Department of Applied Health Research Translating genomics into population-based cancer screening strategies opportunities and challenges Nora Pashayan DAHR & HBRC Seminar March 8th 2016 Today’s presentation Is a different approach of screening needed? How risk-stratification could be used in cancer screening programmes? Polygenic risk score Polygenic risk score combined with non-genetic risk factors Implications on screening outcomes Examples from prostate, breast and ovarian cancers What are the challenges? Generating evidence Implementation 2 Benefit-harm balance of screening ‘All screening programmes do harm; some also do good, and of these, some do more good than harm’* Sir Muir Gray Potential benefits * Potential harms 3 Gray et al. BMJ 2008:336(7642):480-483 Overdiagnosis Detection of disease through screening that otherwise would not have been diagnosed within a patient’s lifetime Occurs in both indolent, non-progressive cancer, and progressive cancer in an individual that dies before the cancer progresses to manifest clinically Epidemiological concept 4 Controversies in cancer screening Screening for breast cancer Estimates of overdiagnosis vary widely The Independent UK Panel on Breast Cancer Screening estimated that for every breast cancer death averted three cases are likely to overdiagnosed * Screening for prostate cancer by PSA In the European Randomised Study of Screening for Prostate Cancer (ERSPC), at 13 years of follow up, for every cancer death averted, 12 to 36 excess cases were detected ** USPSTF recommends against screening * ** Independent UK Panel on Breast Cancer Screening. Lancet, 2012: 380(9855):1778-86 Auvinen et al. Clin Cancer Res, 2015:pii: clincanres.0941.2015 5 To improve benefit to harm balance… Conventionally, age is used to define the target population If breast screening programme provided to women 47-73 years: • 29% of the English women population; accounting for Population of women • 60% of breast cancer cases 47-73 35-79 80+ Breast cancer cases (25/100,00 population) Based on population size and number of cancer registrations in England 2002-2006 6 Tumour progression is not homogeneous Schematic diagram of tumour progression The outcome of screening depends on the behaviour of the tumour The outcome may be improved by varying age of start of screening and frequency of screening 7 Diagram adapted from: Esserman et al. JAMA 2009: 302(15):1685-92 Shift in screening approach A shift from ‘One size fits all’ to ‘Risk-tailored screening strategy’ (precision screening / risk-stratified screening/ personalised screening) If it • improves the benefit-harm balance of screening • is cost-effective • is acceptable to the users and to the providers • is accessible to all • is feasible to implement 8 Risk-tailored screening: two tiered screening Risk-assessment (using age, family history, genetic profile, etc.) Q. What factors to include? Stratify population into several groups Q. How many? Q. What thresholds? -4 -2 0 Risk 2 4 Tailor screening to each risk group (different screening modality, age for start / end of screening, inter-screening interval) Q. Which interventions work for different risk groups? Diagram adapted from: Burton et al. IJPH 2012: 9(4) 9 Risk-assessment: at two different points Pre-screening * Risk assessment of all Based on the risk estimates, tailor screening Post-screening Baseline screening of all Based on the findings of the baseline screening, tailor screening Models of risk-tailored screening More intensified screening for those at higher than population average risk e.g. (more often, start earlier age, stop later age) Less intensified screening for those at lower than population average risk Fully stratified – covering the entire spectrum of risk ** * ** Dent et al. PH Genomics 2012 : doi: 10.1159/000345941 Chowdhury et al. Genet Med 2013: 15(6):423-32 10 Risk assessment Risk score based on: Non-genetic risk factors Genetic risk factors Common susceptibility variants at GWAS significance