New Dimension Reduction Methods for Combining Longitudinally Measured Biomarkers

• Activity Number: 278 - Combining Markers for Classification in Practical Tasks
• Type: Invited
• Date/Time: Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Ruth Pfeiffer * and Wei Wang and Efstathia Bura
• Companies: National Cancer Institute and George Washington University and Vienna University of Technology
• Keywords: Inverse regression; Kronecker product; Non-linear dimension reduction

Abstract

Sufficient dimension reduction (SDR) aims to find a low dimensional transformation of the predictors that preserves all of most of their information about the outcome. In earlier work we developed nonparametric SDR methods to combine several diagnostic markers that are measured longitudinally, using information on correlations over time and across markers (Pfeiffer et al., 2012). Here, we extend parametric SDR approaches, using least squares methods (parametric inverse regression, PIR, Bura and Cook, 2001) and maximum likelihood estimation (Principal Fitted Components, PFC, Cook and Forzani, 2008) to accommodate the longitudinal structure of multiple biomarkers measured over time and improve efficiency of the estimation of the reduction. Robustness to parametric assumptions and hte performance are studied in simulations, and we illustrate the methods with a real data example. This is joint work with Wei Wang and Efstathia Bura