6/9/2017 iCARE Breast Cancer Risk Model Development and Validation NO DISCLOSURES Montserrat Garcia-Closas, M.D. Dr.P.H. Senior Investigator and Deputy Director Division of Cancer Epidemiology and Genetics 2 1. iCARE risk modelling approach 1. iCARE risk modelling approach 2. PRS development in BCAC 2. PRS development in BCAC 3. Building integrated risk models 3. Building integrated risk models 4. Validation in prospective cohorts 4. Validation in prospective cohorts 5. Population risk stratification 5. Population risk stratification 3 4 1
6/9/2017 Breast cancer risk models Breast cancer models with hormonal/environmental factors Model Prediction Density? BBD Target Externally Online path? population validated in tool? prospective cohorts? • Many existing models for different uses and target populations: BCRAT (Gail) Invasive No Yes General Yes Yes BCRAT(Chen) Invasive Yes Yes General No No • 10 hormonal/environmental models Rosner–Colditz Invasive No No General Limited No Tworoger Invasive No No General No No • 12 hereditary models BCSC (Barlow) Inv+ DCIS Yes No Screening No No • 1 hormonal/environmental and hereditary model BCSC-BBD (Tice) Inv+ DCIS Yes Yes Screening No Yes • Many have not been externally validated in prospective cohorts BBD-BC Inv+ DCIS No Yes BBD biopsy No No • Many don’t have online tools for risk calculation Bodian Inv+ DCIS No Yes LCIS Limited Yes IBIS (Tyrer-Cuzick) Inv+ DCIS No Yes General & high Yes Yes genetic risk For a comprehensive review of risk models see Cintolo-Gonzalez et al. Breast Cancer Res Treat 2017 Cintolo-Gonzalez et al. Breast Cancer Res Treat 2017 5 6 Development of a synthetic risk model Why a new breast cancer risk model ? Individualized Coherent Absolute Risk Estimator (iCARE) • Flexible modelling approach: Used for model Population Population Population Population distribution distribution calibration Disease Disease • Comprehensive and easier to update of risk of risk and imputation of Incidence Incidence factors factors missing data • Incorporation of polygenetic risk scores • Updated rates and risk factor distributions in the target population Rate of Rate of Risk factor Risk factor competing competing • Accommodate missing data in risk factors RR RR mortality mortality Models for Models for • Predict subtype-specific risk Absolute Absolute Risk Risk Chatterjee et al. http://dceg.cancer.gov/tools/analysis/icare 7 8 2
6/9/2017 B reast CA ncer ST ratification Understanding the determinants of risk and prognosis of molecular subtypes 1. iCARE risk modelling approach Risk factors Tumor Prognosis 2. PRS development in BCAC subtypes Molecular subtypes Questionnaires Recurrence 3. Building integrated risk models SD vs Interval Breast density Mortality Early vs late onset Genetics (PRS) Aggressive 4. Validation in prospective cohorts 5. Population risk stratification Translating knowledge into risk stratification for precision prevention 6. Future work Risk model Validation Online tools / development implementation iCARE Prospective cohort General population BOACIDEAPlus studies 10 Polygenic risk score (PRS) based on 77 SNPs Distribution of risk alleles for 77 SNPs Analyses included women of European origin in iCOGS: (N=33,381) (N=33,673) 33,673 cases (21,365 ER+ and 5,738 ER-) 33,381 controls per-allele log (OR) for risk allele at locus j from logistic regression adjusted for study and 7 PCs β j x j number of risk alleles at j locus (0, 1 or 2) Mavaddat E et al, JNCI 2015 Mavaddat E et al, JNCI 2015 3
