A Nonparametric Bayesian Approach to Biomarker Discovery

• Activity Number: 497
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Peter Mueller*+ and Alejandro Jara
• Companies: MD Anderson Cancer Center and Pontificia Universidad Católica de Chile
• Keywords: screening test ; dirichlet process ; nonparametric Bayes ; biostatistics

Abstract

We develop a nonparametric Bayesian model to implement retrospective validation and refinement of a biomarker panel.

We focus on two inference problems, (i) the specification of a screening test based on a panel of multiple biomarkers; and (ii) the assessment of operating characteristics summarized in an ROC curve. The motivating application is the development of a screening test for high risk ovarian cancer patients on the basis of a panel of 7 potential markers. The desired biomarker development is usually carried out in a multi-step process, using independent studies for steps (i) and (ii). Exploiting the flexible nature of non-parametric models allows a more honest representation of uncertainty about the distribution of biomarkers in the target population. This makes it possible to attempt joint inference for both steps in a single study. We use a DP mixture model.