Description

"Statistical Methods for Biomarker Discovery."

Tatsuki Koyama, Ph.D., Assistant Professor of Biostatistics

Abstract:
Advancement in quantifying biomarker expression levels in recent years has led to the development of numerous predictive models in a wide range of medical fields. Logistic regression models remain very popular for diagnosis and prognosis of disease. Comparisons of predictive models based on area under receiver operating characteristic (ROC) curves are popular but known to have low power. Analyses based on reclassification tables [Cook 2007] suffer from conceptual problems. Recently, the integrated discrimination improvement measure [Pencina et al. 2008] is gaining popularity. In this talk, some aspects of diagnostic / prognostic model building and evaluation will be reviewed with emphasis on select topics.

Topics:

1. Unnecessary categorization/dichotomization of a continuous variable
2. Sensitivity and specificity
3. Receiver operating characteristic (ROC) curve and area under ROC curves (AUC)
4. Reclassification tables
5. Integrated discrimination improvement (IDI)

Goals:

1. To be familiar with some fundamental concepts in diagnostic/prognostic model building
2. To be able to critically review research presenting new predictive models

Intended Audience:
Anyone who is interested in or involved in statistical model building. Basic knowledge of statistics is assumed.

Speaker Description:
Dr. Koyama is an assistant professor of biostatistics in the Cancer Biostatistics Center at Vanderbilt-Ingram Comprehensive Cancer Center. He received his Ph.D. in statistics from the University of Pittsburgh, and he has been a faculty member at Vanderbilt University School of Medicine since 2003. He has extensive methodology research experience in the field of adaptive and flexible designs for clinical trials. His research interests also include survival data analysis and statistical graphics and their proper application to biological and medical data.

Beverages and snacks will be provided.