Selecting Treatment Based on Generated Effect Modifiers

• Activity Number: 638
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Eva Petkova*+ and Thaddeus Tarpey and R. Todd Ogden and Zhe Su
• Companies: New York University School of Medicine and Wright State University and Columbia University and New York University
• Keywords: biosignitures ; treatment selection ; moderator of treatment effect

Abstract

This talk will introduce a new algorithm for combining predictor in a way that makes the combination a strong modifier of treatment effects. The generated effect modifier can be used for judging whether a given patient would do better on one treatment or another. An alternative approach for deciding what treatment to be given when several treatment options are available, is to develop models for predicting outcome under each of the treatment options based on baseline patient characteristics. These models can then be employed to predict the outcomes for a specific subject under different treatments and a treatment choice can be made based these predictions. We compare the two approaches under various conditions that are common in depression research, including the case of large set of complex pre-treatment patient characteristics that can be used as predictors of treatment response.