Rural Population’s Preferences Matter: A Value Set For The EQ-5D-3L Health States For China’s Rural Population

doi:10.21203/rs.3.rs-746682/v1

Purpose: To develop an EQ-5D-3L social value set based on Chinese rural population’s preferences using the time trade-off method, and to compare the differences in health states preferences between China urban and rural population.

Methods: Between Sep 2013 and Nov 2013, a total of 1201 participants were recruited from rural areas of five Chinese cities (Beijing, Chengdu, Guiyang, Nanjing, and Shenyang) using a quota sampling method. A total of 97 EQ-5D-3L health states were valued for estimating the value set. Various models with different specifications were explored at both aggregate and individual levels. The final model was determined by a set of predefined selection criteria.

Findings: An ordinary least square model at the aggregate level included 10 dummy variables for specifying the level 2 and 3 for each dimension and a N3 term presenting any dimension on level 3 was selected was selected as the final model. The final model provides a value set ranges from -0.218 to 0.859. The predicted utility values were highly correlated with but consistently lower than that of the published Chinese EQ-5D-3L value set (for urban population).

Conclusion: The availability of the China rural value set provides a set of social preferences weights for researchers and policy decision-makers for use in China rural area.

Health Economics & Outcomes Research

EQ-5D

China

rural resident

time trade-off

quality of life

value set

The socioeconomic status and demographic characteristics vary between urban and rural population in China. This study is the first attempt to provide a value set of weights for health states based on the preferences of Chinese rural registrants, and offers an evidence-based approach to health utility measurement in policy decision-making for urban-rural health care system integration and health equity promotion as an important supplement to the Chinese urban populations.

Health preference, which usually measured as utility or value, is people’s preference towards a particular health outcome. Utility and value are used to calculate quality-adjusted life years (QALY) in cost-utility analysis (CUA) which allows comparison across different health programs and can provide compelling evidence for medical decision-making [1].

Health preferences are not universal across jurisdictions [2], even vary among subpopulations in a jurisdiction [3, 4], many jurisdictions recommend the health preference from a representative sample be used to develop the value set of utility-based instruments such as the EuroQol Group’s EQ-5D which is one of the most widely used generic measures. Thus, utilities derived from value sets contextualizing the indirect preference-based instruments should be based on locally defined weights to ensure valid inference [4–6]. Using a value set derived from an unrepresentative sample in medical decision-making may lead to inequality and unethical consequences.

In China, residents in rural areas accounted for more than 50% of the whole Chinese population before 2011 [7]; since then, with the continuous advancement of China’s urbanization process, the number of rural residents has declined, but still accounts for a considerable proportion. In 2020, the proportion of residents in China’s rural areas was around 40% of the whole population [8].

The socioeconomic status and demographic characteristics vary between urban and rural population in China. More specifically, from the perspective of income, between 2013 and 2020, the annual per capita disposable income of the urban population (43,834 RMB ≈ 6,838 USD in 2020) in China was approximately three times that of the rural population (17,131 RMB ≈ 2,673 USD in 2020) [7]; in rural areas, the number of people receiving minimum living security benefits is 4.5 times that of urban people [8]. For the education level, the proportion of the Chinese rural population with a high school or higher education is about 10%, and this proportion exceeds 38% in urban areas [9]. For family structure, on average, each household in the rural areas has one more person than each household in urban areas [9]. In addition, the proportion of people over 60 years old in rural areas is 15.0%, while in urban areas is 11.7%, and the total dependency ratio is 51.8% in rural areas and 34.7% in urban areas [9].

Medical resources allocation between urban and rural areas in China is also very uneven. Medical institutes in rural areas of China are mainly Grade II institutions and village clinics, while the vast majority of Grade III institutions are gathered in urban areas. The number of hospital beds per 1,000 people was 4.8 in rural areas and 8.8 in urban areas, and the number of physicians per 1,000 people was 2.0 rural areas and 4.1 in urban areas in 2019 [10]. Furthermore, in general, rural residences are covered by resident basic medical insurances, which provide narrower coverages and lower reimbursement rates than Urban Employee Basic Medical Insurance (UEBMI). Health expenditure as a percentage of consumer expenditure in rural population (10.3% in 2020) is higher than that in urban population (8.0% in 2020) [7]. For the population-level health status, the highest mortality disease among rural residents is heart disease (164.66/100,000), while the highest mortality disease among urban residents is cancer (161.56/100,000) [11]. The proportion of people over 60 who self-reported unhealthy or unable to take care of themselves in rural and urban areas were 20.3% and 12.3%, respectively [9].

