Talks

An Exact Interval Estimation on Paired Areas under the ROC curves with Multiple Markers

130
reads

2008 - 01 - 29 (Tue.)
13:30 - 15:00
405, Mathematics Research Center Building (ori. New Math. Bldg.)

Microarray is one of popular biotechnologies that enable researchers to quickly screen large numbers of biological analytes for a variety of purposes, from disease diagnosis to detection of bioterrorism agents. In breast cancer study, Swain (2006) pointed out that the based on 70-gene MINDACT (Microarray In Node negative Disease may Avoid ChemoTherapy) is an important implication for the future individualized treatments for thousands of patients. On the other hand, the 21-gene Oncotype DX is the first available breast cancer assays. The area under the ROC curve (AUC) is a useful index to evaluate the accuracy of diagnostic device. Su and Liu (1993) proposed the best linear combination of multiple markers that maximizes the AUC. In this talk, we take our attention to accuracy comparison for paired diagnostic devices like 70-gene MINDACT and 21-gene Oncotype DX. We apply the concept of generalized confidence intervals presented by Weerahandi (1993) to develop the interval estimation for the difference of paired AUCs.