Threshold Analysis of Decision Tables

Abstract

When a decision table is used to find a maximum expected utility testing strategy, it is based on a given prior probability distribution of diseases. In the two-disease situation, a threshold analysis over all prior probabilities can be done using threshold transformations of the points of indifference between treatments. This results in a set of prior probability intervals each with its own unique decision rule. The Boolean expression for the table indicates the ac ceptable testing strategies. A decision table analysis may then be extended to include invasive or costly investigations. The technique represents a saving in time and effort com pared with standard decision tree approaches, especially where investigative recommen dations are to be made for a broad range of prior probabilities, e.g., where initial symptoms and signs are considered before the investigations.