EQ-5D is the most commonly used generic preference-based instrument in economic evaluations around the world. It has five dimensions including mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. For the 3L version (EQ-5D-3L), each dimension contains three levels of severity i.e., no problems, some/moderate problems, and extreme problems, which can generate a total of 243 (= 3⁵) unique health states ranging from 11111 (full health) to 33333 (the worst health state). For the 5L version, each dimension has five response levels.

China Guidelines for Pharmacoeconomic Evaluations (2020) recommends that in cost-utility analysis, researchers should use generic preference-based instruments which have a value set based on the Chinese population’s preferences [12]. The instruments that meet the above requirements in China are EQ-5D-5L, EQ-5D-5L and SF-6D V2 [12]. The EQ-5D-3L (Liu et al., 2014) and EQ-5D-5L (Luo et al., 2017) Chinese value sets that endorsed by the EuroQol group were only based on the urban population’s preferences [13, 14]. Zhou and colleagues published another Chinese EQ-5D-3L value set based on an urban and rural mixed sample [15]. Zhou’s EQ-5D-3L value set consistently provides higher utility values (238 out of the 243 eq-5d-3l health states) than Liu’s 3L value set. The largest absolute difference in predicted utility values between the two value sets is 0.5288 (health state 22332), and 28.1% health states have a larger than 0.3 absolute difference in predicted utility values between the two value sets. Previous study and the differences between the two value sets have suggested that the health preferences of urban and rural populations in China are considerable different. And considering the afore-mentioned socio-economic demographic differences and the differences in the accessibility of medical resources between the rural population and the urban population in China, using existing value sets to support health care resources allocation in the rural areas may cause inequity. Therefore, a value set of EQ-5D based on the Chinese rural population’s preferences is warranted. Such a value set can help researchers better explore and understand the differences in health preferences between China rural and urban populations, and can also provide some support for decision-makers in avoiding inequity decisions.

It is admitted that the EQ-5D-5L has demonstrated better measurement properties than the EQ-5D-3L in many populations [16, 17], considering the relatively lower education level and larger proportion of older people of the rural population, less response levels may help people to better understand the TTO tasks and the instrument per se in valuation studies and health surveys in rural areas [18, 19]. The present study primarily aimed to establish an EQ-5D-3L value set based on a sample which can represent the China's rural population. The secondary objective was to compare the differences in health states preferences between China urban and rural population.

Study Overview

The China Rural population EQ-5D-3L valuation study was conducted between September 2013 and November 2013, and was carried out in rural areas of five Chinese cities including Beijing, Chengdu, Guiyang, Nanjing, and Shenyang. Data were collected through face-to-face interviews. Paper based traditional time-trade-off (TTO) method was used as the main preference elicitation technique.

Sampling

From the highest to lowest, there are three hierarchical levels in Chinese rural area, i.e., county, township, and village. Participants were recruited from the five cities’ rural areas using quota sampling in terms of age and sex according to the Sixth National Population Census. Within each city’s rural area, 3 geographically dispersed counties were selected. Within each county, respondents were recruited from at least two townships’ village area. In order to ensure geographic distribution of the sample, no more than 40 respondents were recruited from the same township. In addition, in order to avoid mutual influence of health preferences among household members, no more than one participant per household was selected. The target sample size of each city’s rural area was 240 respondents.

Participant inclusion criteria were as follows: 1) were rural Hukou holders; 2) had inhabited the selected village for at least six months; 3) aged 16 years or older; 4) were able to read in mandarin; 5) did not have any cognitive impairment and mental disease; and 6) did not have any serious vision, hearing and communication problems. During the interview, if an interviewee could not understand the survey in any aspect, the interviewer terminated the interview.