level Common susceptibility variants at or below GWAS significance level Common variants plus the high penetrance alleles like in BRCA1/2 Combination of genetic and non-genetic risk factors 11 Human genetic variation Single nucleotide polymorphism (SNP) Any two individuals share about 99.9% of their DNA sequence Of the remaining 0.1%, about 80% consists of variation at single nucleotides A sequence variation may be described as polymorphic if occurs with frequency ≥ 0.01 Most SNPs are bi-allelic Allele – alternative form of a genetic marker Genotype - the diploid combination of alleles at a particular locus in an individual. For a SNP with two alleles (A, a) there are 3 possible genotypes (AA, Aa, aa) 12 Genetic susceptibility to prostate and breast cancers Common low penetrance variants, (currently known 94 loci), explain ~15% of excess Familial Relative Risk (FRR) Subtype specific Rare high penetrance alleles in BRCA1 and BRCA2 explain ~15% of FRR Rare moderate penetrance alleles in TP53, LKB1, CDH1, PTEN, BRIP1, PALB2, ATM and CHEK2 explain ~ 6% of the FRR Common low penetrance variants, currently known 100 loci, explain ~33% of the FRR 13 Polygenic susceptibility Prostate cancer common susceptibility loci SNP Locus Risk-allele frequency Risk ratio per allele* Variance** 1 2q31 0.94 1.30 0.008 2 2p15 0.19 1.15 0.002 3 2p21 0.23 1.08 0.002 4 3p12 0.11 1.18 0.002 5 3q21.3 0.28 1.12 0.002 7 4q24 0.55 1.09 0.004 8 4q22 0.66 1.10 0.003 9 4q22 0.46 1.08 0.007 10 6q25 0.29 1.17 0.013 … 100 Total SNP: Normal variation in DNA sequence At each locus, 2 possible alleles - Risk allele - Protective allele Risk allele frequency >5% Relative risk per allele carried <1.8 Alleles combine multiplicatively for each locus on relative risk scale (or additively on logarithmic scale) Polygenic risk score: weighted sum of the risk alleles 0.47 * Based on individuals of European ancestry **Variance of the risk distribution due to an allele is derived from RR conferred by each allele and its population frequency 14 Polygenic risk distribution 3 loci 6 alleles 27 genotypes Aabbcc AaBbcc AabbCc aaBbCc aabbCC aaBBcc Aabbcc aaBbcc aabbCc aabbcc AABbcc AAbbCc AaBbCc AabbCC AaBBcc aaBBCc aaBcCC AABBcc AABbCC AAbbCC AaBBCC AaBbCC aaBBCC AABBCc AABbCC AaBBCC AABBCC For 94 SNPs there are 188 alleles and 394 possible genotype combinations Each combination is associated with different risk and occurs at different frequency in the population 15 Log-normal distribution of polygenic risk 2 Cases Population 2 = Mean = Variance log(RR) ' 2 Mean among cases: Variance among cases: 2 = variance among the population 16 Pharoah et al. Nat Genet 2002 31(1): 33-6 Distribution of polygenic risk –Prostate cancer Risk distribution based on 100 Prostate cancer susceptibility loci (polygenic variance 0.47) Centile in pop Population -4 Cases -2 0 log(RR) 2 4 RR 1st 0.16 10th 0.33 90th 1.90 99th 3.88 RR compared to population average risk 17 Risk-stratification - case of prostate cancer Proportion of prostate cancer cases explained by the proportion of the population at highest risk (polygenic risk based on the known 100 prostate cancer loci) 1 50% FRR Proportion of cases at highest risk .9 • AUC based on the 100 SNPs: 0.69 • AUC based on 50% Familial relative risk RR (FRR): 0.74 .8 .7 .6 100 SNPS .5 .4 .3 .2 .1 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Proportion of population at highest risk AUC (area under the receiver operating characteristic curve), overall measure of predictive ability. AUC values range from 0.5 (total lack of discrimination) to 1.0 (perfect discrimination) 18 Combined risk score Discrimination – case of breast cancer Assuming multiplicative interaction between genetic and non-genetic risk factors: Models – Breast cancer AUC % Cases accounted by 50th percentile of population at highest risk Limited non-genetic risk factors 0.58 62 Non-genetic risk factors 0.62 65 94 SNPs (15% FRR) 0.65 70 94 SNPs + non-genetic risk factors 0.68 75 94 SNPs + non-genetic risk factors + breast density 0.71 79 FRR 50% + non-genetic risk factors + breast density 0.73 81 Non-genetic risk factors*: menarche age, no. full tem pregnancies, oral contraceptive use, benign lesion, family history, BMI, smoking and alcohol intake Limited non-genetic risk factors**: age, pregnancies, OCP, FH FRR: familial relative risk • • • * Garcia-Closas et al, JNCI 2014; 106(11) | **Pharoah et al, NEJM 2008; 358(26):2996-803 19 Utility of risk-stratification - Examples Preventative interventions Ovarian cancer Screening strategy Breast cancer Prostate cancer 20 Preventative intervention - Ovarian cancer Ovarian cancer • 18 common susceptibility loci at GWAS significance • Polygenic variance of 0.09 and AUC=0.58 • Other known risk factors: • first-degree family history of ovarian cancer endometriosis parity OCP use tubal ligation Decision modelling showed that risk reducing salpingo-oophorectomy is cost-effective in post-menopausal women with ≥ 5% lifetime risk of ovarian cancer (ICER <£15,000/QALY)* *Manchanda et al. Gynaecol Oncol 2015; 139:487-94 21 Preventative intervention Distribution of the lifetime risks of ovarian cancer based on the observed combinations of the risk factors (polygenic and non-genetic) Lifetime risk range: 0.35% to 8.78% LR 5% • • • • Population average LR 1.4% The blue lines along the X axis represent the observed 214 combinations of risk factors and the height of the line the frequency of the group Lifetime risk based on the US population 7% of the population at <0.5% lifetime risk; and 1.5% of the population at > 4% lifetime risk 73% of women with lifetime risk >4% had no family history of ovarian cancer Pearce et al. Cancer Epidemiol Biomarkers Prev 2015;24:671-676 22 Screening strategy - age of start of screening Breast cancer 10-year absolute risk of developing breast cancer for women with and without family history by polygenic risk percentiles Women with Family History Women without Family History 0.10 0.10 >80% 0.09 60-80% 60-80% 0.08 40-60% 0.07 20-40% 0.06 <20% 0.05 Reference 0.04 20-40% 0.06 <20% 0.05 Reference 0.04 0.03 0.03 0.02 0.02 0.01 0.01 0.00 0.00 20 25 30 35 40 45 50 55 60 Age (years) • 40-60% 0.07 10 year risk 10 year risk 0.08 >80% 0.09 65 20 25 30 35 40 45 50 55 60 65 Age (years) Reference: 2.5% 10-year absolute risk for developing breast cancer corresponds to risk of UK women aged 47 , i.e. age of invitation to the UK NHS Breast Screening programme Mavaddat et al. JNCI 2015: 107(5): djv036 23 Utility – screening programme efficiency Screening eligibility based on age alone vs. based on absolute risk (dependent on age and polygenic profile) Risk threshold is equivalent to risk threshold for eligibility based on age alone (10-year absolute risk of 2.5%) Population eligible for screening Cases potentially screen-detectable NHS Breast screening (age-based) Women 47-73 years* 7.44x106 22,359 Risk-stratified screening Women 35-79 years & 10-yr AR 2.5% (currently known 94 SNPs) 5.66x106 21,934 -1.78x106 (- 24% ) -423 (-2%) Difference per screening round Based on population and cancer registrations in England 2002-2006 * Currently NHSBSP covers women 50-70 years Pashayan et al. BJC 2011: 104(10):1656-63 24 Reclassification: Eligibility for screening Prostate cancer Population of 100 men, 45-79 years, stratified by age group and risk threshold categories (i), that would be eligible for screening based on age (ii) and based on age and polygenic risk (iii) (i) (ii) (iii) < 55 years of age and 10-year absolute risk < 2% 55 years of age and 10-year absolute risk < 2% < 55 years of age and 10-year absolute risk 2% 55 years of age and 10-year absolute risk ≥ 2% Pashayan et al. BJC 2011; 104(10):1656-63 25 Reclassification: cases detectable Prostate cancer Population of 100 men with prostate cancer, 45-79 years, stratified by age group and risk threshold categories (i), that would be detectable screening based on age (ii) and based on age and polygenic risk (iii) (i) (ii) (iii) < 55 years of age and 10-year absolute