Accurate Statistical Method for Diagnostic Testing Suited for Clustered Data

Document Type

Presentation Abstract

Presentation Date

3-23-2009

Abstract

Consider a comparison of two diagnostic tests for a breast cancer -- Computer Aided Diagnostic (CAD) versus without CAD. Data are typically collected from both breasts of a sample of women. To reduce the effect of operator to operator variability, the tests are conducted by three medical experts for each individual. Classical methods for comparing diagnostic tests would assume the data from the two breasts and by the three experts on the same subject to be independent. That is, the within subject correlation is not accounted for and this undoubtedly results in a low efficiency. Other more recent methods that are designed for clustered data make some restrictive assumptions which may not be plausible for some data. In the last couple of years, methods that do not make such assumption have been devised but they have some technical limitations as a result of which their accuracy suffers.

In this talk, we will discuss the most popular criteria for assessing accuracy of a diagnostic test and briefly review some of the existing methods for comparing such tests. A new method which is more robust in terms of its accuracy than existing methods will be presented. The breast cancer example mentioned above will serve to explain the main ideas in the talk. The talk is based on years of experience of the speaker as a medical statistics consultant in the department of Medical Statistics at the University of Gottingen and his recent research as part of his PhD Thesis.

Additional Details

Frank Konietschke is a senior PhD student in Applied Statistics and Empirical Methods at the University of Gottingen and is currently on a research visit in the department of Mathematical Sciences at The University of Montana.

Monday, 23 March 2009
3:10 p.m. in Math 103
4:00 p.m. Refreshments in Math Lounge 109

This document is currently not available here.

Share

COinS