Date(s) : 16/01/2017 iCal
14 h 00 min - 15 h 00 min
Objectives: Firstly, we recall how to handle uncertainty in medico-economic evaluations in general, in the aim of providing a reliable decision-making in terms of allocating resources. In particular, the method we developed to build confidence regions around the Incremental Cost-Effectiveness Ratio, solving three problems simultaneously is presented: 1) the mathematical problem of instabilities of some methods when the denominator of the ratio approaches zero statistically, 2) the “mirror decision-making” problem where two opposite ratios provide the same decision, and finally 3) Interpretation in terms of decision-making of confidence regions having non standard form with Fieller’s method.
Actually, the cost-utility analyses (CUA) are internationally recommended by the National Institute for Health and Care Excellence. Utility measure accounts for patient preferences and their quality of life by measuring Quality Adjusted Life Years, which are gained years multiplying by utility or preference scores. This makes more complex the handling of uncertainty.
In addition, in CUA, utility values are rarely available and they are generally predicted using a “mapping” interpolation from a functional status questionnaire. This mapping method is not accounted for in pharmaceutical industry and in literature studies, when building confidence regions around the incremental cost-utility ratio, leading to a wrong confidence region and consequently, to a wrong decision-making. Thus, the purpose of this research is to build a confidence region around the Incremental Cost-Utility Ratio, accounting for the uncertainty coming from the “mapping” interpolation.
Methods: Analytical, parametric and nonparametric Bootstrap methods are developed to handle the fact that utility values are interpolated. Linear, multilinear, and nonlinear mapping are considered and compared to a “naïve” method, used in practice, not accounting for mapping. Monte Carlo experiments are carried out to compare the performance of these various methods, which are then applied on data issued from a clinical trial about hepatitis C treatment, measuring the impact of therapeutic education. Utility values are assessed from a SF-12 questionnaire and some of these values are interpolated from the Nottingham Health Profile functional status questionnaire.
Results: Monte Carlo experiments show that the analytic and bootstrap 95% CI display coverage between 94% and 96% for various sample sizes. If mapping is not accounted for (“naive method”), the coverage is between 61% and 95%. The cross validation shows similar results.
Conclusion: In CUA, decision-making based on utility values interpolated from mapping is not reliable and the uncertainty due to mapping has to be accounted for. Our analytic and bootstrap procedures, integrating the mapping, provide very accurate results.