A Comprehensive Evaluation Of Breast Cancer Risk Prediction Using Uk Biobank Data.
This thesis is restricted to Yale network users only. It will be made publicly available on 08/28/2020
Introduction: Both genetic and clinical risk factors play roles in breast cancer onset. Developing comprehensive and accurate breast cancer prediction model will enable effective identifications of individuals at high risk, and facilitate personalized decision on screening and preventive strategy.
Objectives: Develop a risk prediction model for breast cancer including both genetic and clinical risk factors, and quantify the benefit of adding polygenic risk score (PRS) in risk prediction.
Methods: Analysis was based on data collected through the UK Biobank which consist of 13,851 breast cancer cases and 206,865 controls. The clinical factors that we considered include baseline demographics, lifestyles, family history, reproductive history, medication status and operation history. The PRS for breast cancer was constructed from 2,994,056 single nucleotide polymorphisms (SNPs) using the AnnoPred approach. The area under the curve (AUC) was used to evaluate the performance of different prediction models. The cumulative risk of breast cancer was compared between participants in different risk groups.
Results: Breast cancer risk prediction based on AnnoPred derived PRS had a comparable prediction accuracy (AUC=0.646, 95%CI 0.642-0.651) with that based on all the 19 clinical factors (AUC=0.657, 95%CI 0.652-0.662). Combining PRS and clinical factors further improved the prediction accuracy (AUC=0.708, 95%CI 0.704-0.713). Based on the combined model, the estimated lifetime risk of developing breast cancer up to age 70 among individuals in the top 1% risk group (40.1%) was more than 28-folds higher than that in the bottom 1% risk group (1.4%).
Conclusion: Breast cancer risk prediction based on genetic factors only can achieve comparable performance compared to that using well established risk factors, demonstrating the significant progress that has been made towards breast cancer genetics. It is important to include PRS in deriving risk prediction for personalized breast cancer screening and prevention.