Simulating medical decision trees with random variable parameters

This paper describes many of the issues that arise when incorporating the distribution of the uncertainty of the parameters into a medical decision tree. Most importantly, two different formulations of simulations of the tree yield two very differently distributed measures of the value of a tree. One measure is transaction-based reflecting the perspective of an individual patient, while the other provides the distribution of the global average value of the tree. Additionally, much care must be taken to represent duplication or other forms of dependence of distributions properly when calculating the tree average of stochastic trees. Finally, a practical example of a decision tree with random variable parameters, comparing the cost-effectiveness of using a new imaging agent to existing post-myocardial infarction testing protocols is presented.