Recent efforts and incentives designed to contain rising health-care costs, coupled with the goal of providing high-quality medical care, underline the importance of practicing medicine in a cost-effective fashion. Several projects, involving the use of computers and mathematical techniques to manipulate available sources of data, will seek to improve the selection, sequencing, and interpretation of diagnostic tests. A theoretical model of sequential testing will be developed and implemented on computer. This will extend algebraic techniques used previously to analyze single tests. It will be applied to specific diagnostic problems, including the use of computed tomography and ultrasound in the diagnosis of adrenal disease. The cost-effectiveness of the exercise tolerance test and thallium scan, used alone or in sequence to evaluate patients with suspected coronary artery disease, will be assessed. An interactive computer program to aid the physician in the workup of patients with chest pain will be developed. Using a Markov statistical model of sequential testing, the benefits and costs of screening for cancer in patients exposed to carcinogens will be evaluated, so that guidelines for screening such patients can be established. Techniques to maximize the accuracy of diagnostic test interpretation will be developed, using chest radiography and obstetrical ultrasound as examples. Receiver operating characteristic (ROC) analysis will be used to study perceptual mechanisms underlying the effect of clinical information on chest x-ray interpretation, and to determine how such information can best be used by the radiologist. The use of mathematical models to improve predictions based on radiographic measurements will be examined, with fetal age and weight estimation from obstetrical ultrasonic measurements as a prototype. On a more general level, because medical decisions must frequently be made in the face of imprecise information, computer-implemented statistical techniques will be developed to assess how uncertainty in available data affects the confidence in decisions based on those data.
