Quantification of health states with rank-based nonmetric multidimensional scaling.

摘要

OBJECTIVES: An alternative methodology is introduced to generate cardinal valuations of health states. This methodology is based on the ranking of differences between health states combined with an associated scaling model that transforms the individual rank data into group values on the interval level. METHODS: Data were collected in a Dutch EuroQol EQ-5D valuation study, in which a representative sample (n = 212) of the Dutch population valued a set of 18 EQ-5D health states and death. Three computational steps were undertaken: 1) differences in visual analog scale (VAS) values were computed for each pair of health states based on individual data; 2) the rank ordering of these pairwise differences was derived; 3) nonmetric multidimensional scaling was used to recover cardinal scale values for each state based on these rankings of differences. RESULTS: Scaling of ranked differences between health states using multidimensional scaling produced cardinal values that were nearly identical to the mean VAS valuations. The rank-based values explained 98% of the variance in the VAS values. CONCLUSION: Ordinal data collection techniques, combined with scaling models, may offer an attractive alternative to direct cardinal elicitation methods for valuing health states.