Modeling sequential decision-making in pharmacoeconomics

Abstract

Most decision problems in pharmacoeconomics involve sequential decision-making, where the initial decision is followed by outcomes that affect subsequent decisions. However, future decisions are rarely explicitly modeled in most pharmacoeconomic studies. In this workshop we examine the problems associated with modeling future decisions as chance events, and describe the advantages of explicitly incorporating these future decisions in the decision model. We start by examining the different classes of decision problems, including decision-making under certainty, uncertainty, and risk. We describe how decision trees are an ideal framework for representing sequential decision-making problems. We then use an example to illustrate how modeling future decisions as chance nodes can lead to suboptimal decisions. We also use the example to illustrate how failure to explicitly model future decisions can lead to erroneous results during sensitivity analysis. We describe how models that explicitly incorporate future decisions can be used to develop optimal treatment pathways. Further, we argue that we cannot optimize the initial treatment decision without identifying the optimal treatment pathway. We explore possible reasons why future decisions are often represented as chance nodes in pharmacoeconomic models. Finally, we extend the discussion to Markov models. We show how future decisions can be modeled through the use of Markov decision models, which identify the optimal treatment strategy for each health state. We will conclude the workshop with an interactive discussion of the benefits and drawbacks of explicitly modeling future decisions in pharmacoeconomic models.