Hispanic Valuation of the EQ-5D Health States: A Social Value Set for Latin Americans

Introduction

The measurement of health-related quality of life (HrQoL) is of growing importance in assessing the effectiveness of health interventions at all levels of the health-care system. Although a large number of available HrQoL instruments are available, not all of these have the necessary properties to support their legitimate use in cost-effectiveness analysis. Profile measures such as Short Form 36 (SF-36) [1] or Nottingham Health Profile (NHP) [2] that generate concurrent observations on separate dimensions of health cannot be used as they lack the capacity to represent outcomes in terms of a single summary score. This property is designed into the construction of index measures such as Health Utilities Index (HUI) [3] and EQ-5D [4] in which health status is represented as a single numeric value. A further desirable attribute evidenced by these particular HrQoL measures is that they are generic in scope and were designed as general purpose tools. This generic status facilitates the comparison of cost-effectiveness results by those concerned with making system-wide decisions about the allocation of health-care resources.

A key consideration in regard to any generic HrQoL measure used for such social decision-making is the origin of its weighting system. Following the recommendations of the Washington Panel on Cost-Effectiveness in Health and Medicine [5] and more recently published guidance of the UK National Institute for Health and Clinical Excellence (NICE) [6], it is generally accepted that when generic HrQoL measures are used in the computation of quality-adjusted life years, then the values applied should represent the social preferences of the relevant population. Strictly interpreted, this requirement means that valuation systems developed in other countries should be calibrated in terms of the domestic population before being applied by decision-making agencies. This is a major challenge because the resources in terms of research capacity and cost are distinctly nontrivial. This has led to the unsatisfactory short-term expediency of uncritically adopting social preferences without the consideration of evidence to support that decision.

EQ-5D is a generic measure of HrQoL developed by the EuroQoL Group, an international research network established in 1987, and has been used extensively to investigate the value of health in many countries [7]. From the outset, visual analog scaling (VAS) rating was used as the standard method of valuing EQ-5D health states adopted by the Group. However, local innovation and experimentation was also encouraged, and this led to the development of a multimethod valuation protocol that included ranking, VAS rating, and time trade-off (TTO) utility elicitation. The protocol formed the centerpiece of the Measurement and Valuation of Health Project (MVH), which conducted a national survey of EQ-5D values in the United Kingdom in 1993. The social preference weights generated as a consequence of that survey [8] remain in current use within the United Kingdom but by custom and practice have become regarded as a form of default value set in the absence of local national alternatives.

In 2002, a national study was conducted in the United States to establish a set of population-based preference weights for EQ-5D [9]. Funded by the Agency for Healthcare Research and Quality, this study was largely based on the MVH methodology and included TTO utility elicitation. One of the few differences between this study and the earlier MVH study was in the design of the sampling frame. Two language versions were fielded in the U.S. study to take account of the growing minority of Hispanics, which in 2000 represented 13.3% of the total U.S. civilian noninstitutionalized population [10]. The investigation of the preferences of key population subgroups is a major issue in itself because the representation of national values involves a form of aggregation for which no fixed methodology has so far been established. The aim of this article was to examine methods that might be used to develop a set of weights for EQ-5D based on the social preferences of Hispanics who had participated in the U.S. valuation survey.

Methods

Results

Sociodemographic and self-reported health characteristics of the respondents in the U.S. Valuation survey are described in Table 1. Both Hispanic subgroups tended to be younger, less educated, reported less health problems on EQ-5D, and had a higher self-rated health status than their non-Hispanic nonblack counterparts. Rates of reported problems were lower among _SPHispanics for mobility, usual activities, and pain/discomfort dimensions than among their English-speaking counterparts. Although _SPHispanics were older and less educated, their self-rated health status was significantly higher (P < 0.001) than for the _ENHispanics.

Mean TTO values for the 42 EQ-5D health states are described in Table 2. The highest valued state for non-Hispanic nonblacks was 11121 (indicating a moderate problem with pain/discomfort). The highest valued state for both Hispanic groups was 21111 (indicating moderate problem with mobility). The worst health state for all subgroups was 33333 with values of −0.113, −0.057, and −0.103 for non-Hispanic non-blacks, _ENHispanics, and _SPHispanics, respectively. The _ENHispanics tended to have higher values than non-Hispanic nonblacks (32 out of 42), but only seven of the paired tests are statistical significant. Mean TTO values for 22 health states were higher in _SPHispanics than in non-Hispanic nonblacks of which 13 were statistical significant. The mean absolute difference in TTO scores between non-Hispanic nonblacks and _ENHispanics was smaller than that between non-Hispanic nonblacks and _SPHispanics, being 0.054 and 0.070, respectively. In general, Hispanics had higher mean valuations for some of the mild and most of the severe health states compared with their non-Hispanic nonblack counterparts as seen in Figure 1.

