ارزش نهادن به EQ-5D و حالات سلامت SF-6D با استفاده از بهزیستن ذهنی: تجزیه و تحلیل ثانویه داده های بیمار

Abstract The economic evaluation of health care technologies employs a standard economic approach based on preferences to provide utility information. Previous studies have used happiness rather than preferences to weight health states using general population data. However, these data may not reflect the full range and scope of health and happiness experienced by patients. This paper applies a similar approach to a large patient sample (N = 15,184) from a hospital in Wales, UK collected between 2002 and 2004. Logit regression models were used to assess the relationship between happiness and the health state classifications of two measures, the EQ-5D and the SF-6D. The results suggest a different weighting across dimensions to that from preference elicitation techniques such as time trade-off and standard gamble. While mental health (depression and anxiety), vitality and social functioning were found to have a large significant association with the patients' own happiness assessment, pain was less so and physical health had no association.

Introduction Background There has been an increased interest in measuring subjective well-being (SWB) to inform public policy in the United Kingdom as well as in other countries (Abdallah et al., 2011; Forgeard, Jayawickreme, Kern, & Seligman, 2011; Waldron, 2010). This has been driven in part by evidence that increase over time in measures of material well-being such as gross national income has not been matched by increase in SWB. As one of the aims of public policy is to increase overall well-being, including both material and subjective aspects of well-being measures to measure progress and inform public policy decisions is important. In neo-classical economics, income has been used as a proxy for well-being or utility as higher income provides individuals with opportunities to choose more goods or services. Individuals are assumed to be utility maximisers and choices indicate preferences for goods or services that will increase their well-being. This is in contrast to the approach taken by classical economists such as Jeremy Bentham who proposed measuring the well-being of an object based on its ability to increase pleasure or happiness, or to reduce pain which is directly related to SWB (Bentham, 1781/2000; Layard, 2005). Perceived problems with measuring SWB led to the move away from this definition of utility and towards preferences. Kahneman distinguishes between the two forms of utility by referring to the former as decision utility and the latter as experience utility (Kahneman, 2000). Health care policy makers rely on information from the economic evaluation of health care technologies to make resource allocation decisions. Cost utility analysis is an economic evaluation technique which uses the Quality Adjusted Life Year (QALY) to measure the health effects of conditions and associated medical interventions. The QALY is estimated by weighting survival with the health related quality of life (HRQoL) enjoyed in each time period using health state utility values (Torrance & Feeny, 1989). The health state utility values are obtained using preference based HRQoL measures, such as the EQ-5D, SF-6D and the Health Utilities Index (HUI 2 and HUI 3). These measures have a health state classification (HSC) describing health states typically in terms of physical, mental, role and social functioning which is completed by patients. The completed HSC is then scored using existing values obtained via preference elicitation techniques such as the standard gamble (SG) and time trade-off (TTO). SG and TTO aim to elicit utility values associated with hypothetical health states by asking individuals to trade in uncertainty or healthy life years. The values obtained represent preferences or decision utility for hypothetical health states. These values are obtained from members of the general population rather than patients following the recommendations of the Washington Panel and the National Institute of Health and Clinical Excellence (NICE) in England (Gold, Siegel, Russell, & Weinstein, 1996; National Institute for Health and Clinical Excellence, 2008). This approach is based on the premise that the cost and consequences of health care are borne by the general population and their preferences should therefore inform decision-making (Gold et