Subjective well-being (SWB) has been suggested for valuing health through patient experience. With the standard valuation regression model approach applied to SWB, the impact of impaired health appears very different from the preferences of the general population, with coefficients for the anxiety/depression dimension vastly larger, and impairments of mobility close to zero. However, SWB has previously been linked to mood. We hypothesized that if the primary impact of impaired health on mood is through depression, other health problems’ impact on SWB would be effectively negated in regression modeling including depression due to construct overlap, rendering standard regression modeling meaningless for causal interpretation. This study analyzed the impact of the other EQ-5D dimensions on SWB contrasted with self-reported general health (GH) using regression models with and without dummy variables for anxiety/depression.
Method:
We used data from the Health Survey for England from 2003 to 2010. SWB was measured using GHQ12 (General Health Questionnaire), GH was assessed using the question “How is your health in general?”, with five options ranging from “very good” to “very bad”, and respondents filled out the EQ-5D-3L. 61,599 respondents with complete data were analyzed.
GH and SWB scores were rescaled to a best score of 0 and worst of -1. In normal regression, 10 dummy variables representing moderate and extreme problems on each EQ-5D dimension were used. In the modified approach, anxiety/depression was first excluded, then assessed in a subsequent regression predicting residuals from the first. Coefficient sizes were compared between the two approaches.
Result:
When predicting SWb with anxiety/depression included, the coefficients (-.147 and -.347) were more than three times the their usual activities counterparts (-.04 and -.098). The remaining coefficients were smaller, and moderate problems on mobility was slightly positive. With the exception of moderate problems on mobility, all coefficients were substantially increased when removing anxiety/depression from the model. The difference between the two approaches was very small for GH.
Conclusion:
Impact of health impairment on SWB may be interesting, but the normal regression model approach appears to obscure impact of other health problems due to construct overlap between SWB and depression. SWB as a criterion for resource distribution in health care seems questionnable given the overlap with depression and relative insensitivity to other health problems.