Choice Theories

Decision Analysis

Medical decision making relies heavily on rational choice theory. One common way of analyzing treatment choices, decision analysis, employs the same structure. Constructing a decision tree requires specifying the possible actions (“choice nodes”), specifying the possible outcomes of each action (“chance nodes”), attaching probabilities to each outcome (which must sum to 1.0), and then affixing costs and consequences to each outcome. The tree is then “folded back” by computing the expected value at each node by multiplying the probability by the costs and by the consequences.

For example, in their five-part primer, Medical Decision Analysis, Allan Detsky and colleagues work through the example of how to model the choice of management strategies for patients presenting with clinical features that suggest giant cell arteritis (GCA). In this simplified model, the only treatment considered is treating with steroids, which can involve side effects. The rational model they employ thus involves a choice between three possible actions at the choice node—treating, not treating, and testing and treating only if the test result is positive. The possible outcomes can be simplified to four possibilities, depending on whether or not there was an adverse outcome as a result of the disease (in that case, blindness), and whether or not the person had side effects as a result of the treatment. Note that some of these outcomes cannot occur on some branches—for example, someone who did not receive treatment could not experience any outcomes involving side effects. The next step for the decision maker is to determine how likely each of these possible outcomes would be at each choice node (e.g., how likely would an untreated individual with those symptoms be to experience blindness if the person was not treated). Next, the decision maker would affix costs and utilities to each possible outcome. For example, these papers assigned a value of 1.0 to the state with no disease and no side effects, and a value of .5 to the state of having the disease without treatment (or side effects) but ending up with blindness. Sensitivity analysis can be used to modify these values (e.g., change the probability of adverse outcomes or the value attached to particular outcomes) and see how much they affect the resulting choices.

...