Start Your Search
MS29 - Selection into Screening Programs: Interplay of Risk Algorithms, Genetic Markers and Biomarkers (ID 807)
- Event: WCLC 2018
- Type: Mini Symposium
- Track: Screening and Early Detection
- Presentations: 1
MS29.03 - Polygenic Risk Score for Risk Assessment (ID 11527)
14:00 - 14:15 | Author(s): Paul Brennan
Background: Genome-wide association studies uncovered multiple lung cancer susceptibility genes, and consortium efforts greatly increased our ability to investigate the genetic architecture of histological subtypes. However the clinical utility of these genomic discoveries remains unclear. Method: We therefore constructed a risk prediction model with polygenic risk score (PRS) based on 18,316 lung cancer patients and 14,025 controls with European ancestry, via 10-fold cross-validation with elastic net penalized regression. Model calibration was assessed, and was validated with UK biobank data (N=336,911 unrelated participants with European ancestry). To evaluate its potential clinical utility, the PRS distribution was simulated in the National Lung Screening Trial (NLST, N=50,772 participants). Absolute risk was estimated based on age-specific lung cancer incidence and all-cause mortality as competing risk. Added value of PRS to the risk prediction model was assessed by Net Reclassification Index. Results: A PRS was constructed based on 128 independent lung cancer variants using regularized penalized regression. The lung cancer ORs for individuals at the bottom 5% and top 5% of the PRS distribution were 0.49 (95%CI=0.43-0.56, P=2.7e-26) and 2.23 (95%CI=1.93-2.58, P=2.3e-27) in the training set, and 0.46 (95%CI=0.34-0.64, P=2.50e-6) and 1.33 (95%CI=1.08-1.64, P=7.10e-3) in the testing set, versus those at 40 to 60% as the referent group. The OR per standard deviation of PRS was 1.43 (95%CI=1.39-1.47. P=7.8e-138) for overall lung cancer risk in the training set and 1.24 (95%CI=1.18-1.30, P=2.59-e19) in the testing set. When considering age as the time scale, PRS separated out the curve of 5-year absolute risk and cumulative risk. When simulating the PRS distribution in the NLST population, we estimated 47.4% of cases occurred in the top 20% of the individuals with highest lifetime cumulative risk. Discussion: Including well-established genomic information in the risk model can contribute to the risk stratification of the population.
S01 - IASLC CT Screening Symposium: Forefront Advances in Lung Cancer Screening (Ticketed Session) (ID 853)
- Event: WCLC 2018
- Type: Symposium
- Track: Screening and Early Detection
- Presentations: 1
- Moderators:John Kirkpatrick Field, James L Mulshine
- Coordinates: 9/23/2018, 07:00 - 12:00, Room 203 BD
S01.07 - The U19 Plans for Integration of Biomarkers into Future Lung Cancer Screening (ID 11888)
08:00 - 08:50 | Presenting Author(s): Paul Brennan
We are performing a series of three integrated research projects with the unifying goal of reducing mortality from LC by applying targeted approaches to its prevention or early detection. These projects study (1) genetic susceptibility to nicotine dependence and lung cancer, (2) biomarkers for early detection, and (3) application of the results for LC screening. This proposal leverages an extensive collaborative framework and wealth of data from the International Lung Cancer Consortium (ILCCO), the Transdisciplinary Research in Cancer of the Lung (TRICL) Consortium and the Lung Cancer Cohort Consortium (LC3). Epidemiological data from 60 LC studies have been harmonized within ILCCO including 71,000 cases and more than 1 million cohort individuals.
Aims and Results
Project 1: Genomic Predictors of Smoking and Lung Cancer Risk. This project extends and augments genomic analyses that have been completed on 16,000 LC cases and 50,000 controls and extensively characterizes the contribution that genetic variation makes to LC susceptibility. The four aims are. Aim 1: To precisely characterize the contribution of common genetic variation to LC etiology. We will analyze a GWAS of LC of 47,506 genotyped LC cases and 63,687 controls. Aim 2: To investigate uncommon genetic variants using imputation approaches. Aim 3: To identify genetic effects on smoking behavior. Aim 4: To characterize joint effects of environmental and genetic interactions on LC risk. For this aim we will take advantage of novel statistical approaches (Mendelian Randomization, Mediation analysis, gene by environment interactions and pathway based analyses) developed by our team to provide a comprehensive approach to evaluating the impact of environmental factors according to genetic background. Recent findings from project 1 include identification of 10 new loci influencing lung cancer risk, the identification of 3 novel gene-smoking interactions contributing to lung cancer risk, identification and validation of two rare variants that convey an over four fold higher risk for lung cancer among carriers, and Mendelian randomization studies that show excess BMI and shorter telomere lengths increase lung cancer risk in a histology-dependent fashion.
