Inferring causality in observational epidemiology: Breast Cancer Risk as an Example Mary Beth Terry, PhD Department of Epidemiology and Environmental Sciences
Cancer Genes vs Environmental Risk Factors Risk Factor Cancer Site Tobacco Lung Asbestos Lung H. Pylori Infection Gastric Cancer P53 Rb MLH1 MSH2 BRCA1 BRCA2 APC and many, many others G E Human Papillomavirus (HPV) Infection Vinyl Chloride Hepatitis B and C Infection Maternal Diethylstilbestrol (DES) Use Ionizing Radiation Cervical Cancer, Head and Neck Cancer, Anal Cancer Liver Liver Vaginal Adenocarcinoma Leukemia All found through observational designs
Studying GXE in those at highest risk to make inferences that are valid for the population G E
Evidence- based Medical Research Inference Efficiency Randomized Control Trial Cohort Case- Control Cross- Sec8onal Generalizability
Enriched cohorts can be smaller and follow for less time
Family-based Cohorts Have Power at the Tail Prospective Family Study Cohort (PROF-SC) Terry MB et al Int J Epidemiol. 2015
So what about the Environment and Breast Cancer? Lindsey A. Torre et al. Cancer Epidemiol Biomarkers Prev 2016;25:16-27 Gosavi RA, Knudsen GA, Birnbaum, LS, Pedersen LC. EHP 2013 Zhang J, Huang Y, Link K, Wu K. PLOS ONE 2015 March 21, 2017 Page 7
Most empirical evidence relatively modest 1) Most studies fail to factor in windows of breast cancer susceptibility 2) Most studies estimated in average risk cohorts More modest the evidence, the greater the need for replication across DIFFERENT study designs to rule out the role of BIAS (information, selection and confounding) March 21, 2017 Page 8
Most Breast Cancer Risk Factors Are Modest < 1 < 2-Fold > 2-Fold NSAIDs Early Menarche Family History Tamoxifen Physical Activity Never Pregnant Never Breastfed Late Age at First Birth Late Menopause Hormone Use (HRT, OC) Overweight Alcohol Benign Breast Disease Mammographic Density March 21, 2017 Page 9
Summary of Breast Cancer Risk Assessment Models Model Family History Mutations Polygenes Risk Factors Model Family History Mutations Polygenes Risk Factors Claus Claus Multigenerational Multigenerational Gail/BCRAT Gail First-Degree First-Degree Yes Yes BRCAPRO IBIS Multigenerational Multigenerational BRCA1/2 BRCA1/2 Yes BOADICEA IBIS Multigenerational Multigenerational BRCA1/2 BRCA1/2 Yes Yes BOADICEA Multigenerational BRCA1/2 Yes
Large Differences in Absolute Risk IBIS Life6me Risk IBIS Ten Year risk BOADICEA Life6me Risk BOADICEA Ten Year Risk Quante AS, Whittemore AS, Shriver T, Hopper JL, Strauch K, Terry MB. J Natl Cancer Inst. 2015
Model with n-genetic Factors (Red) Fit Better than Model with only Genetic Factors (Blue) to the Observed 10-year Breast Cancer Risks, Observed > Expected Overall (N=16,316) Under 50 (N=10,186) Red IBIS Blue BOADICEA Red IBIS Blue BOADICEA Overall 10- year Observed (95% CI) 5.5% (5.2% - 5.9%) IBIS, Mean 4.8% BOADICEA, Mean 4.2% Age < 50yr 10- year Observed (95% CI) 4.6% (4.1% - 5.0%) IBIS, Mean 4.0% BOADICEA, Mean 3.6% March 21, 2017 Page 12
So what about the Environmental Risk Factors in Women at High Risk for Breast Cancer? What else can help explain risk in high risk women? What else can be used to improve risk assessment? Fraga et al. PNAS 2005
Differences Observed within Families in Biomarkers < 2 fold 2 fold 3-4 fold citation Shorter Telomere Length Lower Global Methylation Cancer Research 2007 Epigenetics 2012 Lower Sat2 repetitive element methylation Plasma Protein Carbonyls Carcinogenesis 2012 Cancer Research 2009 Poorer Nucleotide Excision Repair JNCI 2005 Adjusted for age at blood draw * disintegra6ons per minute Double Strand Break Repair Carcinogenesis 2008 Wu HC, et al. In prepara6on
Polycyclic Aromatic Hydrocarbon (PAH) Exposure Jung KH et al, Environ Res 2014
PAH-DNA Adducts and Breast Cancer Risk in a Population- Based Case Control Study (LIBCSP, PI M Gammon) 3.4 3.2 3.0 2.8 2.6 Odds Ratios 2.4 2.2 2.0 1.8 1.92 1.92 1.83 1.6 1.61 1.4 1.2 1.0 1.26 1.40 1.09 1.41 0.8 Q2 Q3 Q4 Q5 XPD XRCC1 ERCC1 XPD Gammon (2004) - Overall PAH by quantiles Terry (2004) - Shen (2005) - Detectable n-smokers PAH-DNA with dectable adducts ( PAH-DNA median) adducts Crew (2007) - Detectable PAH- DNA adduct levels Gammon MD, et al., Arch Environ Health, 2004; Terry MB et al., Cancer Epidemiol Biomarkers Prev., 2004; Shen J et al., Cancer Epidemiol Biomarkers Prev., 2005; Crew KD et al., Cancer Epidemiol Biomarkers Prev., 2007.
Increase in breast cancer risk from PAH by absolute risk of breast cancer, New York site of BCFR as esqmated by the BOADICEA, New York site of the BCFR BOADICEA 10- year Breast Cancer Risk 3.4% 10% Mean vs n- detect, OR (95% CI) 2.35 (1.13, 4.91) 2.14 (1.00, 4.60) 75th % vs n- detect, OR (95% CI) 2.48 (1.14, 5.41) 2.74 (1.18, 6.36) 90th % vs n- detect, OR (95% CI) 2.80 (1.05, 7.46) 4.84 (1.41, 16.5)
Why size ma7ers in Observa9onal Epi < 1 < 2-Fold > 2-Fold NSAIDs Early Menarche Family History Tamoxifen Physical Activity Never Pregnant Never Breastfed Late Age at First Birth Late Menopause Hormone Use (HRT, OC) Overweight Alcohol Benign Breast Disease Mammographic Density PAHs? March 21, 2017 Page 18
Summary and Implications 1) Most causal factors for cancer found through observational designs 2) Family-based cohorts are an efficient design to investigate GXE 3) Larger associations are less likely driven by bias 4) Just like with genes, finding GXE in familybased studies, results still generalizable to those without a family history March 21, 2017 Page 19
Acknowledgements Columbia Collaborators Regina Santella, PhD Rachel Miller, MD Jing Shen, PhD BCFR Collaborators Australia John Hopper, Ph.D. Melissa Southey, Ph.D. rthern California Esther John, Ph.D. Alice Whittemore, Ph.D. Ontario Irene Andrulis, Ph.D. Philadelphia Mary Daly, M.D. Utah Saundra Buys, M.D. Funding U01 BCFR Registry U01CA069398 (Terry) U01 NIEHS BCERP CCCEH (Terry, Miller) R01 Birth Cohort R01CA104842 (Terry) U01 Pedigree U01ES019471 (Terry, Cohn) R01 Legacy R01CA138822-01A2 (Terry) Breast Cancer Research Foundation (Santella, Terry) R01 ProFC 1 R01 CA159868-01 (Terry, Hopper)