al., 1996). The question is whether these general population preferences match the experiences of patients as health care resource allocation decisions based on these values impact directly on patients. For preferences to match experience, individuals have to accurately predict what the impact of different health states will be on their actual well-being (Dolan & White, 2006). Evidence shows that patient values of experienced health states tend to be higher than values of similar but hypothetical health states generated by the general population for several reasons (De Wit, Busschbach, & De Charro, 2000). Adaptation to poorer health states in patient groups leads to a reduced perceived effect of these health states (Ubel, Loewenstein, & Jepson, 2005). Individuals also mispredict how quickly they adapt to changes in their own lives (Kahneman & Sugden, 2005). The general population is likely to ignore adaptation when undertaking valuations as they may not have experienced the health states they are valuing leading to overestimation of the duration and intensity of experiencing the imagined health states. However, this problem can also affect patients as they may overestimate the duration and intensity of positive changes to health (Dolan & Kahneman, 2008). In addition individuals may focus on specific aspects of being in the hypothetical health state such as the impact of being immobile on their work or transition into these particular health states at the expense of other domains of life, described as focussing illusions, which may lead to overestimation of the negative impact of health states (Ubel, Loewenstein, & Jepson, 2003; Ubel et al., 2005). Loss aversion, where individuals value losses more than gains, may also have an impact. Those who do not have the experience of a health state may place larger weight on the potential loss of health while patients may place relatively smaller weight on gains (Baron et al., 2003). There is evidence that patients are unwilling to trade life years or take risks which may be as a result of loss aversion whereas those who do not suffer from the condition do not have these problems (De Wit et al., 2000; Menzel, Dolan, Richardson, & Olsen, 2002). Differences between the anticipated effect of experiencing particular health states and actual experience of health states can lead to sub-optimal resource allocation decisions. It is therefore important to consider valuation methods that reflect the experiences of patients (Dolan, 2007; Kahneman, 2009; Kahneman & Sugden, 2005; Kahneman, Wakker, & Sarin, 1997). Economist have began to explore the use of SWB measures in health state valuation as direct measures of health state utility instead of preferences (Dolan, 2007; Dolan & Kahneman, 2008). A number of recent studies have focused on this approach (Dolan, Lee, & Peasgood, 2012; Dolan & Metcalfe, 2012; Graham, Higuera, & Lora, 2011). Graham et al. (2011) assessed the relationship between the EQ-5D and both life and health satisfaction in Latin American countries (n = 14,000). Their results indicate that anxiety and pain have relatively stronger associations with SWB than physical health, with larger impacts on health satisfaction compared to life satisfaction. Dolan and Metcalfe (2012) use data from the US (n = 1173) that includes the day affect measure, a momentary measure of well-being experienced the day before, and report similar findings for mental health although pain has a smaller effect in their study. Dolan et al. (2012) show that these results persist when unobserved characteristics such as personality are taken into account using data in the British Household Panel Survey (BHPS) (n = 19,230) and the SF-6D. These results provide evidence that there are differences between preferences and SWB as valuations based on TTO or SG give relatively larger weight to physical functioning and pain compared to mental health ( Brazier, Roberts, & Deverill, 2002; Brazier & Roberts, 2004; Dolan, 1997). However, the datasets used in these studies are general population data rather than patient data. As the focus is on experience, it is important to assess the outcomes of patients who may have experienced a wider range of health states than the general population. In this paper we use a large patient dataset containing both routinely collected data and survey data to assess the impact of valuing the EQ-5D and the SF-6D health states using well-being.