risk < 2% 55 years of age and 10-year absolute risk < 2% < 55 years of age and 10-year absolute risk 2% 55 years of age and 10-year absolute risk ≥ 2% 26 Polygenic risk and overdiagnosis - Prostate cancer I. UK based study Total number of UK men 50-69 years: 15,747 - with screen-detected prostate cancer: 2,148 (ProtecT) - with clinically-detected prostate cancer: 5,991 (SEARCH, UKGPCS) with no prostate cancer: 7,608 (ProtecT, SEARCH ) Polygenic risk score based on 66 prostate cancer SNPs The observed prevalence of screen- detected prostate cancer is a combination 50 Pj MSI j oI' zj of non-overdiagnosed and overdiagnosed cancers % Overdiagnosis 45 40 35 30 25 20 15 10 5 0 Q1 Q2 Q3 Q4 Quartiles of polygenic risk score distribution in the population Pashayan et al. Genet Med 2015: doi: 10.1038/gim.2014.192 27 Polygenic risk and overdiagnosis - Prostate cancer (2) II. ERSPC – Finland section Screening trial based on 71,502 men 55-67 years, followed up for 13 years, screened with PSA every 4 years up to 3 rounds or age 71 Derived mean sojourn time and PSA test sensitivity from the incidence of interval cancers Using these estimates, calculated the expected number of non-overdiagnosed cancers for each round of screening Proportion of prostate cancers likely to be overdiagnosed derived from the observed and expected cancers Controls stratified to below and above 50th centile of polygenic risk distribution derived based on 66 prostate cancer SNPs Overall % cancers overdiagnosed (95% CI) 42 (37-52) Lower risk group 58 (54-65) Higher risk group 37 (31-47) 28 Pashayan et al. BJC 2015: doi: 10.1038/bjc.2015.289 Targeted screening 100 men with screen-detected prostate cancer stratified by polygenic risk and probability of overdiagnosis (i) and impact of targeting screening to men above 50th centile (higher) polygenic risk (ii) – based on ERSPC-Finland findings i ii Lower polygenic risk overdiagnosed Higher polygenic risk overdiagnosed Lower polygenic risk Non-overdiagnosed Higher polygenic risk Non-overdiagnosed Targeting screening to men at higher polygenic risk: 50% less screening rounds 38% less overdiagnosed cacers 20% of non-overdiagnosed cancers missed 29 Implementation challenges Chowdhury et al 2013, Genet Med 15(6):423-32 | http://www.phgfoundation.org/news/15513/ 30 Key implementation issues Feasibility: The set up and delivery of risk-stratified screening programme is more complex than a programme with eligibility based on age alone • Dynamic risk score • Data safeguarding • Ethical,legal, social implications (ELSI) and organisational complexities Acceptability to the public, health professionals and policy decision makers to use genetic profile in risk estimation Risk perception and attitude of public and of professionals Risk communication Accessibility to ensure equitable delivery and uptake of personalised screening programme Healthcare professional training for interpretation of genetic risk score and better understanding of advantages and challenges of stratified intervention Evidence on that risk-stratified screening can do more good than harm at affordable cost 31 Summary There is a need for different approaches of screening to improve the benefit to harm balance of screening and to better use of resources Risk scores based on genetic and non-genetic risk factors could be used for risk-stratification for risk-tailored screening Risk-stratified screening programmes could improve the efficiency of the screening programme The set up and delivery of risk-stratified screening programme is more complex than a programme with eligibility based on age alone Robust evidence is needed on the effectiveness, cost-effectiveness of such programmes and on best way of implementation 32 Acknowledgements Funding Cancer Research UK training and clinician scientist fellowships EU-FP7 funded Collaborative Oncological Gene environment Study (COGS) Enabling Prof Paul Pharoah Prof Stephen Duffy Foundation for Genomics and Population Health (PHG Foundation) 33