Accounting for Uncertainty in Cost-Effectiveness Studies

Abstract

Due to an increasing demand from decision makers for proper economic evaluations of health care services, cost-effectiveness analyses are becoming increasingly frequent. The statistic of interest in cost-effectiveness analysis is the incremental cost effectiveness ratio (ICER). When patient-specific data on costs and effects of alternative interventions is available, it can be used to quantify the uncertainty in the estimate of the ICER. Expressing this uncertainty by using confidence intervals has been recommended. However, because the statistic of interest is a ratio of two correlated random variables, its variance cannot be estimated exactly. Furthermore, the distribution of the ratio is unknown. Recently, several approximate methods have been proposed for calculating confidence intervals for the incremental cost-effectiveness ratio. These include two parametric methods: one which relies on a Taylor's Series approximation of the variance, and one based on Fieller's theorem; as well as a number of methods which rely on bootstrapping methodology. In this manuscript, these methods were applied to data obtained from a randomized clinical trial in which both health resources consumed and health outcomes were observed. Furthermore, several variations of the bootstrapping methods were proposed and applied to the same data set. Probabilities of the true ICER being in given ranges were also estimated using a bootstrapping approach. Finally, issues of sample size and power were briefly considered. The relative advantages and disadvantages of the different approaches were discussed.