The purpose of this proposal is to develop statistical methodology for analyzing the process of medical diagnosis. We conceive of the process as involving three major stages. The first stage occurs when a patient presents with a clinical problem, manifested by signs, symptoms or other indications. We are required to determine, on the basis of this clinical information, a suspected diagnosis, selected from many potential diagnoses. The classical statistical method for performing this determination is usually referred to as discriminant analysis. The second stage involves assessing the properties of an available diagnostic test or tests. Measures frequently used for such an assessment are the sensitivity, specificity, ROC curve, and predictive values. The third stage involves determining whether it is cost-effective to perform the diagnostic test. (The diagnostic test under consideration may be one of several alternatives). This determination should ideally involve a decision analysis, in which the utilities of the possible test outcomes and subsequent management strategies are assesed. We plan to examine available statistical methods for each of these three stages of analysis, identify problems with the available methodology, and develop new methodology where appropriate.
Showing the most recent 10 out of 12 publications