Project 2: Biomarkers of Lung Cancer Risk. Multiple preliminary studies have implicated a wide range of circulating biomarkers in risk prediction for lung cancer. In Project 2, we hypothesize that a comprehensive and extensively validated risk prediction model that incorporates such risk biomarkers has the potential to substantially improve the selection of subjects at a high risk of developing LC and that these individuals are most likely to benefit from CT screening. This project involves three aims. Aim 1: To organize the LC3, including identifying the study population of 2,300 former and current smoking LC cases that were diagnosed within 5 years of donating their blood sample along with one smoking-matched control per case; and organize sample shipments and database preparation. Aim 2: To replicate a comprehensive panel of promising risk biomarkers and identify those that may be useful for risk prediction. This will involve assaying pre-diagnostic plasma samples for immune biomarkers, protein biomarkers such as pro-surfactant protein B, micro RNAs, methylation markers, and 34 additional promising biomarkers implicated in lung cancer. We will base this initial analysis on 800 case-control pairs from three LC3 cohorts, and define a panel of replicated risk biomarkers that provide non-redundant information on disease risk. Aim 3: To extensively evaluate all replicated risk biomarkers from Aim 2, identifying a minimum set of validated risk biomarkers, and ultimately evaluate the extent to which they improve risk prediction models. This will involve performing additional assays for 1,500 additional case-control pairs selected from 16 separate LC3 cohorts. The final outcome of this work will be risk prediction models incorporating a distinct set of biomarkers that provide meaningful information on disease risk, and these biomarkers will finally be evaluated in CT screening studies in collaboration with Project 3. This project recently completed analysis of a set of 4 biomarkers that improve the classification accuracy in prediction of lung cancer risk by 14% compared with a model that only included demographic and smoking information.
Project 3: Translating Molecular and Clinical Data to Population Lung Cancer Risk Assessment will evaluate radiographic models using data from the National Lung Screening Trial (NLST), lung cancer CT screening programs in British Columbia Cancer Agency (BCCA), Early Detection of Lung Cancer – a Pan-Canadian Study (PanCan), and the International Early Lung Cancer Action Program- Toronto (IELCAP-Toronto) along with the UK Lung Screen Trial (UKLS), and the Dutch Belgian randomized Lung Cancer Screening trial (NELSON) trial. Data from Projects 1-2 will be used to improve the risk prediction model and the nodule probability models. There are 2 specific aims. Aim 1 will establish an integrated risk prediction model to identify individuals at high risk of lung cancer, initially analyzing epidemiological and smoking related phenotypes and then integrating targeted biomarker, genomic profile, and lung function data applied to LC CT screening populations. We will study 950 CT-detected LC patients with biosamples from 46,057 screening individuals. Specific Aim 2 will establish a comprehensive LC probability models for individuals with LDCT-detected non-calcified pulmonary nodules. In this aim we will (a) first establish the 2D diameter-based probability model in N. American CT programs based on 36,481 participants, and then validate it based on 9,576 participants in the European LDCT programs; (b) establish the volume 3D and radiomics-based probability model in European CT programs based on 9,576 participants in European CT programs, and then validate it in the North American CT screening populations; and (c) assess the added predictive value and clinical usefulness of targeted genomic and molecular profiles in both the 2D diameter- and 3D and radiomics volume-based LC probability models based on risk stratification table analysis and decision curve analysis. Finally we will (d) compare the model performance with the existing classification system such as Lung-RADS. This project has developed and evaluated a polygenic risk score using data from project 1 which highly significantly improves risk prediction for lung cancer risk, but has a limited impact on prediction accuracy.