Health State Selection

We followed the Paris protocol and proposed a direct valuation of a total of 97 impaired EQ-5D-3L health states (included “33333”), to estimate the value sets. Except for “33333”, the 96 selected health states were categorized into three levels: mild, moderate, and severe. Mild states were those health states have no level 3 on any dimension and only have level 2 on up to three dimensions; severe states had no level 1 on any dimension; all other health states were considered moderate (more detailed information on selecting health states was described in the revised MVH protocol [20]). A total of 24 mild, 48 moderate and 24 severe states were randomly assigned into eight fixed blocks. Each block included 12 health states with three mild, six moderate, and three severe states. In addition, states 11111, 33333, and death were assigned to each block, making a total of 15 health states in each (Table 1). Health state blocks were randomly assigned to each participant during the interview.

Table 1

EQ-5D-3L Health States Distributions
Severity	Group
	A	B	C	D	E	F	G	H
Mild	11112	11221	11212	22211	21111	12221	22111	11121
	12122	11222	12211	21211	22121	21121	22112	21122
	12112	11122	12121	11211	21112	21221	12212	12111
Moderate	31213	11313	12123	21332	21331	11332	12312	23321
	31311	32123	21313	21133	11123	13222	13211	21231
	23132	21311	12313	33122	22232	12331	23311	11232
	21123	11323	23313	22221	11312	33221	32111	21312
	23231	33121	33313	13123	13232	31222	22313	31131
	22113	33211	33231	11223	23222	33312	23131	31313
Severe	22233	23233	33332	33233	33223	32332	33232	22333
	22332	23322	32322	23323	33323	23223	23332	22323
	23333	33222	32223	33322	32233	32333	32323	32232
Others	11111	11111	11111	11111	11111	11111	11111	11111
	33333	33333	33333	33333	33333	33333	33333	33333
	Death	Death	Death	Death	Death	Death	Death	Death

Data Collection

The interview process included the following tasks: 1) assessing respondent self-reported health state using the Chinese version of EQ-5D-3L instrument; 2) ranking the 15 health states in terms of their severity; 3) rating the 15 health states using a vertical hash-marked visual analogue scale (VAS); 4) evaluating the same set of health states using the TTO except for 11111 and death; 5) collecting participant’s socio-demographic information.

Each health state was described on a separate card, and the level of severity was marked by different colors: green, yellow, and red indicate no problem, moderate, and severe problem, respectively. In step 2 and 3, each health state had a 10-year duration followed by death. A double-side time board was used in step 4, the TTO exercise, to illustrate the different lengths of hypothetical lives, with one side representing states better than death and the other side representing states worse than death [21]. In order to minimize the memory effect, the cards were reshuffled at the beginning of each step.

In step 4, each TTO task started with asking participant whether health state be valued was better or worse than or equals to death. If the state was considered better than death, then the iteration process was trading off t years in state 11111 (Life A) against the 10 years in the being valued health state (Life B). If the state was considered to be worse than death, the participant was asked to compare living in the target state for (10-t) years followed by t years in the state 11111 (Life A) and immediately death (Life B). The minimal changeable unit in the iteration process was 6-month. If the state was considered as equivalent to death, the valuation of this state is completed and the interviewer would move on to the next one. The TTO utility score (U) was calculated as t/10, -t/(10-t) and 0 for better than death states, worse than death states and equivalent to death states, respectively, where the t is life years in the state 11111 when Life A and Life B were considered about the same.

A total of 61 graduate students and faculties were recruited as interviewers from one university in each one of the five cities and trained by Peking University’s China Center for Health Economic Research (CCHER). All the interviewers participated in a three-day standardization training session, where they were trained to grasp the skills and procedures for interviewing through mock exercises and were required to accomplish at least one pilot interview in the selected city.

Data analysis

Data Logic and Transformation

In line with the previous studies [14, 22], data exclusion criteria were: 1) data missed for all; 2) only 1 or 2 states were valued; 3) all states were reported with the same value; 4) all states were valued as worse than death; and 5) had logical inconsistencies for 4 or more pairs of states. Logical consistency was defined as: for a given pair of health states, if at least one dimension of state A (such as 11121) is better than the corresponding dimension in state B (such as 11123) and other dimensions are not worse than their counterparts in state B, then the valuation for the state A should be at least as good as the valuation of the state B.

