Comparison of the Effectiveness of Three Novel Statistical Methods for Biomarker Selection with Application to an HIV Infection Data Set

• Activity Number: 84
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Bosny J. Pierre-Louis*+
• Companies: The University of North Carolina at Chapel Hill
• Keywords: Biomarker selection ; Variable importance measure ; targeted maximum likelihood ; generalized propensity score ; incremental value estimation ; causal inference

Abstract

We consider the selection of biomarkers from a pool of candidates, based on inference from variable importance measures (VIM). We compare three approaches to estimating VIM: Targeted maximum likelihood estimation (TMLE), flexible propensity score weighting (PSW), and incremental value estimation based on partial area under the Receiver-Operating Characteristic (ROC) curve. Simulation studies suggest that TMLE and PSW outperform the incremental value technique, notably in small and moderate sample sizes. We apply these methods to a dataset from the Genital Shedding and HIV Infection (GS-HIV) study to select, among many candidate biomarkers, the best subset that is significantly associated with CD4 cell count.