risk factors are similar to those published in other epidemiological studies,36 indicating that important levels of biases are most likely not present. In addition, as mammographic density measurements are not usually revealed to screening participants in BC, it is implausible that breast density influenced enrolment in the study. Last, replication using larger independent datasets is necessary to confirm these results.

Conclusion

In conclusion, our findings indicate that mammographic density parameters, although important risk factors for breast cancer, are not differentially associated with breast cancer tumor characteristics.

Abbreviations

aOR, adjusted odds ratio; BC, British Columbia; BMI, body mass index; CBCS, Canadian Breast Cancer Study; DA, mammographic dense area; DAG, directed acyclic graph; ER, estrogen receptor; FISH, fluorescence in situ hybridization; HER2, human epidermal factor receptor 2; HRT, hormone replacement therapy; IHC, immunohistochemistry; NDA, mammographic non-dense area; PDA, mammographic percent dense area; PR, progesterone receptor.

Ethics approval and informed consent

Ethical approval for this study was provided by the University of British Columbia, British Columbia Cancer Agency Research Ethics Board (reference #H14-01614).

Data availability

The analyzed datasets are available from the corresponding author on reasonable request.

Acknowledgments

We thank Drs. Gertraud Maskarinec, Jennifer Stone, and Martin Yaffe for their kind assistance to ascertain our mammographic density readings’ consistency. We would like to express our deep gratitude to Ms. Karen Locken and Mrs. Christine Lam (BC Cancer Agency, Diagnostic Images), as well as the staff of the Screening Mammography Program of British Columbia, particularly Mrs. Carla Brown–John for their invaluable help. We highly appreciate the extensive support provided by Ms. Anoma Gunasekara, Mr. Gord Mawdsley (Sunnybrook Research Institute), Mrs. Zenaida Abanto (BC Cancer, Cancer Control Research), and the staff of the BC Cancer Registry, Breast Cancer Outcomes Unit.

Funding for the original study was provided by a grant from the Canadian Institutes of Health Research (PI: KJA, Funding Reference #69036). HAVG was supported by a Four Year Doctoral Fellowship Award from the University of British Columbia, and a Canadian Breast Cancer Foundation Fellowship (award #319404).

Disclosure

The authors report no conflicts of interest in this work.

References

1. 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):1159–1169. doi:10.1158/1055-9965.EPI-06-0034

2. Baglietto L, Krishnan K, Stone J, et al. Associations of mammographic dense and nondense areas and body mass index with risk of breast cancer. Am J Epidemiol. 2014;179(4):475–483. doi:10.1093/aje/kwt260

3. Bertrand KA, Scott CG, Tamimi RM, 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. doi:10.1158/1055-9965.EPI-14-1136

4. Potter JD, Cerhan JR, Sellers TA, 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;4(4):319–326.

5. 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;151(7):703–714.

6. 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;96(3):218–228. doi:10.1093/jnci/djh025

7. 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;29(5):419–426.

8. 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;8(4):R39. doi:10.1186/bcr1514

9. 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. doi:10.1186/bcr1525

10. Gaudet MM, Press MF, Haile RW, 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;130(2):587–597. doi:10.1007/s10549-011-1616-x

11. Turkoz FP, Solak M, Petekkaya I, et al. Association between common risk factors and molecular subtypes in breast cancer patients. Breast. 2013;22(3):344–350. doi:10.1016/j.breast.2012.08.005

12. 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;13(12):2090–2095.

13. 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;8(3):R30. doi:10.1186/bcr1507

14. Yang W-T, Dryden M, Broglio K, et al. Mammographic features of triple receptor-negative primary breast cancers in young premenopausal women. Breast Cancer Res Treat. 2008;111(3):405–410. doi:10.1007/s10549-007-9810-6

15. 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;18(2):479–485. doi:10.1158/1055-9965.EPI-08-0805

16. 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;19(6):1643–1654. doi:10.1158/1055-9965.EPI-10-0188

17. Phipps AI, Buist DSM, Malone KE, et al. Breast density, body mass index, and risk of tumor marker-defined subtypes of breast cancer. Ann Epidemiol. 2012;22(5):340–348. doi:10.1016/j.annepidem.2012.02.002

18. Eriksson L, Hall P, Czene K, et al. Mammographic density and molecular subtypes of breast cancer. Br J Cancer. 2012;107(1):18–23. doi:10.1038/bjc.2012.234

19. 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):337–347. doi:10.1007/s10549-012-2362-4

20. Pollán M, Ascunce N, Ederra M, 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;15(1):R9. doi:10.1186/bcr3380

21. 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;49:156–160. doi:10.1016/j.canep.2017.07.003

22. Krishnan K, Baglietto L, Stone J, et al. Mammographic density and risk of breast cancer by tumor characteristics: a case-control study. BMC Cancer. 2017;17(1):859. doi:10.1186/s12885-017-3871-7

23. Yaghjyan L, Colditz GA, Collins LC, et al. Mammographic breast density and subsequent risk of breast cancer in postmenopausal women according to tumor characteristics. J Natl Cancer Inst. 2011;103(15):1179–1189. doi:10.1093/jnci/djr225

