Accounting for Skewness with a Nonparametric Estimate of the Incremental Cost Effectiveness Ratio

• Activity Number: 572
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): David K. Blough*+
• Companies: University of Washington
• Keywords: bivariate location ; skewness ; incremental cost effectiveness ratio

Abstract

Typically, mean costs and mean measures of effectiveness are used in the estimation of the incremental cost effectiveness ratio (ICER). However, cost and effectiveness distributions are often skewed. The goal of this work is to use a nonparametric bivariate generalization of location to obtain a location region in the cost effectiveness plane. Analogous to Mann-Whitney, all pairwise cost and effectiveness treatment differences are computed. Using univariate methods, an estimate of a location rectangle is obtained as the Cartesian product of marginal location intervals, intervals that account for skewness. By considering the intersection of such rectangles over all orthogonal transformations, a closed convex region is obtained. Kolmogorov-Smirnov provides simultaneous confidence bands for each bivariate rectangle. Any point of this region is an estimate of the ICER; all such points satisfy basic axioms of bivariate location. Comparing the location region with the usual bootstrap estimates for the ICER, it was found that the location region approach produced confidence intervals that were considerably larger for both skewed and symmetric distributions of costs and effectiveness.