The Optimal Thresholds of Biomarkers by Maximizing Various Metrics

• Activity Number: 522
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Kelly Zou*+ and Martin M Carlsson and Kezhen Liu and Ching-Ray Yu
• Companies: Pfizer Inc. and Pfizer Inc. and State University of New Jersey and Pfizer Inc.
• Keywords: Optimal Threshold ; Youden's Index ; Kappa Statistic ; Euclidian Distance ; Mutual Information ; Biomarker

Abstract

Using biomarkers, optimal thresholds are typically determined as cutoff points to dichotomize each biomarker. Depending on different objectives, for example, either to minimize the classification error or to improve the agreement between the biomarker and a binary gold standard, such thresholds may be obtained accordingly. In this research, four metrics, i.e., Youden's index, kappa statistic, Euclidian distance, and mutual information, are individually maximized in order to derive the corresponding optimal threshold. These metrics are pre-defined for the two-sample problem against the gold standard. The gold standard variable is a binary indicator for disease status, cancer stage, or post-treatment outcome. Mathematical solutions to these optimal thresholds are obtained using these various objective functions under parametric modeling assumptions. Monte-Carlo simulation studies are conducted to compare the performances of these different methods. In addition, two previously-published datasets are used to illustrate and compare the estimated thresholds both empirically and parametrically.