Generating Treatment Effect Modifiers from Complex Data Modalities

• Activity Number: 303
• Type: Topic Contributed
• Date/Time: Tuesday, August 2, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Mental Health Statistics Section
• Author(s): Thaddeus Tarpey* and Eva Petkova and Robert Todd Ogden
• Companies: Wright State University and New York University and Columbia University
• Keywords: functional data ; precision medicine ; index models

Abstract

A growing challenge in precision medicine is to combine complex baseline data modalities into powerful biosignatures that can inform optimal treatment decisions. Closely related is the problem of how to construct biosignatures that are capable of distinguishing a specific drug response from a placebo response. This is an issue of critical importance in mental health research where there are often high placebo response rates. One approach when using scalar predictors is to consider a biosignature defined as a linear combination of variables. However, many modern modalities are functional in nature, such as EEG signals and brain scans using fMRI data. This talk will discuss approaches to combining these complex modalities to form powerful biosignatures for treatment decisions as well as distinguishing placebo response from specific drug response. Data from a large clinical trial for treating depression will be used to illustrate the ideas.