Modeling of TTO Values

By design, the TTO utility values originally ranged from − 19 to 1 [23]. In order to eliminate the outlier effects of extremely low values in subsequent analyses, a linear transformation formula applied in a previous Chinese study [14], was used to rescale the negative values to range from − 1 to 0 (U^∧=U/19). State 11111 and death were anchored as 1 and 0, respectively.

The dependent variable of all models was disutility, which was defined as 1 minus the TTO value. The basic model included only 10 dummy variables present level 2 and level 3 problems for each one of the five dimensions. The following interaction terms were considered in our study: 1) N3 term, which equalled to 1 if the health state being valued included at least 1 dimension at level 3, otherwise equalled to 0; 2) D1 term, which was the number of dimensions with problems beyond the first one; 3) I2 and I3 terms, which were the number of dimensions at level 2 and level 3 beyond the first one, respectively; and 4) I2sq and I3sq terms, which were the square I2 and I3, respectively.

To estimate the TTO values on all health states, models at both aggregate and individual levels were constructed. At the aggregate level of analysis, the mean of the TTO values was used to summarize the value of each health state. Both ordinary least square (OLS) and weighted least square (WLS) regression were employed, with the number of respondents who rated a particular health state as the weight. At the individual level of analysis, Pooled OLS, fixed effect/random effect estimation model were taken into account.

The Breush-Pagan test was performed to test for heteroscedasticity and the Jarque-Bera test evaluated whether the residual term in the regression models had skewness and kurtosis consistent with abnormal distributions. A Hausman’s test was used to decide between random effects and fixed effects models. The Ramsey Regression Equation Specification Error Test (RESET) for model misspecification was also examined.

Model Selection

Four criteria were used to select the final model for the value set: 1) logical consistency; 2) sign and significance of regression coefficients: the coefficients of main effects should be statistically significant and positive, and the coefficient of level 3 was expected to be higher than that of level 2 for each dimension; 3) goodness of fit, the mean absolute error (MAE) was calculated, and the number of health states that had an absolute error greater than 0.025, 0.05 and 0.10 was estimated; and 4) parsimony, if several models performed similarly to the criteria specified above, the most parsimonious model would be preferred.

The robustness of the final model was assessed by using a split-half strategy: a subset of two-thirds of the observations was randomly selected and used to re-estimate the model [23]. The estimated coefficients were then used to generate predicted values, which were then compared with the observed values of the remaining one-third of observations. The Pearson’s product-moment correlation coefficient between the observed and the predicted values of health states is presented.

Comparison with the Urban Value Set

The preference of five dimensions in the allocation of disutility was compared among rural and urban registrants in order to analyze which dimension was the leading factor affecting health utility. The Pearson correlation analysis was used to test the correlation between the urban and rural value set. In addition, we also compared the correlation coefficient between the one-third sample’s mean observation values and the rural predicted values and that between the one-third sample’s mean observation values and the urban predicted values.

All statistical analyses were conducted using STATA/SE 11.2.

Participants Characteristics

A total of 1201 respondents completed the interview. Twenty-eight respondents were excluded due to quality issues (2 respondents gave the same values for all 13 TTO tasks and 1 valued all states worse than death, 25 respondents had four or more logical inconsistencies). As a result, a total of 1173 respondents formed the valuation sample, of which 592 (50.47%) were females. The mean (standard deviation) age was 43.21 (SD = 15.75) years, (more details about the characteristics of participants see Table 2). At the aggregate level, the means of all the 97 health states did not show any logical inconsistency in pairwise comparison.

Table 2 Socio-demographic Characteristics of Valuation Sample (N=1173)

Characteristic	Percentage	Characteristic	Percentage
Sex		Working status
Male	49.53	Formal	13.13
Female	50.47	Temporary	10.91
Age		Freelance	17.05
16-20	8.53	Retired	2.47
21-30	16.54	Student	6.31
31-40	19.27	Farmer	39.05
41-50	21.31	Unemployed	10.57
51-60	17.14	Other	0.51
60+	17.22	Monthly income†
Ethnic group		0-1000	17.14
Han	84.65	1001-5000	53.45
Minority	14.66	5001-10000	16.03
No answer	0.68	>10000	2.64
Education level		Missing	10.74
Primary and lower	24.21	Self-reported health status
High school	68.37	Very good	25.58
College and higher	7.42	Good	33.25
Marital status		Fair	37.34
Unmarried	14.07	Poor	3.41
Married	81.42	Very poor	0.43
Divorced	1.19	EQ-5D-3L any problem
Widowed	3.24	Mobility	5.29
Other	0.09	Self-care	2.30
Having chronic condition		Usual activities	4.43
Yes	25.83	Pain/discomfort	20.12
No	71.70	Anxiety/depression	11.76
Unclear	2.47

^†RMB, Renminbi.

