Beyond expected utility: rethinking behavioral decision research.

Much research in psychology has evaluated the quality of people's decisions by comparisons with subjective expected utility (SEU) theory. This article suggests that typical arguments made for the status of utility theory as normative do not justify its use by psychologists as a standard by which to evaluate decision quality. It is argued that to evaluate decision quality, researchers need to identify those decision processes that tend to lead to desirable outcomes. It is contended that a good decision-making process must be concerned with how (and whether) decision makers evaluate potential consequences of decisions, the extent to which they accurately identify all relevant consequences, and the way in which they make final choices. Research that bears on these issues is reviewed.

Sensitivity analysis on a chance node with more than two branches.

Sensitivity analysis is an essential part of decision analysis. The literature on medical decision analysis suggests the use of two-branch chance nodes in decision trees to avoid logical inconsistencies during sensitivity analysis. The authors show that the two-branch decomposition is not appropriate for sensitivity analysis when multiple outcomes from a single state cannot be disentangled into a sensible sequence of events. They recommend retaining the natural structure of the tree and propose two sensitivity-analysis methods for use on chance nodes with three or more branches.

The clinical utility index as a practical multiattribute approach to drug development decisions.

We identify some innovative approaches to predicting overall patient benefit from investigational drugs to support development decisions. We then illustrate calculation of a probabilistic clinical utility index (CUI), an implementation of multiattribute utility that focuses on clinical attributes. We recommend use of the CUI for the support of early drug development decisions because of its practicality, reasonable accuracy, and transparency to decision makers, at stages in which financial factors that may dominate later-phase decisions are less critical.

Analysis of colorectal cancer screening regimens.

We analyze several colorectal cancer screening methods. We begin with an existing deterministic model of the colorectal cancer growth-and-development model. Using judgments from two knowledgeable experts on colorectal cancer, we incorporate probability distributions for important parameters in the model. The analysis proceeds in three phases: First is a straightforward Monte Carlo simulation that includes uncertainty about structural parameters, the results of which identify five dominant screening strategies in terms of the expected number of cancers prevented and expected cost per life-year saved. The next part of the analysis develops a two-attribute utility function to rank order the screening regimens. The results show the same top five, with the top-ranked strategy being colonoscopy every three years. Sensitivity analysis demonstrates the robustness of the results.