Results Table 1 reports respondents' socio-demographic characteristics. The mean age (s.d.) was 57.5 (17.5) and 45.7% were male. Mean (s.d.) self-reported happiness was 3.48 (0.98) with 4% reporting being happy ‘none of the time’ while 57% report being happy ‘most or all of the time’. A larger proportion of older people (over 65) reported higher levels of happiness (levels 4 and 5) than younger people (61% vs. 55%, χ21 = 53, p = 0.000). There were also gender differences with slightly more men reporting higher levels of happiness than women (59% vs. 55%, χ21 = 24.93, p = 0.000). There are no significant differences in happiness when comparing inpatients to outpatients. Table 1. Sample characteristics. All % Male 45.7 Mean age (s.d.) 57.5 (17.5) Age distribution 18–40 20.0 41–65 41.9 Over 65 38.1 Marital status Single 12.9 Married 61.7 Remarried 2.0 Cohabiting 4.9 Separated 1.2 Divorced 6.9 Widowed 10.4 Employment status Professional 38.5 Skilled 15.6 Skilled-manual 16.8 Manual 7.4 Unskilled 6.7 Never employed 1.8 Unemployed 12.7 Full-time education 0.4 Ethnic group White 97.5 How much of the time during the last 4 weeks have you been happy? None of the time 4.1 A little of the time 12.0 Some of the time 27.0 Most of the time 45.7 All of the time 11.2 Mean self-reported happiness (s.d.) 3.48 (0.98) Mean EQ-5D score (s.d.) 0.68 (0.31) Mean SF-6D score (s.d.) 0.66 (0.15) N 15,184 Table options Mean (s.d.) EQ-5D and SF-6D scores were 0.68 (0.31) and 0.66 (0.15) respectively which are lower than UK general population values of 0.85 (0.23) and 0.80 (0.15) (Petrou & Hockley, 2005). Compared to general population values, a smaller proportion report being in full health (1.0) for both EQ-5D (26% vs. 52%) and the SF-6D (0.7% vs. 8.3%). The reverse is true for lower levels of health with 6% reporting levels of health worse than dead compared to 1.6% for general population in the EQ-5D (Petrou & Hockley, 2005). A large proportion report being in the worst levels of health in the HSC dimensions in both measures (EQ-5D: 17%; SF-6D: 70%) although this is still small for the lowest level in mobility (0.3%) and self-care (1.0%) in the EQ-5D. As would be expected, EQ-5D and SF-6D scores varied by age and gender as well as whether patients were inpatients or outpatients. Correlation analysis results (Table 2) show that as would be expected, EQ-5D and SF-6D scores were positively correlated with happiness while the HSC dimensions were negatively correlated with happiness with correlations ranging from −0.25 to −0.57 indicating small to moderate strength associations. The smallest correlations were with dimensions related to physical functioning while those related to mental health had the largest correlations. Table 2. Spearman rank correlations: happiness with preference-based measures, dimensions and scores. Happiness EQ-5D Mobility −0.246*** Self care −0.243*** Usual activities −0.304*** Pain/discomfort −0.300*** Anxiety/depression −0.535*** EQ-5D utility score 0.428*** SF-6D Physical functioning −0.294*** Role limitation −0.443*** Social functioning −0.485*** Pain −0.374*** Mental health −0.570*** Vitality −0.502*** SF-6D utility score 0.541*** ***p < 0.001. Table options Regression results EQ-5D The ordered logit model results for EQ-5D are presented in Table 3. The proportional odds assumption was not violated for the EQ-5D dimensions. McKelvey and Zavoina's R2 is 0.34 compared to 0.05 when only control variables are included ( Appendix B) indicating that including the HSC dimension variables improves the explanatory power of the models. Inclusion of the control variables does not improve the R2 significantly. Table 3. Multivariate ordered logit regression: EQ-5D dimensions. Explanatory variables Model 1 ordered logit Model 1 + controls ordered logit Model 2 OLS UK TTO (Dolan) Odd ratio SE Odds ratio SE Mobility 2 1.038 (0.048) 0.889* (0.043) 0.001 (0.005) −0.069 