Quantitative Issues in Cancer Research Working Seminar

Jane Liang
Doctoral Student, Department of Biostatistics, Harvard University

“Aggregating Across Genes and Cancers in Mendelian Risk Prediction Modeling”

ABSTRACT: Identifying individuals whose risk of cancer is increased as a result of carrying heritable pathogenic mutations is important for clinical management and research. Using principles of Mendelian genetics, Bayesian probability theory, and mutation-specific knowledge, Mendelian models identify those at high risk for carrying a pathogenic mutation and assess future risk of cancer, based on family history. These quantitative risk measures can then be used to tailor personalized prevention programs. PanelPRO is a generalizable, computationally efficient Mendelian risk prediction model that incorporates an arbitrary number of gene-cancer associations. However, there are pragmatic challenges in the implementation of such a comprehensive model with many genes and cancers. There may be uncertainty in estimating the necessary population-level model parameters (prevalences and penetrances) among rare genes and cancers. Obtaining accurate and comprehensive patient family history information about a large number of cancers may also be impractical for some clinical settings. To help overcome uncertainties in the user-specified family history and model parameters, we investigate simplifying modeling assumptions that reduce the amount of patient information that needs to be collected and allow for more robust parameter estimation. These aggregation approaches are evaluated through simulations to determine the trade-offs between simplicity and accuracy.