Targeted Methods for Biomarker Discovery: The Search for a Standard

• Activity Number: 20
• Type: Topic Contributed
• Date/Time: Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Catherine Tuglus*+ and Mark J. van der Laan
• Companies: University of California, Berkeley and University of California, Berkeley
• Keywords: Biomarker ; Causal Inference ; Robust ; Targeted Maximum Likelihood

Abstract

Biomarker detection methods aimed at identifying causally related biomarkers often result in measures that are difficult to interpret and nearly impossible to compare across studies. In response to the FDA call for method regulation, we present targeted variable importance (tVIM) as a standardized method for biomarker discovery. Based on targeted maximum likelihood theory, these estimates are robust to model misspecification and, under specified conditions, are interpretable as causal effects and reproducible across populations. The ability of the tVIM method to detect "true" relevant variables is remarkably resilient under increasing correlation conditions, outperforming univariate linear regression and LASSO penalized multiple regression in simulation. Applying tVIM to leukemia shows an increased ability to detect relevant genes over other methods.