Risky Choices in Medical Decision Making: A Survey

Abstract

Economic models of decision making under uncertainty presume five features of a decisionmaking situation. The first is a set of well-defined alternatives from which to choose. The second is uncertainty in the relation between the chosen alternative and the actual consequence. The third feature is a set preferences held by the decision maker and applied to the possible outcomes. Fourth, one or more constraints may limit the set of available alternatives. Finally, and most importantly, the decision must be made; to postpone a decision or to preserve the status quo is itself a decision with its attendant consequences. While these features are shared by many economic decisions under uncertainty, such as in purchasing insurance or in assembling an investment portfolio - and it was surely these sorts of decisions that von Neumann and Morgenstern (1947) had in mind when they developed their now classical theory of economic choice under uncertainty - these are features also shared by many decisions that arise in clinical medicine and public health. Physicians must choose between alternatives: whether to perform surgery or to administer medication, which antibiotic to use, whether to perform a risky test. Consequences of action are uncertain: Will the patient survive the operation? What disease does the patient have? What will the test show? Value judgments arise in evaluating outcomes: Is the risk of surgical mortality worth taking to relieve the pain? Are the side effects of the medicine worse than the disease? Budgetary constraints limit today's health care provider in the same way that the classical economic consumer is limited by his or her income. Finally, the decision cannot be avoided, however painful the decision-making process may be. By imposing various axioms of rational behavior on the decision maker - such as adherence to the laws of probability and transitivity of preferences - models such as the expected utility model of von Neumann and Morgenstern (1947) have been developed. These models can, and have, been used either to describe how decisions are made in fact, or to prescribe optimal decisions, that is, to help decision makers decide in a manner consistent with their own beliefs and preferences.