EQ-5D-3L, three-level EuroQol five-dimensions.

Modeling

At the individual level, the Hausman’s test didn’t reject the null hypothesis of no inconsistencies between coefficients (χ² = 4.39, P = 0.9280). Therefore, we only conducted pooled OLS and random effects models rather than fixed effects estimation. At the aggregate level, the results of OLS and WLS regression are presented. We only report the main effects, N3, and D1 models results because alternative models did not perform better.

To select the best performing model, various tests were used to evaluate the specification of all models mentioned above. At the individual level, no model passed the Jarque-Bera test for normality of the residuals (P < 0.001). At the aggregate level, three models passed all of the tests, with the exception of the N3 model estimated by OLS regression (Breusch-Pagan test for heteroscedasticity, χ² = 22.91, P = 0.0182). It seems that models at the aggregate level performed better.

As shown in Table 3, at the aggregate level (the coefficients estimates and fit statistics results at the individual level of analysis are reported as Appendix 1 in Supplemental Materials), all the estimated coefficients of main effects in each model (10 dummy variables), no matter using OLS or WLS regression, are positive and significant. In N3 model, estimates for the N3 term using OLS regression were significant, reflecting the much greater disutility associated with extreme problems, but this term turned to be insignificant in WLS regression. In contrast, the D1 model included two negative interaction terms which remained consistently significant in both of the OLS and WLS regressions. One of the negative estimates was D1, which suggested a declining marginal disutility associated with additional shifts away from perfect health beyond the first dimension. The other significant term was I3, while estimates for I3sq were insignificant, indicating that the effect of having multiple dimensions at level 3 was linear. Estimates for I2 and I2sq terms were both insignificant.

Table 3 Parameter Estimates and Fit Statistics of Aggregate Level Models Using OLS and WLS Regression

Variable	Main Effects				N3				D1
	OLS		WLS		OLS		WLS		OLS		WLS
	coef.	s.e.	coef.	s.e.	coef.	s.e.	coef.	s.e.	coef.	s.e.	coef.	s.e.
Constant	0.071	0.007	0.071	0.007	0.067	0.007	0.070	0.007
MO2	0.102	0.005	0.100	0.005	0.101	0.005	0.099	0.005	0.167	0.007	0.166	0.007
MO3	0.279	0.007	0.280	0.007	0.275	0.007	0.279	0.007	0.365	0.014	0.370	0.014
SC2	0.102	0.005	0.101	0.005	0.103	0.005	0.101	0.005	0.169	0.007	0.169	0.007
SC3	0.242	0.006	0.244	0.006	0.239	0.007	0.243	0.007	0.330	0.015	0.336	0.015
UA2	0.087	0.006	0.085	0.006	0.086	0.006	0.084	0.006	0.151	0.007	0.150	0.007
UA3	0.222	0.006	0.223	0.006	0.217	0.007	0.222	0.007	0.308	0.013	0.313	0.014
PD2	0.110	0.006	0.110	0.006	0.110	0.006	0.109	0.006	0.175	0.007	0.175	0.007
PD3	0.237	0.006	0.240	0.006	0.232	0.007	0.239	0.007	0.323	0.014	0.329	0.014
AD2	0.075	0.005	0.074	0.005	0.074	0.005	0.073	0.005	0.139	0.009	0.138	0.009
AD3	0.177	0.006	0.180	0.006	0.172	0.007	0.178	0.007	0.262	0.014	0.267	0.014
N3					0.016^‡	0.009	0.005^§	0.009
D1									-0.073	0.013	-0.077	0.014
I2									0.009^§	0.017	0.015^§	0.018
I2sq									-0.000^§	0.003	-0.00^§	0.003
I3									-0.022^‡	0.013	-0.028^†	0.013
I3sq									0.001^§	0.003	0.003^§	0.002
Fit statistics
Adjusted R²	0.993		0.995		0.993		0.995		0.999		0.999
MAE	0.018		0.017		0.017		0.017		0.017		0.017
No. (of 97)>0.025	28		29		27		28		27		28
No. (of 97)>0.05	2		3		2		3		2		2

