The health-economic models: practical aspects and management of uncertainty

Abstract

Analytic models are a powerful instrument to develop pharmacoeconomic analyses and their importance is growing as they are being increasingly used to make predictions of the consequences of a particular intervention. It is possible to group the most commonly used techniques in three families: decision trees, Markov chains and probabilistic simulation models. Only the last ones take into account a wide range of uncertainties and have the capability to make probabilistic predictions. Discrete-state, discrete-time Markov models are the most used technique, but have some limits due to their structural rigidity that can make appropriate representation of clinical reality difficult. First-order simulation of Markov models produces deterministic results and can be conveniently implemented in a matrix algebra formal framework. In order to take decision based on models prediction deterministic results are not sufficient and it is widely recognized the need to handle uncertainty in its various forms. The task could be accomplished with traditional (deterministic) and/or probabilistic sensitivity analysis. Both analyses provide complementary information on how the parameters and assumptions uncertainty spreads trough the model and are recommended by ISPOR (International Society for Pharmacoeconomics and Outcomes Research) modelling guidelines.