6/9/2017 Polygenic risk stratification by family history of breast cancer Risk stratification for ER-positive than ER-negative disease Women with a family history Women without a family history (10% of the population) (90% of the population) Risk for ER-positive disease Risk for ER-negative disease Quintiles of 77-PRS Quintiles of 77-PRS 25% Quintiles of 77-PRS Quintiles of 77-PRS Quintiles of 77-PRS Life-time breast cancer risk Life-time breast cancer risk 16% 16% 16% 7.5% 10% 9% 4% 5% 1% Mavaddat E et al, JNCI 2015 Mavaddat E et al, JNCI 2015 Improved PRS based on iCOGS + OncoArray analyses P-value cut off Number OR per SD AUC for SNP selection of SNPs (95% CI) 1. iCARE risk modelling approach 2. PRS development in BCAC 3. Building integrated risk models 77 SNP PRS 77 1.49 (1.43-1.57) 0.61 (P< 5x10 -8 in 2015) 4. Validation in prospective cohorts 5. Population risk stratification P< 10 -5 268* 1.65 (1.58-1.73) 0.64 *Selected based on data from 94,094 cases and 75,017 controls (European) from 69 studies in BCAC, divided into training (90%) and test (10%) sets. Mavaddat E et al, In preparation 16 4
6/9/2017 iCARE breast cancer risk model for the UK Multiplicative effects of PRS and risk factors Distribution of risk Distribution of risk Breast cancer Breast cancer Log linear and non-parametric risk scores from a model including factors from Health factors from Health incidence from incidence from Survey for England, Survey for England, Office of National Office of National 77- SNP PRS and hormonal/environmental risk factors (BCAC) Fertility Tables, Fertility Tables, Statistics, UK Statistics, UK others others Non-parametric smoothing 3 500 Linear logistic RR RR 2 Rate of Rate of Goodness of fit tests 400 competing competing Reproductive, Reproductive, P=0.252 global test 1 mortality from mortality from R Models Models BMI, OC, HRT, BMI, OC, HRT, Fitted log O requency 300 Office of National Office of National P= 0.179 tail-based test alcohol, BBD, alcohol, BBD, 0 for for Statistics, UK Statistics, UK F family history; family history; 200 -1 PRS PRS Absolute Absolute -2 100 Risk Risk -3 Literature review 0 BPC3 (8 cohorts) -3 -2 -1 0 1 2 3 Rudolph et al. In submission BCAC (PRS) Log linear risk score 17 18 Alcohol and breast cancer risk by PRS percentiles Studies to evaluate PRS and density joint effects Study Name Cases Controls Bavarian Breast Cancer Case-control study (BBCC) 512 367 77-SNP PRS OR [95%CI] Percentiles Mayo Mammography Health Study (MMHS) 456 1166 Goodness of fit tests Nurses Health Study (NHS) 850 849 Genetic risk P=0.013 global test European Prospective Investigation into Cancer (EPIC) 86 968 P= 0.18 tail-based test Mayo Clinic Breast Cancer Study (MCBCS) 677 864 Melbourne Case-control study (MCCS) 68 28 Multi-ethnic Cohort (MEC) 110 101 Singapore and Sweden Breast Cancer Study (SASBAC) 869 783 TOTAL 3,628 5,126 Rudolph et al. In submission Vachon et al In preparation 19 20 5
6/9/2017 PD OPERA (95%CI) Association of PD and 77-SNP PRS with breast cancer risk Study Name Percent density 77-SNP PRS OPERA* (95%CI) OR per SD (95%CI) BBCC 1.18 (1.0, 1.40) 1.30 (1.13, 1.49) Percent density and MMHS 1.67 (1.45, 1.92) 1.59 (1.42, 1.78) breast cancer risk by NHS 1.45 (1.32, 1.59) 1.58 (1.43, 1.75) PRS percentiles EPIC 1.32 (1.04, 1.67) 1.41 (1.12, 1.77) MCBCS 1.89 (1.65, 2.17) 1.46 (1.32, 1.62) MCCS 1.62 (1.03, 2.56) 1.21 (0.74, 1.96) MEC 1.53 (1.21, 1.92) 1.57 (1.17, 2.11) SASBAC 1.29 (1.17, 1.42) 1.58 (1.43, 1.75) Overall 1.45 (1.38, 1.52) 1.51 (1.44, 1.59) Vachon et al. In preparation Vachon et al. In preparation 21 * Age and BMI adjusted Multiplicative effects of PRS and mammographic breast density Log linear and non-parametric risk scores from a model including 77- SNP PRS and mammographic density 1. iCARE risk modelling approach 2. PRS development in BCAC 3. Building integrated risk models 4. Validation in prospective cohorts Goodness of fit tests P=0.164 global test 5. Population risk stratification P= 0.372 tail-based test Vachon et al. In preparation 23 24 6
Recommend
More recommend