24. Ding J, Warren R, Girling A, Thompson D, Easton D. Mammographic density, estrogen receptor status and other breast cancer tumor characteristics. Breast J. 2010;16(3):279–289. doi:10.1111/j.1524-4741.2010.00907.x

25. Conroy SM, Pagano I, Kolonel LN, Maskarinec G. Mammographic density and hormone receptor expression in breast cancer: the multiethnic cohort study. Cancer Epidemiol. 2011;35(5):448–452. doi:10.1016/j.canep.2010.11.011

26. Heusinger K, Jud SM, Häberle L, et al. Association of mammographic density with hormone receptors in invasive breast cancers: results from a case-only study. Int J Cancer. 2012;131(11):2643–2649. doi:10.1002/ijc.27592

27. Bertrand KA, Tamimi RM, Scott CG, et al. Mammographic density and risk of breast cancer by age and tumor characteristics. Breast Cancer Res. 2013;15(6):R104. doi:10.1186/bcr3570

28. Kerlikowske K, Gard CC, Tice JA, et al. Risk factors that increase risk of estrogen receptor–positive and –negative breast cancer. J Natl Cancer Inst. 2017;109(5):djw276. doi:10.1093/jnci/djx007

29. Aiello EJ, Buist DSM, White E, Porter PL. Association between mammographic breast density and breast cancer tumor characteristics. Cancer Epidemiol Biomarkers Prev. 2005;14(3):662–668. doi:10.1158/1055-9965.EPI-04-0132

30. 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;15(5):R76. doi:10.1186/bcr3470

31. Baré M, Torà N, Salas D, 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;154(2):403–415. doi:10.1007/s10549-015-3623-9

32. Sala E, Solomon L, Warren R, et al. Size, node status and grade of breast tumours: association with mammographic parenchymal patterns. Eur Radiol. 2000;10(1):157–161. doi:10.1007/s003300051020

33. Porter GJR, Evans AJ, Cornford EJ, 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;188(3):676–683. doi:10.2214/AJR.05.1950

34. Kerlikowske K, Cook AJ, Buist DSM, et al. breast cancer risk by breast density, menopause, and postmenopausal hormone therapy use. J Clin Oncol. 2010;28(24):3830–3837. doi:10.1200/JCO.2009.26.4770

35. Evans DGR, Warwick J, Astley SM, 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;5(7):943–951. doi:10.1158/1940-6207.CAPR-11-0458

36. Grundy A, Richardson H, Burstyn I, et al. Increased risk of breast cancer associated with long-term shift work in Canada. Occup Environ Med. 2013;70:1–8. doi:10.1136/oemed-2013-101482

37. Velásquez García HA, Sobolev BG, Gotay CC, et al. Mammographic non-dense area and breast cancer risk in postmenopausal women: a causal inference approach in a case–control study. Breast Cancer Res Treat. 2018;170:159–168. doi:10.1007/s10549-018-4737-7

38. Byng JW, Boyd NF, Fishell E, Jong RA, Yaffe MJ. The quantitative analysis of mammographic densities. Phys Med Biol. 1994;39(10):1629–1638. doi:10.1088/0031-9155/39/10/008

39. Glymour MM, Greenland S. Causal Diagrams. In: Rothman KJ, Greenland S, editors. Modern Epidemiology. 3rd ed. Philadelphia: Lippencott-Raven Publishers; 2008:183–209.

40. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10(1):37–48. doi:10.1097/00001648-199901000-00008

41. Pearl J. Causality: Models, Reasoning, and Inference. Cambridge: Cambridge University Press; 2000.

42. Textor J, Hardt J, Knüppel S. DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology. 2011;22(5):745. doi:10.1097/EDE.0b013e318225c2be

43. Dubin N, Pasternack BS. Risk assessment for case-control subgroups by polychotomous logistic regression. Am J Epidemiol. 1986;123(6):1101–1117. doi:10.1093/oxfordjournals.aje.a114338

44. Begg CB, Zhang ZF. Statistical analysis of molecular epidemiology studies employing case-series. Cancer Epidemiol Biomarkers Prev. 1994;3(2):173–175.

45. Buuren SV, Groothuis-Oudshoorn K. Mice : multivariate imputation by chained equations in R. J Stat Softw. 2011;45(3):1–67. doi:10.18637/jss.v045.i03

46. Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15(5):615–625. doi:10.1097/01.ede.0000091604.32542.97

47. Shrier I, Platt RW. Reducing bias through directed acyclic graphs. BMC Med Res Methodol. 2008;8(1):70. doi:10.1186/1471-2288-8-70

48. Shepherd JA, Kerlikowske K. Do fatty breasts increase or decrease breast cancer risk? Breast Cancer Res. 2012;14(1):102. doi:10.1186/bcr3169

Creative Commons License © 2019 The Author(s). This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms.php and incorporate the Creative Commons Attribution - Non Commercial (unported, v3.0) License. By accessing the work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.