Composite Score-Based Decision Rule as a Predictive Signature for Patient Subgroup Identification

• Activity Number: 550
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xin Huang* and Yan Sun and Lu Tian and Viswanath Devanarayan
• Companies: and AbbVie and Stanford University and AbbVie
• Keywords: Multivariate ; Variable selection ; Personalized Medicine ; Biomarkers ; Nonparametric

Abstract

Development of predictive signatures for identifying patient subgroups that benefit more from a certain therapeutic is an important first step towards personalized medicine. When multiple candidate biomarkers are available, most predictive signature development methods in the literature force strict cut-points on each biomarker, thus potentially limiting their full use for predicting treatment response (Lipkovich et al., 2011, 2014, Tian et al., 2010, Chen et al., 2014). Utilizing all the data from important biomarkers by combining them first and then applying a cut-point to the final composite score may improve the discriminatory ability for identifying patient subgroups with promising treatment effect. In this research, we generalize the method developed for prognostic applications for binary responses (Huang et al., 2011) to derive optimal linear combination that maximizes the difference of the area between the receiver operating characteristic curves of the treatment and control groups without assuming any parametric model. The performance of this proposed method is evaluated and compared to existing methods via simulations and real clinical trial data.