There is an urgent need to better understand the value of diagnostic tests. Their use is critical to timely and appropriate selection of medical treatment. This project investigates potential test interpretation aids based on partial knowledge about the statistical relations between serial tests. One method is based upon a concept to be called contingent information display. This involves using mathematical transformations of test patterns to remove redundancies, contingent on the diagnostic hypotheses being considered. One experiment to test this would would randomly assign physicians to one of three experimental groups, given different displays of serial creatine phosphokinase (CPK) assays to detect reinfarction or extension of an acute myocardial infarction. The hypothesis is that physicians' predictions of extension will be improved with transformed displays. A second method attempts to improve the interpretation of serial tests by clarifying the relevance of pattern features rather than by clarifying their display. An experiment will test whether logical explanations of statistical parameters describing these features improve the assessment of serial carcinoembryonic antigen (CEA) assays to detect a recurrence of colorectal cancer. Various other applications (multiple serum enzymes in myocardial infarction, radionuclide scan patterns) also are planned.