Mobility 3 0.903 (0.254) 0.859 (0.246) −0.022 (0.031) −0.314 Self care 2 0.870** (0.043) 0.849** (0.043) −0.015** (0.005) −0.104 Self care 3 0.820 (0.135) 0.766 (0.128) −0.024 (0.018) −0.214 Usual activities 2 0.712*** (0.033) 0.696*** (0.032) −0.032*** (0.005) −0.036 Usual activities 3 0.677*** (0.065) 0.607*** (0.058) −0.034*** (0.010) −0.094 Pain/discomfort 2 0.760*** (0.032) 0.714*** (0.030) −0.025*** (0.004) −0.123 Pain/discomfort 3 0.721*** (0.067) 0.704*** (0.066) −0.030** (0.010) −0.386 Anxious/depressed 2 0.149*** (0.006) 0.156*** (0.006) −0.203*** (0.004) −0.071 Anxious/depressed 3 0.034*** (0.003) 0.040*** (0.004) −0.385*** (0.010) −0.236 N3 0.794* (0.076) 0.854 (0.082) −0.029** (0.010) −0.269 Male 1.007 (0.035) Age 0.863*** (0.020) Age2 1.342*** (0.058) Age3 0.848*** (0.022) Single 0.777*** (0.044) Remarried 1.130 (0.125) Living with partner 0.971 (0.074) Separated 0.591*** (0.084) Divorced 0.736*** (0.046) Widowed 0.698*** (0.041) Mixed 0.644* (0.125) Asian 0.721* (0.114) Black 0.826 (0.173) Other ethnicity 1.074 (0.330) Skilled 0.977 (0.046) Skilled manual 1.002 (0.048) Manual non-skilled 0.867* (0.055) Unskilled 0.880 (0.058) Never employed 1.177 (0.143) Unemployed 0.852** (0.044) In full-time education 0.985 (0.262) Threshold 1 −5.11*** (0.057) −7.26*** (0.395) Threshold 2 −3.34*** (0.041) −5.47*** (0.393) Threshold 3 −1.50*** (0.032) −3.59*** (0.391) Threshold 4 1.34*** (0.031) −0.69 (0.390) Constant 0.744*** (0.003) −0.081 Observations 15,184 15,184 15,184 R2 0.30 McKelvey & Zavoina's R2 0.32 0.34 Log likelihood −17,683 −17,476 Likelihood ratio χ2 5416 5829 Degrees of freedom 11 32 Exponentiated coefficients; standard errors in parentheses *p < 0.05, **p < 0.01, ***p < 0.001. Reference categories: no problems – walking about, with self-care & performing usual activities; no pain or discomfort; not anxious or depressed; female, married, white, professional. Table options Model 1 has odds ratios less than one for all the dimensions in the EQ-5D apart from mobility where odds ratios are greater than one for level 2 indicating that those with poor functioning have greater odds of reporting higher levels of happiness when controlling for all other dimensions. The ordering within each dimension is consistent, with lower odds ratios as functioning worsens. The odds ratios are ranked in size from mobility followed by self-care, pain/discomfort, usual activities and anxiety/depression. The odds ratios are statistically significant at the 5% level apart from the mobility dimension. This indicates that the relationship between the EQ-5D dimensions and happiness is as expected for 4 of the dimensions. Having a problem in any of the dimensions therefore increases the likelihood of reporting lower levels of happiness compared to having no problem. Inclusion of control variables has a significant impact on the coefficients of the mobility dimension. Odds ratios are less than one and statistically significant at the 10% level. Self-care and usual activities also have lower odds ratios when the control variables are included. The age variables have the biggest impact on these dimensions as would be expected as older people are more likely to have problems with mobility, self-care and usual activities. Comparison of the OLS coefficients and Dolan (1997) preference-based weights indicates that there are differences. Mobility, self-care and pain/discomfort have lower weights when SWB is used compared to the preference-based weights based on general population. Usual activities are not very different, while anxiety/depression has a relatively larger impact when based on SWB compared to preferences. The weight associated with being in the worst level for any dimension is also smaller. Coefficients are broadly similar to Dolan and Metcalfe (2012) results for the affect measure of SWB and smaller than those for life satisfaction although they use the