The regression analyses showed no statistical significance for the _ENHispanic dummy after adjusting for age, sex, and self-reported health problems (P > 0.05). This finding suggests that differences between non-Hispanic nonblacks and _ENHispanics reflected in aggregate-level data were affected by variations in respondent-level characteristics. The _SPHispanic dummy was statistically significant when the regression model was adjusted by the same background variable (Table 3). This suggests that, on average, the _SPHispanic community values EQ-5D health states 0.034 point higher than non-Hispanic nonblacks. Further analysis that allowed for interaction effects found that 5 out of 10 terms involving EQ-5D dimensions reached statistical significance. From these five terms, having moderate self-care or having severe pain/discomfort and anxiety/depression problems were valued higher in the _SPHispanics group than in non-Hispanic nonblacks. The opposite situation occurred with having severe usual activities and moderate pain/discomfort problems, which were valued lower in the _SPHispanics than in the non-Hispanic nonblack group.

Because differences were statistically significant in the regression analysis only when non-Hispanic non-blacks were compared against the _SPHispanics, a number of RE models were developed to fit the valuations of the latter. Only models with the highest goodness of fit are presented in Table 4. In addition to the 10 EQ-5D dummy variables included in the basic (main effect) regression, two other terms that accounted for interactions were statistically significant with a P-value < 0.001: N3 and X4. The estimates for these two terms (N3 and X4) were higher in magnitude than the lowest of the 10 EQ-5D dummy variables. Only X4 presented a negative estimate, which produces a positive adjustment to health states that have four or more dimensions on level 2 or level 3. Despite accounting for the same proportion of variance, the N3 + X4 model generated the lowest number of errors greater than 0.05 in absolute magnitude (i.e., 8 out of 42). The exclusion of an intercept did not improve the model fit.

Residuals of the N3 + X4 model appeared normally distributed in the scatter plot. As expected and seen in previous studies, heteroskedasticity was present in the model [12,13]. When robustness was explored by splitting the sample randomly in two and using the value set of one-half to predict the values of the other half, both estimated and predicted values were highly correlated (R² = 0.96) with an MAE of 0.044.

Table 5 reports a goodness-of-fit comparison between the Latin N3 + X4 model, the UK (MVH) N3, and the U.S. D1 model based on their original estimates. From the results, it seems clear that neither the N3 nor the D1 model adequately fit the valuations from the _SPHispanics. The Latin N3 + X4 model provided the best fit for the valuation data, with few predicting errors and no bias across the 42 EQ-5D health states. The full set of the Spanish-speaking Hispanic preference weights for the 243 EQ-5D health states is provided in Table 6.

Discussion

This study confirmed that differences in EQ-5D valuations exist when population subgroups are compared. Similar results have been recently reported by Shaw et al.; however, preferred language of the respondents was not taken into account in their subgroup analysis [14].

Differences in the observed data are mainly found in the most severe health states where Hispanics generally recorded higher valuations. Regression analyses showed that, on average, _SPHispanics valued EQ-5D health states 0.034 point higher than their non-Hispanic nonblack counterparts. Further regression analyses accounting for interaction suggested that differences in valuations are not constant across the EQ-5D dimensions but differ in magnitude depending on the level of the problem and dimension.

_SPHispanics attached more importance to health states in which individuals have severe problems in pain/discomfort and anxiety/depression than non-Hispanic nonblacks (reflected by the positive sign before the coefficients of the interaction model in Table 2). By way of contrast, having severe problems on usual activities was a situation considered to be of less importance for the _SPHispanics than for their non-Hispanic nonblack counterparts. Opposite findings were reported in a previous study that compared UK and Spanish EQ-5D TTO valuations where Spaniards ascribed lower weight to more symptom-based dimension of pain/discomfort and anxiety/depression [15]. This difference between _SPHispanics in the United States and the European Spaniards certainly requires further analyses; nonetheless it might be reflecting a different attitude toward the valuation of health that goes against the idea of cultural heritage [16].

The development of the N3 + X4 model based on the valuations of the _SPHispanics who participated in the U.S. EQ-5D valuation study represents, as far as is known, the first social value set estimated for general Latin American population. The extent to which its application might generate significant differences in cost-effectiveness ratios within the United States certainly requires more research, although we agree with Shaw et al. on the idea that national preferences should be preferred in domestic economic evaluations unless a strong justification exists [14].

The comparison of estimated values reported for the N3 and D1 indicates that the N3 + X4 model reflects more appropriately the preferences of the Spanish-speaking community, which now stands as the best available proxy of social values in many Latin American countries. It is still unclear how this model might fit the preferences of French- or Portuguese-speaking Latin Americans, but given that these populations share not only the Latin culture but also the urgent need for methods to support the cost-effective allocation of resources, the N3 + X4 model may constitute a reasonable option when comparing alternative health-care interventions in their local context.

Conclusion

Despite the diversity and limited number of respondents, the N3 + X4 model developed for the _SPHispanic community produces a value set with good fit, with an MAE of 0.031 and only eight prediction error exceeding 0.05 in absolute magnitude. We believe that the use of this new model will ease the progress of economic evaluations in Latin American populations and at the same time will aid decision-makers from both the private and the public health-care sectors.

Source of financial support: Victor Zarate was sponsored by the “Alan Williams Fellowship,” University of York, York (UK), and Paul Kind was supported by the EuroQol research fund.