P<0.01 and Heteroskedasticity-robust standard error for all regression coefficients unless otherwise stated. ^†0.01≤P≤0.05, ^‡0.05<P≤0.1, ^§P>0.1.

The results of the main effects, N3, and D1 models from modeling of aggregate-level data using WLS regression were generally worse than those based on OLS regression. Specifically, the number of states with an absolute error greater than 0.025 in each model using WLS regression was one more than the number using OLS regression, which potentially resulted from the outweighed number of values for state “33333” (n = 1173, SD = 0.41). Finally, the N3 model based on OLS regression with robust standard error at the aggregate level was selected as the best performing model. The selected model passed the Jarque-Bera test for normality of the residuals (χ² = 5.99,P = 0.6582). There was no model or functional form misspecification as suggested by the Ramsey RESET test (F = 0.03, P = 0.9916). Figure 1 shows the estimated values plotted against the mean observed TTO values for the 97 health states used in this model (see Appendix 2 to find the utility of 243 health states).

Robustness

Figure 2 presents each health state’s the mean observed value of the one-third sample and the predicted values for the two-thirds sample in the split-half validation process. The correlation coefficient was 0.9937.

Comparison with Urban Study

The N3 model based on an OLS regression at the aggregate level turns out to be the best performing model. The correlation coefficient between the rural value set predicted values for the 242 impaired health states and the urban value set predicted values was 0.9922. As shown in Fig. 3, the predicted values of the two value sets generally showed the same trend regardless of small fluctuations, however, the predicted values of 242 impaired health states for the rural final model were generally lower than that of the corresponding health states for the urban model, Especially the predicted value of state 33321 for rural registrants is 0.124 lower than that for urban counterparts. Among all 242 impaired health states, only 13 states yield higher value in the rural model than the urban model.

Of the five EQ-5D dimensions in levels 3, the marginal effect of mobility on health utility is bigger than the other four dimensions in both rural and urban respondents. The dimensions at level 3 exerting the least influence on health utility are different between rural and urban respondents; the former is anxiety/depression, while the latter is usual activities.

The correlation coefficient between the mean observed values of all health states in the one-third validation sample and the urban model predicted values was 0.9892 which was smaller than that of the mean observed values in the validation sample and the rural model predicted values (r = 0.9937) .

This study applied a similar study design and research methodology that has been proved to be effective in previous urban study to analyzing the health preferences of the rural population particularly, which is of great importance for a country with significant urban-rural differences like China. This is the first study to provide a set of weights for the 243 EQ-5D health states based on the preferences of Chinese rural registrants. The N3 model based on an OLS regression at the aggregate level was finally selected as the tariff for the rural population.

The comparison between rural and urban studies shows a convergence of health state valuation between the two subgroups. Nevertheless, differences do exist: 1) residents with rural registration are generally more sensitive than urban registrants when facing the same health problems, especially when faced with extreme problems associated with physical condition and self-care; 2) rural registrants are less concerned with mental problems like anxiety and depression compared to urban registrants. These differences reflect an existing urban-rural gap in health preferences shaped by socioeconomic and institutionalized disparities, and hopefully could shed light on more efficient health policies and welfare package design with more accurate health valuations.

