Medical Decision Making

Expected utility theory, and the Bayesian probability theory on which it is based, form the normative basis of most work in medical decision analysis. Recent work in the psychology of judgments and decisions indicates that people do not conform to the axioms of this theory and that these deviations occur in clinical reasoning as well as in the psychology laboratory. At issue is what to do now. The authors argue that the important next steps lie at the interface between descriptive, prescriptive, and normative accounts, all of which affect each other. They point to examples in which the simplest application of supposedly normative theory seems inappropriate, and suggest ways in which the tension between normative and de scriptive models may be resolved. Key words: decision analysis; expected utility theory; clinical reasoning; cognitive processes. (Med Decis Making 7:203-211, 1987)

When considering two dichotomous tests in combination for reaching a treatment decision, the choice between single and multiple testing depends, in part, on the pretest probability of disease. The authors show that two tests are never preferred to a single test for all disease probabilities, regardless of whether the two tests are performed in parallel or in series. Key words: decision analytic model; tests with dichotomous outcomes; parallel testing; series testing. (Med Decis Making 7:212-219, 1987)

The authors review the probability threshold approach to test/treatment decisions developed by Pauker and Kassirer, ^6,7 emphasizing that certain aspects of the nature of medical decisions call for a new approach. The utility threshold approach, while maintaining all the advantages of threshold methods in general, brings improvements. It diminishes the need to accurately assess one of the decision's parameters: the patient's utility for the outcome states. For a simple case of one disease with three outcome states (cured, diseased, dead) and one test, three utility thresholds are derived. The treat/no treat threshold, denoted by u*, separates the utility space in two. If the patient's value for the diseased state is greater than u*, the analyst can feel confident in recommending the patient forego treatment. Similar interpre tations are developed for u₁, the no treatment/test utility threshold (the value u must take, given a positive test result, for the patient to be indifferent between foregoing and receiving treatment), and u₂, the test/treatment utility threshold (the value u must take, given a negative test result, for the patient to be indifferent between foregoing and receiving treatment). Key words: medical decision analysis; threshold analysis; utility threshold. (Med Decis Making 7:220-233, 1987)

The receiver operating characteristic (ROC) curve was applied to observer performances in a multiple-alternative decision task. It was shown that the probability of correct classification, a performance criterion often maximized in multiple-classification procedures, corresponds to the area under an appropriately constructed ROC curve. Degrees of confidence in the observer's judgment of 0, 1, ... , 10 were used for both classification and ROC rating. To demonstrate the validity of the method, 1,190 photofluorograms were examined by experi enced staff radiologists to identify four cardiovascular conditions distinguishable on the basis of images of structural elements of the contours of the heart and great vessels. The clas sification matrices for three radiologists who achieved high, medium, and low performance ratings in this experiment are reported. The ROC curves are symmetric, with their points located around the off-diagonal. Differences between the overall probability of correct clas sification and the ROC curve index calculated from the same evaluator's data were very small, 0.004 to 0.011. Key words: receiver operating characteristic (ROC) curve; multiple classification; medical decision making; cardiovascular diagnostics; thorax, radiography. (Med Decis Making 7:234-237, 1987)

This paper describes a simple method for constructing receiver operating characteristic (ROC) curves based on boundary conditions that restrict its theoretical shape. The resultant curve is identical to that for logistic distributions with equal variance. Although the underlying assumptions are highly restrictive, this method nevertheless possesses several attractive properties which are illustrated by its application to 2,567 patients undergoing a variety of noninvasive diagnostic tests for coronary artery disease: 1) the area under the curve is defined by a single parameter; 2) the associated standard deviation is analytically defined and readily computed; 3) the curve is completely invariant with respect to selection bias. Key words: coronary artery disease; diagnostic tests; ROC analysis; ROC curves; selection bias. (Med Decis Making 7:238-243, 1987)