کره جنوبی ارزش زمان تجارت کردن برای EQ- 5D ایالات سلامتی: مدل سازی با مقادیر مشاهده شده برای 101 ایالات بهداشت

Objectives This study establishes the South Korean population-based preference weights for EQ-5D based on values elicited from a representative national sample using the time trade-off (TTO) method. Methods The data for this paper came from a South Korean EQ-5D valuation study where 1307 representative respondents were invited to participate and a total of 101 health states defined by the EQ-5D descriptive system were directly valued. Both aggregate and individual level modeling were conducted to generate values for all 243 health states defined by EQ-5D. Various regression techniques and model specifications were also examined in order to produce the best fit model. Final model selection was based on minimizing the difference between the observed and estimated value for each health state. Results The N3 model yielded the best fit for the observed TTO value at the aggregate level. It had a mean absolute error of 0.029 and only 15 predictions out of 101 had errors exceeding 0.05 in absolute magnitude. Conclusions The study successfully establishes South Korean population-based preference weights for the EQ-5D. The value set derived here is based on a representative population sample, limiting the interpolation space and possessing better model performance. Thus, this EQ-5D value set should be given preference for use with the South Korean population.

Economic evaluations of health-care interventions provide important evidence to decision-makers in charge of making efficient resource allocations within their jurisdictions. Qualityadjusted life year (QALY) is one of a number of measurement units in cost-utility analysis for economic evaluation. QALY stands for both quantity and quality of life. To calculate the value of a QALY, a set of value scores needs to be assigned to each of the various health states indicating weights for quality of life, also known as health-related quality of life (HrQoL). It is recommended that these values be calibrated using social preference weights elicited from the general population [1]. In addition, because the preferences for health states can differ across cultures [2], many countries have measured their own population-based preference weights for all possible health states. Several methods to quantify people’s preferences for health status have been developed; these include visual analog scale (VAS), standard gamble, time trade-off (TTO), and person trade-off methods [3]. Together with EQ-5D [4], there are other preference-based health status measures that can be used to classify the health state of individuals and summarize the change of health outcome in a single index score. For example, there are the Health Utilities Index [5], SF-6D [6], and Quality of Well-Being Scale [7]. In Korea, as in many other countries, there is growing interest in EQ-5D due to the increasing need of measuring the change in HrQoL as an outcome of the health care program. The Korean version of EQ-5D has been under development for some time. Its reliability and validity has already been proven [8] and it was included in the Korea National Health and Nutrition Survey, designed to measure population health in 2005. In order to develop a population-based preference weights for EQ-5D (also known as EQ-5D value set), a valuation study was conducted, in which a subset of health states defined by the EQ-5D descriptive system was directly valued. Based on these observed values, a regression modeling approach is adopted to exploit values for all 243 health states defined by EQ-5D. It must be noted here that there appears to be reported in the literature only one earlier study that attempted to develop the EQ-5D value set for the population in South Korea [9]. However, due to drawbacks in the design of its valuation study and modeling, the sample was not nationally representative and the average of absolute differences between observed and estimated scores was as great as 0.071. To the authors’ knowledge, to this day the demand for a representative and reliable EQ-5D value set for South Korean population is still not met. The current study establishes the South Korean populationbased preference weights for EQ-5D based on the values elicited from a national representative sample using the TTO method. One of the main features of the survey where the preference data were collected is the number of health states involved in the study. Unlike previous valuation studies performed in Korea or in other countries, where either 43 health states defined by EQ-5D or less were directly valued, here the values for a total of 101 EQ-5D health states have been directly observed. Thus, with this unique dataset it is expected that the interpolation spaces in estimating a value set are minimized in comparison to other value sets.