The generally lower utilities of 243 health states for rural registrants compared to their urban counterparts suggest a tendency to trade length of life for quality of life among rural registrants, due largely to two reasons. On the one hand, although covering more than 90% rural registrants (comprising more than half of China's total population) since the full scale-up of deepening health system reform in 2009, the New Rural Cooperative Medical Scheme has a relative low financing and security level compared to the health insurance scheme for urban population. While the critical illness insurance program being introduced successively, its fundraising process is actually pro-rich. In addition, other social welfare and government assistance policy for rural registrants is not as comprehensive as that for urban residents, resulting in severe dilemma for the rural of falling into or back to poverty due to illness. Therefore, when trading off between the quality of life and the length of life, rural registrants prefer to keep healthy life to avoid potential negative impact caused by the incomprehensiveness of health and social security system. On the other hand, most rural registrants in China are manual-labor workers whose income and life rely heavily on their health. Therefore, they highly value healthy life to ensure stable and substantial income, rather than a long length of life with illness that may result in income reduction and economic burden to their family.

The major limitation in this study is that we could not making accurate interpretations of comparison. Although the same study design and research methodology as urban study was adopted in the rural study, we cannot easily identify whether the difference between rural and urban registrants is a result of genuine differences in health preference due to different final models. Sensitivity analysis should be performed to evaluate the impact of using different models to generate preference weights for the EQ-5D-3L health states. In addition, utility values predicted by Zhou’s mixed population value set are the highest, followed by Liu’s urban value set, and this rural value set predicted the lowest utility values among the three value sets. So, the previous hypothesis that the rural population should be less willing to trade-off life years than the urban population may not be correct. Thus, further investigations on the Chinese urban and rural population’s health preferences are necessary.

Despite these limitations, the value set generated by the current study further supplements previous population-based studies that focused only on urban Chinese registrants by targeting the rural areas. Moreover, the research team largely comprises the researchers engaged in previous urban study whose experience have laid solid foundation for this study, and this serves as a reasonable explanation for better predictive capability of the rural model. This EQ-5D-3L value set is at present the best available EQ-5D-3L value set for health technology assessment and cost-utility analysis for the rural Chinese population.

Considering the difference of EQ-5D-3L health state preference between the Chinese rural and urban population, the rural tariff established in the present study should be used in health surveys of the rural population in the future.

Acknowledgements

We sincerely thank all the participants and interviewers who made this study possible.

Authors’ contributions

All authors contributed to the study design, or data analysis and interpretation; and read and approved the content of the manuscript and take responsibility for the study.

Funding

This study was supported by the National Natural Science Foundation of China (grant no.71273015 and no.71463007), and the Guizhou Science and Technology Department (2013 no.3068).

Availability of data

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Conflict of interest

The authors declare no conflicts of interest exist.

Ethics approval and consent to participate

No personal information was collected to protect privacy and all interviews were agreed by respondents.

Consent for publication

Not applicable.

1. Robinson R. Cost-utility analysis. BMJ. 1993;307(6908):859-862. https://doi.org/10.1136/bmj.307.6908.859.

2. Tsuchiya A, Ikeda S, Ikegami N, Nishimura S, Sakai I, Fukuda T, et al. Estimating an EQ-5D population value set: the case of Japan. Health Econ. 2002;11(4):341-353. https://doi.org/10.1002/hec.673.

3. Pullenayegum EM, Sunderland KM, Johnson JA, Xie F. Handling regional variation in health state preferences within a country: a methodological framework. Med Decis Making. 2017;37(3):252-261. https://doi.org/10.1177/0272989X16651885.

4. Devlin NJ, Shah KK, Feng Y, Mulhern B, Hout BV. Valuing health-related quality of life: an EQ-5D-5L value set for England. Health Econ. 2018;27(1):7-22.

https://doi.org/10.1002/hec.3564.

5. Johnson JA, Luo N, Shaw JW, Kind P, Coons SJ. Valuations of EQ-5D health states: are the United States and United Kingdom different. Med Care. 2005;43(3):221-228.

https://www.jstor.org/stable/3768220.

6. Lamers LM, McDonnell J, Stalmeier PFM, Krabbe PFM, Busschbach JJV. The Dutch tariff: results and arguments for an effective design for national EQ-5D valuation studies. Health Econ. 2006;15(10):1121-1132. https://doi.org/10.1002/hec.1124.

7. National Bureau of Statistics. National Bureau of Statistics of the People's Republic of China [in Chinese]. http://data.stats.gov.cn/easyquery.htm?cn=C01. Acceesd 1 July 2021.

8. National Bureau of Statistics. Statistical bulletin on national economic and social development in 2020 [in Chinese]. http://www.stats.gov.cn/tjsj/zxfb/202102/t20210227_1814154.html. Acceesd 1 July 2021.