US tariffs. SF-6D The ordered logit results for the SF-6D are reported in Table 4. The proportional odds assumption was violated for all the SF-6D dimensions apart from the physical functioning dimension. Analysis using the generalised ordered logit indicates the largest differences occur for individuals reporting the lowest levels of happiness (Mukuria & Brazier, 2011). McKelvey and Zavoina's R2 is approximately 0.5 compared to 0.05 in the model with control variables which suggests that the SF-6D dimensions are better predictors of SWB than the EQ-5D dimensions. Table 4. Multivariate ordered logit Regression: SF-6D dimensions. Explanatory variables Model 3 ordered logit Model 3 + controls ordered logit Model 4 OLS UK SG (Brazier) Odds ratio SE Odds ratio SE Beta SE Physical functioning 2 1.209*** (0.067) 1.086 (0.062) 0.015** (0.005) −0.035 Physical functioning 3 1.444*** (0.095) 1.211** (0.083) 0.031*** (0.006) −0.035 Physical functioning 4 1.834*** (0.159) 1.467*** (0.132) 0.053*** (0.008) −0.044 Physical functioning 5 1.948*** (0.144) 1.540*** (0.121) 0.060*** (0.007) −0.056 Physical functioning 6 2.074*** (0.178) 1.559*** (0.142) 0.068*** (0.008) −0.117 Role limitation 2 1.323*** (0.083) 1.224** (0.078) 0.024*** (0.006) −0.053 Role limitation 3 0.550*** (0.049) 0.498*** (0.044) −0.065*** (0.008) −0.053 Role limitation 4 0.686*** (0.050) 0.613*** (0.045) −0.044*** (0.007) −0.053 Social functioning 2 0.691*** (0.038) 0.736*** (0.041) −0.029*** (0.005) −0.057 Social functioning 3 0.555*** (0.032) 0.600*** (0.034) −0.047*** (0.005) −0.059 Social functioning 4 0.411*** (0.029) 0.448*** (0.031) −0.080*** (0.007) −0.072 Social functioning 5 0.336*** (0.028) 0.346*** (0.029) −0.103*** (0.008) −0.087 Pain 2 0.982 (0.057) 0.976 (0.057) −0.001 (0.005) −0.042 Pain 3 0.840** (0.050) 0.862* (0.052) −0.014* (0.006) −0.042 Pain 4 0.881 (0.059) 0.891 (0.060) −0.009 (0.006) −0.065 Pain 5 0.799** (0.056) 0.842* (0.060) −0.017** (0.007) −0.102 Pain 6 0.622*** (0.055) 0.700*** (0.062) −0.049*** (0.008) −0.171 Mental health 2 0.326*** (0.016) 0.342*** (0.017) −0.073*** (0.004) −0.042 Mental health 3 0.177*** (0.010) 0.190*** (0.010) −0.129*** (0.005) −0.042 Mental health 4 0.063*** (0.005) 0.073*** (0.005) −0.249*** (0.007) −0.100 Mental health 5 0.041*** (0.004) 0.049*** (0.005) −0.287*** (0.008) −0.118 Vitality 2 0.289*** (0.030) 0.284*** (0.030) −0.081*** (0.009) −0.071 Vitality 3 0.126*** (0.013) 0.122*** (0.013) −0.144*** (0.009) −0.071 Vitality 4 0.063*** (0.007) 0.059*** (0.007) −0.213*** (0.010) −0.071 Vitality 5 0.042*** (0.005) 0.038*** (0.004) −0.259*** (0.010) −0.092 Most 1.303*** (0.087) 1.327*** (0.089) 0.030*** (0.006) −0.061 Male 1.116** (0.039) Age 0.862*** (0.020) Age2 1.302*** (0.057) Age3 0.875*** (0.023) Single 0.740*** (0.042) Remarried 1.143 (0.132) Living with partner 0.971 (0.076) Separated 0.567*** (0.081) Divorced 0.720*** (0.046) Widowed 0.688*** (0.041) Mixed 0.819 (0.158) Asian 0.982 (0.157) Black 0.803 (0.166) Other ethnicity 1.170 (0.356) Skilled 1.021 (0.049) Skilled manual 1.003 (0.049) Manual non-skilled 0.908 (0.059) Unskilled 0.983 (0.067) Never employed 1.268 (0.158) Unemployed 0.910 (0.048) In full-time education 1.148 (0.309) Threshold 1 −8.01*** (0.123) −10.41*** (0.417) Threshold 2 −6.14*** (0.116) −8.50*** (0.414) Threshold 3 −4.04*** (0.111) −6.37*** (0.412) Threshold 4 −0.56*** (0.103) −2.86*** (0.408) Constant 0.896*** (0.009) 1.000 Observations 15,184 15,184 15,184 R2 0.42 McKelvey & Zavoina's R2 0.49 0.50 Log likelihood −15,965 −15,816 Likelihood ratio χ2 8852 9150 Degrees of freedom 26 47 Reference categories: health does not limit physical, role and social functioning, has no pain, does not feel tense or downhearted and has a lot of energy all the time; female, married, white, professional. Table options As with the EQ-5D, odds ratios are less than one for all dimensions/levels with the exception of those associated