زمان تجارت کردن با اقتباس وزن EQ-5D برای استرالیا

Background Cost-utility analyses (CUAs) are increasingly common in Australia. The EuroQol five-dimensional (EQ-5D) questionnaire is one of the most widely used generic preference-based instruments for measuring health-related quality of life for the estimation of quality-adjusted life years within a CUA. There is evidence that valuations of health states vary across countries, but Australian weights have not previously been developed. Methods Conventionally, weights are derived by applying the time trade-off elicitation method to a subset of the EQ-5D health states. Using a larger set of directly valued health states than in previous studies, time trade-off valuations were collected from a representative sample of the Australian general population (n = 417). A range of models were estimated and compared as a basis for generating an Australian algorithm. Results The Australia-specific EQ-5D values generated were similar to those previously produced for a range of other countries, but the number of directly valued states allowed inclusion of more interaction effects, which increased the divergence between Australia's algorithm and other algorithms in the literature. Conclusion This new algorithm will enable the Australian community values to be reflected in future economic evaluations.

Economic evaluation of health interventions is integral to the decision-making process in many countries, particularly for government reimbursement decisions. The tools used in the construction of such analyses are, therefore, of increasing importance. Cost-utility analysis (CUA) is the preferred approach in many countries, including Australia. An increasing focus on health-related quality of life has seen the development of standardized descriptive quality of life instruments that allow for direct measurement of the quality of life of patients in clinical settings, trials and observational studies, and valuation via a single index derived from a population-based preference elicitation study. These instruments (termed multi-attribute utility instruments) describe health in terms of a set of dimensions and items and include an algorithm that assigns an index number to each health state (defined as a specific profile of attribute items representing alternative levels of the different dimensions) represented by the instrument space on a scale with one representing full health and zero representing death. Attaching a value greater than zero to a health state implies it is better than dead, whereas a negative value represents a state worse than dead. Existing instruments include the EuroQol five-dimensional (EQ-5D) questionnaire [1], the six dimensional health state short form (SF-6D) [2], the health utilities index 3 (HUI3) [3] and [4], and assessment of quality of life (AQoL) [5]. Australia is an unusual case. Although CUA has become the preferred approach for the evaluation of pharmaceuticals [6], Australian general population specific weights exist for only one of the more common multi-attribute utility instruments (the AQoL). Therefore, Australian CUAs performed using EQ-5D or SF-6D data have relied on weights from other countries, particularly those from the United Kingdom [1] and [2]. Multi-attribute utility instruments have been compared and their role in the economic evaluation of health technologies has been discussed widely in the literature [5] and [7]. In this article, the focus is on the EQ-5D, because it represents the most commonly used generic quality of life descriptive system. The primary aim of this study was to develop Australian based weights for the EQ-5D descriptive system, based on data collected from a sample representative of the Australian general population and using methods that are largely comparable to those used previously to develop weights for other countries. A secondary aim was to explore methodological issues in the derivation of weights for the EQ-5D, particularly in relation to the choice of health states to be directly valued, and the impact of this choice on the weights derived. In this study the choice of health states was informed by undertaking a simulation study. Several different methods were used to define subsets of health states to be directly valued, and simulation data were generated. The results from each of these subsets were analyzed separately and the resulting utility weights were compared for all health states defined by the EQ-5D descriptive system to determine a preferred set of health states to be directly valued. This set was then used for the data collection for the Australian valuation study. The second section of the article briefly describes the EQ-5D and its development, including the methods that underlie the existing algorithms, and in particular the selection of health states for direct valuation. This section motivates the simulation approach used in this study and provides a rationale for the approach. Section 3 describes the methods for the simulation study and for the data collection and analysis for development of the Australian algorithm. Section 4 presents the results and Section 5 discusses the choice of algorithm.