9. National Bureau of Statistics. The 2010 population census of the People’s Republic of China [in Chinese]. http://www.stats.gov.cn/tjsj/pcsj/rkpc/6rp/indexch.htm. Acceesd 1 July 2021.

10. National Bureau of Statistics. China Statistic Yearbook 2020 [in Chinese]. http://www.stats.gov.cn/tjsj/ndsj/2020/indexch.htm. Acceesd 1 July 2021.

11. National Health Commission of the People's Republic of China. China Health Statistic Yearbook 2020 [in Chinese]. Beijing: Peking Union Medical Clooege Press; 2021.

12. Liu GG, Hu SL, Wu JH, Wu J, Dong ZH, Li HC, et al. China Guidelines for Pharmacoeconomic Evaluations 2020 (Chinese-English Version). Beijing: China Market Press; 2020.

13. Luo N, Liu GG, Li MH, Guan HJ, Jin XJ, Rand-Hendriksen K. Estimating an EQ-5D-5L value set for China. Value Health. 2017;20(4):662-669. https://doi.org/10.1016/j.jval.2016.11.016.

14. Liu GG, Wu HY, Li MH, Gao C, Luo N. Chinese time trade-off values for EQ-5D health states. Value health. 2014;17(5):597-604. https://doi.org/10.1016/j.jval.2014.05.007.

15. Zhuo L, Xu L, Ye JT, Sun S, Zhang YG, Burstrom K, et al. Time trade-off value set for EQ-5D-3L based on a nationally representative Chinese population survey. Value Health. 2018;21(11):1330-1337. https://doi.org/10.1016/j.jval.2018.04.1370.

16. Janssen MF, Bonsel GJ, Luo N. Is EQ-5D-5L better than EQ-5D-3L? A head-to-head comparison of descriptive systems and value sets from seven countries. Pharmacoeconomics. 2018;36(6):675-697. https://doi.org/10.1007/s40273-018-0623-8.

17. You RX, Liu JY, Yang ZH, Pan CW, Ma QH, Luo N. Comparing the performance of the EQ-5D-3L and the EQ-5D-5L in an elderly Chinese population. Health Qual Life Outcomes. ;2020;18:97. https://doi.org/10.1186/s12955-020-01324-0.

18. Liang ZR, Zhang TT, Lin TF, Liu LS, Wang BY, Fu AZ, et al. Health-related quality of life among rural men and women with hypertension: assessment by the EQ-5D-5L in Jiangsu, China. Qual Life Res. 2019;28(8):2069-2080. https://doi.org/10.1007/s11136-019-02139-3.

19. Yang F, Jiang S, He XN, Li HC, Wu HY, Zhang TT, et al. Do rural residents in China understand EQ-5D-5L as intended? Evidence from a qualitative study. PharmacoEconomics. 2021;5(1):101-109. https://doi.org/10.1007/s41669-020-00212-z.

20. Kind P. A revised protocol for the valuation of health states defined by the EQ-5D-3L classification system: learning the lessons from the MVH study. Centre for Health Economics: University of York, 2009.

21. Furlong W, Feeny D, Torrance G, Barr R, Horsman J. Guide to Design and Development Health-State Utility Instrumentation. McMaster University Centre for Health Economics and Policy Analysis, CHEPA Working Paper, 1990. Available from: http://ideas.repec.org/p/hpa/wpaper/199009.html. Acceesd 1 July 2021.

22. Lee YK, Nam HS, Chuang LH, Kim KY, Yang HK, Kwon IS, et al. South Korean time trade-off values for EQ-5D health states: modeling with observed values for 101 health states. Value Health. 2009;12(8):1187-1193. https://doi.org/10.1111/j.1524-4733.2009.00579.x.

23. Dolan P. Modeling valuations for EuroQol health states. Med Care. 1997;35(11):1095-1108. https://doi.org/10.1097/00005650-199711000-00002.

Rural Population’s Preferences Matter: A Value Set For The EQ-5D-3L Health States For China’s Rural Population

Status:

Journal Publication

Version 1

Abstract

Figures

Key Points For Decision Makers

Background

Methods