Scale validation of EQ-5D - A statistical analysis of quality of life based on a clinical trial

Health related quality of life is brought more and more into focus in the huge growing health sector. This implies an increased interest from the participants in the health sector eg. medical personal, the industry, researchers etc. This work is based on a clinical study of wound products from the Danish medico company Coloplast A/S. The clinical study is a large post-marketing study with the main purpose of comparing a product with the standard care on the market. It is presumed that the Coloplast product is more e®ective than the standard care products on the market. This work should lead to a reliable scale based on the EQ i 5D which makes it possible to perform an analysis of this presumption. A scale validation of EQ i 5D is a stepwise process of going from the five items (questions) to an estimation of a latent distribution describing each of the patients level of quality of life. The objective of the work is to perform a scale validation of EQ i 5D based on data from the clinical trial at Coloplast with wound patients. The scale validation is performed by using methods from item response theory. This is found appropriate for modeling a concept like quality of life based on the possibilities of adjusting the scale. The aim of this work is a reliable scale of quality of life based on the patients answers to the EQ i 5D. This scale should furthermore be used to compare treatment groups. The scale validation is solved by the use of the correlation measure the partial ° coefficient and a graphical loglinear Rasch model from the item response theory. The partial ° coefficient is used to analyse the existence of local dependence and Differential Item Functioning (item bias/DIF) based on correlations between variables. The model type is chosen because it can deal with violations of some general assumptions. The graphical loglinear Rasch model can handle local dependence between items and DIF between items and covariates. Furthermore the model has the advantages of being capable of using classical methods know from the generalized linear models. The statistical analyses result in a dividing of the items in two dimensions - a physical dimension and psychical dimension. The items contain different types of information which can not be contained in one scale. It has been possible the develop a set of models based on the dividing of items which describes a latent scale of the EQ i 5D. The test of reliability shows that the scale for the physical dimension is reliable which is not the case for the psychical dimension. It can be shown that a significant difference between treatment groups exists using classical comparison tests (t-test, Wilcoxon).