The simulation approach used in this study demonstrates that previous TTO studies designed to develop EQ-5D algorithms lack sufficient coverage of the EQ-5D space to allow identification and estimation of interactions that may be present between dimensions and levels. Our data collection and comparison of models suggests that a more complex algorithm may be appropriate. Current models that include only the N3 term are essentially additive and each level enters as a main effect only (although the N3 term might be considered a constrained interaction term). In this study, the model that provides the best fit includes a more complex set of interactions of dimensions at their worst levels. The fact that these interaction terms are generally positive and, therefore, in the opposite direction to the main effects suggests that there is a multiplicative effect and the additional decrement in utility associated with a worsening in a second dimension is smaller than the decrement for the first worsening dimension. In all specifications, the constant term is significantly different from one. This is consistent with the findings of other studies, and suggests that it is appropriate to include an unconstrained constant term in the TTO algorithm. The inclusion of a constant term that is not constrained to unity is typically interpreted as capturing the effect of any move away from full health. However, it does impact on the valuation of the milder health states, an impact that is particularly evident in comparison of models 1 and 1b. Anchoring prevents a ceiling effect that can be seen in the non-anchored algorithms. However, this ceiling may be justifiable in that the constant plays a role relative to dimensions at level 2 or 3 that the N3 term plays relative to dimensions at level 3 only. The simple main effect models 1 and 1b can be rejected on model fit, with significantly poorer AIC/BIC values than other models. The significant interaction terms present in the other models suggest that neither model 1 nor 1b are appropriate. In addition, the inclusion of an anchoring point on these models has the largest impact on health state valuations. Model 2 is the model that is most consistent with existing international studies and provides a point of comparison between the Australian population's preferences and those of other populations in other countries (Fig. 2). The N3 term is statistically significant and has a similar effect in the Australian models to that seen in other countries. This comparison also suggests broad consistency between Australian valuations and internationally.Models 3 and 4 take a more sophisticated approach to interactions, and both represent an improvement in model fit over model 2. Model 4 includes all interaction terms, whereas model 3 includes only interactions between level 3 of dimensions. Additional combinations of interactions were considered (such as including only interactions involving at least one level 3 dimension, or limiting interactions to specific dimensions), but did not prove better than those reported here. In terms of AIC and BIC, model 3 is preferred to model 4. In both cases some of the interaction terms are not significantly different from zero. This is particularly the case in model 4. Although this may be the effect of sample size given that this model includes a large number of estimated coefficients, the fact that there is not a consistent pattern of interactions also suggests that many of these effects may not impact on the valuation placed on the health state beyond the main effect. As expected, given the pattern of non-statistically significant interaction terms in model 4, it does not provide an improvement over model 3 when compared using the AIC and BIC. In model 3 the interaction terms are more consistently significant, and typically positive. In particular, the interaction terms for mobility with pain/discomfort, self-care with pain/discomfort, and with anxiety/depression, and pain/discomfort with anxiety/depression are all statistically significant (P<0.01) and positive. Comparing models 2 and 3 it can be seen that the main effect for these terms in model 4 is much larger. Although not all interaction terms are significant, the improvement in fit and the significance of interactions between the mobility, pain/discomfort, self-care, and anxiety/depression dimensions suggests that this model is to be preferred over model 2, and provides a more appropriate algorithm for the Australian population. In balancing parsimony with predictive value, we recommend model 3 as the preferred Australian algorithm although we also recommend that the effect of using alternate specifications be considered as part of sensitivity analysis in economic evaluation. There were 14 non-monotonic pairwise orderings of health states in the algorithm implied by model 3. In these pairs, the value placed on the poorer health state exceeded the value placed on the better one by up to 0.079 (mean, 0.028). Because the interaction effects generally offset the main effect (reflecting the fact that the move to a worse level on one dimension depends on the levels of other dimensions and is generally smaller when other dimensions are already at lower levels), it is possible for non-monotonic effects to occur. Given the means and SD of the estimated coefficients that generate these implausible orderings, it is likely that this is a result of sample size rather than valuations (that is, they are generally very small and may result from random error in the data). These non-monotonic orderings are problematic because if used in economic evaluation they would produce implausible cost-effectiveness results, and, therefore, a method was proposed for removing these from the final algorithm implied by model 3. The scores for these health states were then amended by considering each illogically ordered pair and assigning to each the mean value of the two health states under the algorithm. This approach was taken because it minimized the maximum movement of a health state away from the state assigned through the preferred algorithm. Functionally this is equivalent to treating the valuations of the two health states as the same, and treating the non-monotonic effects as random error. In situations in which a health state is in more than one illogically ordered pair, the mean score that does not produce a new illogically ordered pair was selected. The updated valuation of all EQ-5D health states under model 3 with the amendment for illogical pairings is available in Appendix 1 at: 10.1016/j.jval.2011.04.009. The comparability of the amended model 3 algorithm to that produced elsewhere can be addressed using the graphical approach taken by Badia et al. [9]. Ranking each of the 243 states using the Australian algorithm, each state is valued under a selection of the pre-existing algorithms and placed on one graph. Figure 2 compares the Australian weights with a selection of other studies (in this case the UK, Spain, and Japan). This study provides the first Australian general population derived TTO EQ-5D weights for use in Australian CUA. The broad consistency of the health state values predicted by model 3 with those from other studies undertaken elsewhere using the same regression gives us confidence that the valuation studies are comparable. However, the more comprehensive approach taken in this study to both the absolute number and descriptive content of health states included for direct valuation within the preference elicitation study, suggests that a more complex scoring algorithm than traditionally applied may be more appropriate. Further research is required to confirm the pattern of interactions in other countries and settings. Source of financial support: This work was funded by an Australian National Health and Medical Research Council Project Grant403303.