This study collected TTO values for 101 EQ-5D health states from a South Korean representative sample. Based on these values the population-based preference weights for EQ-5D are developed using the N3 model. This model yields the best fit for the observed TTO value at aggregate level, with an MAE of 0.029 and only 15 (out of a total 101) prediction errors exceeding 0.05 in absolute magnitude. At the aggregate level, despite the D1 model producing identical results to the N3 model, the negative sign of coefficient estimations for both interaction terms in the D1 model make it less transparent in calculation. It also becomes conceptually difficult to understand why, for instance, health states with more level 3 problems result in an increased value. Thus, the N3 model is preferred. The empirical comparisons between modeling at aggregate and individual levels in the current study support the use of aggregate level analysis. However, the choice of either aggregate or individual level based analysis is an ongoing debate. Advantages associated with the aggregate level approach include simple modeling, easy interpretation, and being intuitive. On the other hand, the advantages associated with the individual level approach include utilizing the maximum amount of information and treating each respondent’s value on an equal basis. Theoretically, individual level analyses might be expected to produce better results with their capacity to adjust for individual effects. However, in practice it is commonly found that there is too much noise in individual level data that hinders the performance of the estimates. In contrast, aggregate level analysis can alleviate such a problem by regressing at aggregate measures to minimize the unwanted variations. The choice of central tendency measures, such as the mean or median in the aggregate level analysis, is a debatable issue and the exploration of the impact of the choice of central tendency is beyond the scope of this study. Compared with other valuation studies, the major contribution of the current study is the number of health states that were directly valued. Unlike other studies, either following the 43 EQ-5D health states in the MVH project or decreasing the number of health states investigated to fewer than 43, this study increases the number of health states. A total of 101 health states were valued, of which 23 overlap with the health states in the MVH set. Therefore, there are at least 2.4 times more health states investigated than in other studies, covering almost 42% of the total health states (101/243) defined by the EQ-5D descriptive system. As a result, this study provides more information regarding how values (observed) are distributed in the valuation space defined by EQ-5D, and consequently it limited the interpolation space in the estimations. There are three possible ways to transfer the TTO value for states worse than death: monotonic, linear, and truncated transformations [16]. The choice of transformation method in our study was purely based on empirical evidence showing that the linear transformation results in the smallest MAE amongst the three methods. There is no theoretical ground for the choice of one method over another. However, there should be awareness of the effect of applying different transformations in the resulting EQ-5D value set and consequently on the cost-effectiveness analysis. For instance, the value set based on linear transformation produces a smaller range of values and therefore the QALY estimation, and possibly QALY gain, will be smaller than those estimated from a value set based on monotonic transformations. A possible contribution to the discrepancies in observed TTO values and, consequently, coefficient estimation between the current study and Jo et al.’s [9] is the sampling difference. The sampling in our study is from 15 regions representing the whole country (except the Jeju region), while in the latter study it was confined to two adjacent regions only (Seoul and Gyeonggi-do). Our data suggests that the values obtained from the other 13 regions are different from values elicited from Seoul and Gyeonggi-do regions (data not shown). Therefore, the values elicited from these two regions alone cannot be used as a representative preference for the population in South Korea as a whole. Another possible explanation for differences in coefficient estimations between the two studies could be the number of health states involved. In this study there are 101 health states with directly observed values, whereas Jo et al. [9] use only 42 states. In other words, in our study there is more information available regarding the valuation space defined by EQ-5D, and therefore it minimizes the interpolation spaces in the estimation. Particularly, this study values 26 severe health states directly. In contrast, only seven severe health states were investigated in the latter study. Thus, the coefficient estimation for level 3 problems is likely to be more robust in the current study. We are confident that the EQ-5D preference weights developed in this study are better than the ones published previously and should be used preferentially for South Korean population. There are three main reasons for this: First, the data was collected from a national representative sample. Second, it is based on the values of 101 health states, which is twice the number of health states used in the earlier study and, on average each health state (apart from state “33333”) has about 150 observations to provide a reliable estimate. Finally, the performance of the chosen model in our study is superior to the final model in the previous study in terms of the size of MAE and of the proportion of health states with an absolute estimation error greater than 0.05 and 0.10. When considering the correlation coefficients and MADs between the estimated value set in the current study and the official value set in other studies, the estimates here are closer to values in the Japanese study than those in the USA and UK. This observation could represent the cultural similarity between Korea and Japan which was also observed in the previous Korean study [7]. In conclusion, the study successfully establishes a set of South Korean population-based preference weights for the EQ-5D. The value set derived here is based on a population representative sample, limiting the interpolating space and possessing better model performance. Thus, this EQ-5D value set should be used preferentially for the South Korean population. Source of financial support: This study was prepared by financial support of Korea Centers for Disease Control and Prevention (KCDC) in the Ministry of Public Health and Social Welfare.