Swedish experience-based value sets for EQ-5D health states

Purpose To estimate Swedish experience-based value sets for EQ-5D health states using general population health survey data. Methods Approximately 45,000 individuals valued their current health status by means of time trade off (TTO) and visual analogue scale (VAS) methods and answered the EQ-5D questionnaire, making it possible to model the association between the experience-based TTO and VAS values and the EQ-5D dimensions and severity levels. The association between TTO and VAS values and the different severity levels of respondents’ answers on a self-rated health (SRH) question was assessed. Results Almost all dimensions (except usual activity) and severity levels had less impact on TTO valuations compared with the UK study based on hypothetical values. Anxiety/depression had the greatest impact on both TTO and VAS values. TTO and VAS values were consistently related to SRH. The inclusion of age, sex, education and socioeconomic group affected the main effect coefficients and the explanatory power modestly. Conclusions A value set for EQ-5D health states based on Swedish valuations has been lacking. Several authors have recently advocated the normative standpoint of using experience-based values. Guidelines of economic evaluation for reimbursement decisions in Sweden recommend the use of experience-based values for QALY calculations. Our results that anxiety/depression had the greatest impact on both TTO and VAS values underline the importance of mental health for individuals’ overall HRQoL. Using population surveys is in line with recent thinking on valuing health states and could reduce some of the focusing effects potentially appearing in hypothetical valuation studies. Electronic supplementary material The online version of this article (doi:10.1007/s11136-013-0496-4) contains supplementary material, which is available to authorized users.

Thus, the pooled data set for the present VAS analyses consisted of 41,761 individuals. Before carrying out analyses on the SRH question individuals with missing answer on SRH were excluded: 741 individuals for the TTO analyses and 637 individuals for the VAS analyses.
The internal non-responders to the TTO question (7.5%) had a significantly higher mean age and worse health status compared to those who had answer the TTO question. Non-response to the TTO question was more common among women than among men and in groups with lower education and among manual workers. The same pattern was seen in non-responders to the VAS question (15.1%) except that there was no difference in non-response rate between men and women. Due to the higher proportion of non-responders to the VAS question in the 2004 survey we examined the two surveys separately with respect to VAS, resulting in the same pattern as in the pooled data. Burström et al. (2013). Swedish experience-based value sets for EQ-5D health states. Qual Life Res DOI 10.1007/s11136-013-0496-4. 11136_2013_496_MOESM1_ESM.pdf 4 (21) However, in the 2004 survey the non-response rate was higher among women than men, as found in the analyses of non-responders to the TTO question for the pooled data.

Test for parameter homogeneity across surveys
The estimated models were tested for homogeneity in the parameters of the regressors across the two surveys of the pooled model, with the age restricted to 18-64 years, i.e. the intercept and the slope of all regressors. In this F-test analysis, we hypothesised that the intercepts were equal in the two surveys. Given rejection of this hypothesis we secondly tested the hypothesis that the parameters of all socio-demographic regressors were equal. Given rejection of this hypothesis we finally tested the hypothesis that the parameters of all the dimensional regressors (mobility; selfcare; usual activities; pain/discomfort; anxiety/depression) were equal. Burström et al. (2013). Swedish experience-based value sets for EQ-5D health states. Qual Life Res DOI 10.1007/s11136-013-0496-4.

Regression analysis on VAS values for EQ-5D dimensions
Results of the regression analysis on individual VAS values for EQ-5D dimensions are presented in Supplementary Table S4.
The VAS results were consistent with lower values the more severe the health state was (Model 1), except for self-care and mobility where the value for level 3 was not lower compared to level 2. For the self-care dimension the coefficient for level 3 was positive, but not significant. The N3 variable had a negative sign and was significant (Model 2). The results when entering the MO23 and SC23 variables (merged levels 2 and 3 for mobility and self-care) are shown in Model 3 and Model 4 (including N3).
For health states with five or more observations the Spearman correlation coefficient was greatest for Model 3 and MAD was smallest for Model 4 (Supplementary Table S5). For health states with ten or more observations the correlation was highest and MAD was smallest in Model 4. The adjusted R 2 was similar for all models (around 0.48) (Supplementary Table S4).
Supplementary Figure S1 shows the estimated VAS values predicted by the different OLS models compared to the observed mean VAS values for health states with five or more observations.
Looking at the main effects (Model 4), severe problems in the anxiety/depression dimension had the greatest coefficient (23.7), followed by usual activities (15.0) and pain/discomfort (12.9) (Supplementary Table S4). For moderate problems, the greatest coefficient was for usual activities (12.1) followed by anxiety/depression (10.0) and pain/discomfort (6.7). The merged coefficients for mobility (9.8) and self-care (0.8) represented any move away from no problems. The difference Burström et al. (2013). Swedish experience-based value sets for EQ-5D health states. Qual Life Res DOI 10.1007/s11136-013-0496-4.
11136_2013_496_MOESM1_ESM.pdf 6 (21) between the predicted and the observed mean values exceeded 10 for 14% of the health states with five or more observations.

Split sample test
The results for the Spearman correlation coefficients and the MAD in the split sample test were similar to those obtained in the whole sample. For the TTO model, for health states with five or more observations, the Spearman correlation coefficient remained 0.83 and the MAD was still below 0.06. For the VAS model, for health states with five or more observations, the Spearman correlation coefficient was 0.93, and the MAD was 5.0, similar to that of the whole sample. For health states with ten or more observations, the results from the split sample test showed greater Spearman correlation coefficients and smaller MAD, compared to health states with five or more observations, which was similar to the findings from the whole sample.

Homogeneity across surveys
The results of the F-tests showed that homogeneity of the regression coefficients across surveys were rejected irrespectively of whether TTO or VAS were considered or inclusion of sociodemographic control factors; indicating that the effects of the regressors, both socio-demographic and dimensional regressors, on TTO and VAS values were not equal across the two surveys (except for dimensional regressors on TTO values in the SRH regression) (Table 3 main manuscript;   Supplementary Tables S4 and S8). It is likely that this heterogeneity is mainly due to regional heterogeneity since there is only a two-year time difference between the surveys in the different areas. In the TTO regression, the N3 variable was smaller and non-significant in the 2006 survey and the greatest difference was seen in the anxiety/depression dimension (Supplementary Table S7).
In the VAS regression, the differences were similar. Burström et al. (2013). Swedish experience-based value sets for EQ-5D health states.
Higher non-manual