Hi Stephanie,

yes you can estimate a score from the 5 dimensions of the EQ-5D questionnaire. I attached the use guide of the EuroqoL in which this scoring is explained and for further explanations and/or to obtain the macro for SAS as example you can contact EuroQoL itself (([email protected]))

However, this score is not as those of most of the QoL questionnaires, i.e. the score is not a simple mean of the response obtained for each dimension. You have to first create a 5 digit code corresponding to a summary of the response obtained to the five domains and then a total of 243 health state can be defined....

Indeed, this score is not standardized on a 0-10 or 0-100 scale and thus it can make difficult to interpret the reults only using this summary score/index.

Thus, to analyse this kind of questionnaire I recommend you to follow the user guide and to first describe proportion of patients in each response category of each scale (see page 14 of the attached document as example). The VAS scale is easier to exploit and to interpret.

I hope that these explanations could help you

Best regards

amélie

Hi Stephanie

Since you ask for a scoring algorithm I assume you haven't worked with ED-5D before, so here are a few comments.

The properties of EQ-5D index may not be as other indices of QoL but is still interpretable and the theoretical basis is better than indices based on summary scores. Since the value sets used for computing indices are derived with well established methodology from the health economic field of research, there are good reasons why it should approximate interval data, something that simple summary scores and mean values doesn't.

The recommended analysis method (as far as I know) is group comparisons using a t-test and ANOVA or perhaps even better a non-parametric Mann-Whitney U-test and the Kruskal Wallis test. The t-test and ANOVA uses standard deviations which may not be a good description of spread in the data, medians and quartiles are better. The reason for this is that the index suffer from severe ceiling effects, meaning that at lot of people may ha index 1, so a confidence interval based on the normal distribution may extend beyond 1, which of course is ...just wrong.

For regression the recommended method is ordinary linear regression on the index. With large datasets and good spread over the index range the results should be have "approximately good" statistical validity. There are regression methods (in Stata) for interval range data, like EQ-5D index, but these are based on logistic regression and the coefficients are not easy to translate into the index scale.

When computing the index you have determine which value set to use. I believe the most common one is the UK value set from 1997 [1], but perhaps you should use the Canadian [2] since there are cultural differences between countries. Using this value set will give you indices in the interval (approximately) -1.5 and 1. The upper limit is the value for a subject with no problems in any dimension. States with index