Strategies for Identifying Predictive Biomarkers in Clinical Trials Using Variable Importance

• Activity Number: 300
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Ilya Lipkovich*+ and Alex Dmitrienko
• Companies: Quintiles, Inc. and Quintiles, Inc.
• Keywords: SIDES method ; recursive partitioning ; variable importance ; predictive biomarkers ; subgroup identification

Abstract

Several approaches were proposed recently for identification of predictive biomarkers and subgroups of patients with enhanced treatment effect. SIDES method (subgroup identification based on differential effect search) introduced in Lipkovich et al. 2011 adopts recursive partitioning algorithms for screening treatment-by-covariate interactions. In this talk we propose to enhance the subgroup search with several two-stage subgroup identification procedures that first identify few biomarkers with largest variable importance and then identify subgroups using a smaller set of selected covariates. Variable importance allows for screening out noise covariates by averaging over a large number of subgroups thus taking advantage of sparseness of noise covariates. We compared three strategies via a simulation study: (i) applying SIDES method to the initial covariate space, (ii) identifying covariates with largest variable importance first and identifying final subgroups by applying SIDES to the reduced covariate space, (iii) same as (ii) but biomarker screening and subgroup identification are conducted on two independent data sets. The methods are illustrated using a clinical trial example