Subgroup Identification Based on Multiple Outcomes

• Activity Number: 641
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Chensheng Kuang* and Menggang Yu and Sijian Wang
• Companies: University of Wisconsin - Madison and University of Wisconsin - Madison and University of Wisconsin - Madison
• Keywords: Subgroup Identification ; Boosting ; Multivariate regression

Abstract

We propose a novel multivaraite component-wise boosting method for in subgroup identification based on multiple outcomes. Our method is motivated by the situation when multiple outcomes share common influential features in subgroup identification and hence an integrated analysis of all outcomes would provide more power in variable selection and better performance in prediction. The proposed method is based on the contrast classification framework in subgroup identification (Zhang 2012) and multivariate boosting (Xiong 2015), which is built on the well known gradient boosting machine (Friedman 2001). We adapt the original multivariate boosting to incorporate to weights and to binary outcomes and also extend it to other loss functions than the squared loss. The updated algorithm is a highly flexible approach by the choice of loss functions and base learners. Moreover, it is a general variable selection approach that is capable of dealing with high dimension of both covariates and responses. The performance of the method is evaluated through simulation studies.