Generated Effect Modifiers in Linear and Nonparametric Models

• Activity Number: 270 - Intersection of Econometrics and Biometrics in Making Policy and Treatment Determinations
• Type: Invited
• Date/Time: Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economic Statistics Section
• Author(s): Eva Petkova* and Thaddeus Tarpey and Hyung Park and Zhe Su and R. Todd Ogden
• Companies: New York University School of Medicine and Wright State University and Columbia University and New York University School of Medicine and Department of Biostatistics, Columbia University
• Keywords: Single index model ; Moderator of treatment effect ; Treatment decisions rule ; Antidepressant treatment

Abstract

In a randomized clinical trial (RCT), the interest is not only to estimate the treatment, but also to determine if any patient characteristic has a different relationship with the outcome, depending on treatment. In regression models, if there is a non-zero interaction treatment-by-predictor, the predictor is called an "effect modifier". Identification of such effect modifiers is crucial for optimizing individual treatment assignment. In most settings, there will be several baseline predictor variables that could potentially modify the treatment effects. We present a method of constructing a composite variable to generate a strong effect modifier in an RCT setting. This Generated Effect Modifier (GEM) approach represents a parsimonious alternative to existing methods for developing treatment decision rules. While the meaning of a moderator of treatment effect is well understood when the outcome is linearly related to a predictor, it is less obvious when they related nonlinearly. A GEM for a flexible nonparametric model is presented also. We illustrate using data from a randomized clinical trial designed to discover biosignatuires for treatment response to antidepressants.