with physical functioning (Model 3). All the dimensions show decreasing odds ratios as functioning worsens, as would be expected apart from physical functioning and role limitation. For physical functioning, odds ratios increase as functioning worsens indicating that having lower levels of physical functioning is associated with higher odds of reporting being happy. The odds ratio for the second level of role limitation, ‘limitations in work or other activities by physical health’, is larger than one while levels 3 and 4 are less than one but the order of magnitude is reversed. These levels refer to emotional health, either on its own (level 3) or in addition to physical health (level 4) and this may explain why the order is reversed with level 4, which includes references to physical health, higher than level 3. The largest odds ratios are those for physical functioning followed by role limitations associated with physical health (level 2), pain, role limitation (levels 3 and 4), social functioning, mental health and vitality. The odds ratios are all statistically significant at the 5% level apart from levels 2 and 4 in the pain dimension. The ‘most’ variable has an odds ratio greater than 1. Including control variables lowers the odds ratios for physical functioning and role limitation, although these remain above one. Comparison of the weights from the OLS analysis (Table 4) with Brazier and Roberts (2004) preference-based weights indicates that using SWB gives relatively larger weight to mental health and vitality compared to physical functioning and pain. Role limitations associated with physical health also have lower relative weight when SWB is used. For the other role limitation levels and social functioning, the weights are not significantly different when comparing SWB and SG weights. Coefficients are larger in absolute size than those reported in Dolan et al. (2012) but this is as expected as they control for personality. The strong reverse effect for physical functioning and level 2 for role limitation was tested by analysing the association of the individual variables with happiness. Correlations indicate that these dimensions are negatively associated with happiness. Interactions with other dimensions were not significant. Addition of other dimensions to the model with physical functioning indicates that inclusion of vitality has a strong impact on this dimension, reducing its impact and further inclusion of mental health causes reversal of the coefficients. The vitality dimension refers to energy and is therefore correlated with physical functioning although likelihood ratio tests indicate that excluding either physical functioning or vitality does not improve the model. It is also important to note that most of the dimensions of the SF-6D have high correlations (>0.5). Control variables As with previous studies, we find that age is negatively correlated with SWB but age squared is positively correlated with SWB (Dolan et al., 2008). We include age cubed as we expect the very old in a patient dataset to have poorer indicators and as a result poorer SWB and the results indicate that this is the case, suggesting an S-shaped curve. As already noted, inclusion of age has an impact on the HSCs coefficients. Those who are single, separated, divorced or widowed have lower SWB compared to those who are married, a common finding in the literature. Gender is only significant in Model 3 where SF-6D dimensions are included and it indicates men are likely to report being happier than women which is in contrast to other literature (Dolan et al., 2008). Some of the ethnicity and occupation variables are significant but only in Model 2 with EQ-5D dimensions. These variations are unsurprising given the differences in the dimensions and wording of the two HSCs. SF-6D may capture occupation effects better than the EQ-5D because of the focus on role.