Recent Advances in Statistical Methods for Biomarker Evaluation: Naïve Pooling Vs Umbrella Ordering

• Activity Number: 159 - Novel Approaches for Diagnostics and Prediction with Complex Data
• Type: Topic Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Lili Tian* and Yingdong Feng and Dan Wang
• Companies: SUNY at Buffalo and SUNY at Buffalo and Eli Lilly
• Keywords: ROC analysis; diagnostic accuracy; tree or umbrella ordering

Abstract

In biomedical research dealing with biomarker data, “Naïve pooling” is a common practice for combing data, and subsequently, based on pooled data, estimated diagnostic measures such as AUC are presented to assess the diagnostic ability of biomarkers. This talk will point out several severe hidden problems induced by such practice and present a new ROC framework under tree or umbrella ordering (denoted as TROC) for such data. Compared to the traditional AUC based on the pooled data, the area under TROC curve (denoted as TAUC) is a more appropriate diagnostic measure. Furthermore, issues such as optimal threshold selection will be addressed and the resulting diagnostic accuracy will be compared between naïve pooling and umbrella ordering. In the end, a published lung cancer data set will be analyzed.