This proposal seeks to develop statistical methods for medical diagnostic testing and disease screening which will allow a greater range of study designs to be used and a greater range of research questions to be addressed than can be done at present. There are four aims: 1. To develop regression modeling methods for sensitivity and specificity for dichotomous tests. 2. Robust estimation, comparison and optimization of receiver operating characteristic (ROC) curves. 3. For continuous tests,analogous regression modeling methods are proposed for ROC curves. Recently developed marginal regression modeling methods will be applied and modified as necessary. Regression modeling methodology can take account of factors which influence test accuracy as well for comparing different diagnostic tests. Marginal regression methods will accommodate clustered and unbalanced data which frequently arise in studies to evaluate diagnostic tests. 4. In the event that disease status may be unknown for a subset of study subjects, especially in screening studies, it is proposed to apply some missing data techniques to permit valid inference in such settings. In order to evaluate the proposed methodologies, each aim will require the following three steps: 1) development of large sample theory; 2) small sample simulation studies; and 3) application to real data. Four real data sets are available for this project and will guide the development of, and illustrate, the new methodologies.
Showing the most recent 10 out of 74 publications
