Using bayesian techniques to build up an inconsistency free health status index

Abstract

Objectives: 1-To obtain a set of values of health states of the EQ-5D based on self-related health VAS using linear and non- linear models Bayesian techniques. 2 - To analyse “logical consistency” in different models and to derive a model free from logical inconsistencies. 3 - To analyse and compare results of several models when using a priori sources of information. Methods: We apply the usual models and transformations of these, in order to attain logical consistency of the value set. Models proposed are: linear model (1); linear with dummy variables (2) and two models with a logistic structure with different distributions of the coefficients to be estimated (3 and 4). For two of these models new dummies are added in order to obtain logical consistency (2B and 4B). Results: We propose a modelling to guarantee consistency of values of the EQ-5D health states that may be applied to suitable samples at apparently low cost of fit. This model is nonlinear, has distribution Gamma in the coefficients and specific dummy variables. The introduction of priors may reduce the cost of forcing logical consistency.