Skip to Main Content

An official website of the United States government

Assessing the predictive ability of a group of ovarian biomarkers

Principal Investigator
Name
Paul Pinsky
Institution
NCI, DCP, EDRG
Email
About this CDAS Project
Study
PLCO (Learn more about this study)
Project ID
2010-0070
Initial CDAS Request Approval
May 13, 2010
Title
Assessing the predictive ability of a group of ovarian biomarkers
Summary
A widely held viewpoint in the field of predictive biomarkers for disease status holds that no single marker can provide high enough discrimination and that thus a panel of markers, combined in some type of algorithm, will be needed. Motivated by an experimental example where 26 additional markers, many of which had good predictive value alone, failed to substantially increase the predictive ability of the primary marker, we explore the effect of additional markers on the area under the ROC curve (AUC). Specifically, we assume (log) markers follow a multivariate normal (MVN) distribution in cases and in controls and that the predictive algorithm takes the form of a linear combination of (log) markers. We derive conditions under which additional markers will or will not contribute extra predictive ability. In addition, we examine the effect of measurement error on the predictive ability of algorithms.
Aims

1. To explore, using a multivariate normal model, how the inclusion of additional markers may or may not increase the predictive ability of an algorithm in the form of a linear combination of markers. We will derive theoretical results and then show how these apply to 26 ovarian markers assayed from PLCO ovarian cases and controls. 2. To explore the effect of measurement error on the predictive ability of an algorithm. Using duplicate pairs to estimate measurement error, we will develop a methodology to estimate the effect of measurement errors on the ROC AUC of an algorithm. We will use the above 26 ovarian marker values and corresponding algorithms to illustrate the method.

Related Publications