Constructing clinical scoring systems to determine the need for an oral glucose tolerance test

We developed a methodology for constructing scoring systems to support bivalent decision making in clinical medicine. Such systems could be of great benefit in forced-choice decision-making situations as well as in triaging, screening, and diagnostic applications. Our methodology combines medical judgment with explicit computer-derived information to identify an optimum parsimonious set of variables for predicting bivalent (2-class) outcomes. It includes a process for using these variables to produce scores for each negative (d=0) and positive (d=1) class entity and for using these scores in decision making. We produced this methodology while attempting to devise a system that determines which African Americans are most likely to be glucose intolerant (i.e. pre-diabetic) and hence would benefit by taking an oral glucose tolerance test. We achieved excellent results as evidenced by ROC plot areas of .85 for men alone and .94 for women alone. We anticipate actual clinical use of the scoring systems that we constructed. We also anticipate constructing and deploying other similar useful clinical decision support systems with our methodology.