Biologically Informed Development of Treatment Selection Scores from High-Dimensional Omics Data

• Activity Number: 64 - Innovative Practical Improvements in Biomarker Development and Evaluation
• Type: Invited
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Lisa McShane*
• Companies: National Cancer Institute
• Keywords: omics; treatment selection; predictive model; prognostic ; high-dimensional data; cancer

Abstract

Precision medicine therapeutic approaches rely on matching mechanism of action of a therapy to biological and molecular characteristics of a patient or the patient’s disease. Therapies that can correct for aberrant or missing gene products or compensate for a disrupted biological pathway hold promise for the treatment of the corresponding disease. In oncology, many predictors based on multivariable scores generated from high dimensional omics data have been developed for purposes of prognosis, but these are not always helpful for therapy selection. The modified covariates method of Tian and colleagues (JASA 2014;109:2350-2358) is one approach that has been proposed specifically for development of therapy selection predictors. Biologically informed enhancements of the modified covariates approach that use information about biological pathways are proposed in this talk, and their performance is compared with that of the original modified covariates method on some real omics data from patients with cancer. The discussion additionally highlights some general issues regarding appropriate evaluation of treatment selection predictor performance.