The handling of uncertainty in economic evaluations of health care strategies

BACKGROUND: The recent development of prospective economic evaluations in association with randomized controlled clinical trials has provided clinical and economic data allowing a statistical analysis of uncertainty. Pertinent comparison of alternative medical strategies in the cost-effectiveness analysis requires analysis of the incremental cost-effectiveness ratio. In this work, we were interested in developing confidence regions of the cost-effectiveness ratio in order to take into consideration its uncertainty which creates a certain number of problems when the denominator approaches zero. Our objectives were to conduct a critical analysis of the different estimations used and examine their application in the decision making process. METHODS: There are two kinds of methods: those based on the density of the estimated ratio (Taylor method, parametric and non-parametric bootstrap methods such as percentile, percentile-t and bias-corrected and accelerated methods) and those based on the bivariate density of the variable pair: difference of average costs, difference of average effects (box method, ellipse method, and Fieller method). RESULTS: We showed that methods based on the density of the ratio become unstable and inapplicable when the statistical difference of average effects tends towards zero. In practice however, data often take on such a pattern. We give an example of an economic evaluation as an illustration. Bivariate density methods do not have this drawback. Moreover, the "mirror decision" phenomenon requires that the region of confidence be a directed single confidence sector for valid decision making. CONCLUSIONS: Uncertainty must be taken into account for all economic evaluations but the use of undirected confidence regions is inappropriate as a decision-making tool.