Assessing the Reproducibility and Value of Genomic Signatures

• Activity Number: 343
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Prasad Patil* and Jeffrey T. Leek
• Companies: The Johns Hopkins University and Johns Hopkins Bloomberg School of Public Health
• Keywords: genomics ; signature ; replicability ; value

Abstract

A gene signature forms the basis for many laboratory tests designed to take a patient's genomic information into account in the course of treatment. Popular tests such as Mammaprint, Oncotype DX, and Prosigna rely on modeled genomic information to make assessments of future risk of cancer recurrence or cancer severity categorization. Many factors can affect the replicability of predictions made by these models and of the signature itself. We examine these issues and propose approaches to mitigate them. We also explore scenarios where the clinical value of a genomic signature can be calculated. We are especially interested in the benefit a genomic test can provide beyond the standard clinical quantities that are gathered for all patients when they appear in the clinic. We describe a method that allows for the examination of predictive covariates in a clinical trial setting and apply it in an attempt to tease out the relative clinical value of a gene signature.