Introduction
Fiscal sustainability of healthcare has become a key challenge of developed economies, resulting in an increasing focus on the efficiency of public spending on healthcare [
1]. To tackle this challenge, among several policy options, decision-makers may need to be increasingly selective when defining the benefit basket covered by public reimbursement systems [
1].
Can this challenge be met by considering the preferences of the general population as a beacon for decision-making?
Standard economic analysis uses perfect health as a reference point for health gains and applies the same weights to QALYs across the full range of disease severity or age of the target population. However, when making decisions about the public funding of various treatment options in scarcity of healthcare resources, alternative approaches may be considered, which differentiate QALY gains on the basis of various theories of distributional justice [
2]. The ‘fair innings’ principle argues on the grounds of egalitarian ideology about placing greater weights on QALYs in younger individuals, who have not yet had a fair share of lifespan as compared to older individuals [
3]. The ‘worse off’ principle on the grounds of prioritarian ideology favours QALY gains in more severe disease states over less severe ones [
4], while other principles may favour the maximisation of lifetime QALYs or interventions with the greatest benefit [
5]. Using sufficientarian reasoning, Wouters et al. recently explored the principle of differentiating QALY gains based on the acceptability of health states. Instead of being concerned with inequalities, sufficientarianists propose that it is morally important for everyone to have just enough [
6]. It has been shown in empirical studies on the Dutch general population [
7,
8] and a sample of Hungarian patients with rheumatoid arthritis (RA) [
9] that people have internal reference points, against which they compare the acceptability of certain health states via a simple acceptable—not acceptable judgement. The reference points depend on age, suggesting that more health problems are acceptable in older ages [
7‐
9]. According to sufficientarian reasoning, it is desirable to live above the acceptability threshold. Therefore, highest utility could be attached to health gains, that move patients from unacceptable to acceptable health states (AHSs), while depending on the application other sufficientarian criteria, zero or lower utility level could be attached to those health gains below or above the acceptability threshold, that do not cross the reference line [
2].
The relative simplicity of the cognitive evaluation task makes the measurement of AHSs a compelling approach. The EQ-5D-3L descriptive system [
10] provides a standardised framework for measuring AHSs, which is an important feature for public decision-making [
11]. However, rendering a binary acceptable/not acceptable status to all 243 discrete health states across several age-groups poses a feasibility challenge for the comprehensive evaluation of AHSs. To overcome these challenges, previous studies investigated AHSs in different ages separately by each dimension of the EQ-5D-3L [
7‐
9]. These studies left uncertainty about the acceptability of simultaneous health problems in more than one dimension. The acceptability of combined health problems was jointly evaluated using only a few health profiles by Wouters et al. [
7]. Although acceptable health may serve as a reference point for priority setting, it requires further methodological exploration. Wouters et al. pointed out that despite the appealing concept, the challenge of finding a morally acceptable and practically feasible acceptability threshold hampers its application in real practice [
2,
7].
Building on the results and unanswered questions of previous empirical research using the descriptive system of EQ-5D-3L instrument [
7,
8], we explored AHSs using the EQ VAS. The EQ VAS is a feasible and reliable instrument for the evaluation of health states [
12]. We assumed that although the EQ VAS does not inform about the details of subjective criteria when evaluating the acceptability of health states, it may provide meaningful point estimates about where the internal reference health states fall relative to the extremes of best and worst imaginable health. Furthermore, the EQ VAS provides a single score about the individuals’ global evaluation of health, not affected by the properties of the EQ-5D-3L descriptive system and index values, which reflect the average societal preferences attached to discrete health profiles [
12].
Our primary goal was to explore AHSs using the EQ VAS instrument and to analyse the differences compared to the assessment based on the descriptive system of the EQ-5D-3L. Moreover, we aimed to explore how AHSs are influenced by the health status and socio-demographic characteristics of respondents. Another novelty of our study is that we investigated how the reference person imagined by respondents during the evaluation exercise affected the acceptability of health states.
Methods
Study sample and design
Between January and March 2018, we conducted a cross-sectional survey on a convenience sample of 200 subjects from the Hungarian general population. We obtained the approval of the Medical Research Council of Hungary (ID: 5111-2/2018/EKU). Respondents were informed, and provided written consent. Data were collected anonymously. The acceptability of health problems was assessed via computer-aided personal interviews separately by EQ-5D-3L dimensions as well as by joint evaluation of EQ-5D-3L profiles using an adaptive testing algorithm. We recorded socio-demographic and health-related data using a paper-and-pencil questionnaire including the EQ-5D-3L, as well as the evaluation of AHSs by an adapted version of EQ VAS. The electronic and paper questionnaires were joined by a common code, retaining the anonymity of respondents.
Socio-demographic and health-related data
Respondents’ age and gender were recorded, and three main age groups (18–43, 35–64 and 65+ years old) were formed for the analysis. The three education categories (low: primary, middle: secondary, high: tertiary) were based on the highest completed level of education. We recorded the lifespan of close relatives as well as respondents’ informal caregiver experience. We assigned informal caregiver status to individuals who provided at least 6 weeks of informal care over the past 10 years. We also asked health-related lifestyle questions: weight and height for body mass index (BMI), smoking, alcohol intake and physical activity. We considered the following lifestyle parameters as risky: BMI ≥ 25 [
13], smoking at any quantity [
14], ≥ 7 drinks per week or ≥ 3 drinks per any single day for women and ≥ 14 drinks per week or ≥ 4 drinks per any single day for men [
15], and moderate physical activity < 150 min/week [
16].
The EQ-5D-3L instrument
We recorded respondents’ current health status using the paper-based validated Hungarian version of the EQ-5D-3L instrument [
10]. EQ-5D-3L is a generic quality-of-life instrument, which consists of two parts [
10]. The descriptive system assesses self-reported health in five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Respondents are asked to describe their current health in each dimension with one of the following three categories: no problems, some problems and severe problems. The descriptive system defines 243 (3
5) distinct health states, denoted by a five-digit profile comprised of the problem levels in each dimension. (e.g. 21,113 indicates moderate problems with mobility and severe problems with anxiety/depression with no problems in other dimensions.) The EQ-5D-3L index scores (utility values) attached to each health state reflect the preferences of the general population. The EQ-5D-3L index score of 1 represents perfect health, 0 represents death, and negative values represent “worse than dead” health states [
17]. Due to the lack of a Hungarian EQ-5D-3L value set, we used the time trade-off (TTO)-based value set from the UK, which is the most frequently applied EQ-5D-3L value set in the Central and Eastern European (CEE) region [
17,
18].
The second part of the instrument is a 20-cm visual analogue scale (EQ VAS) ranging from 0 (worst imaginable health) to 100 (best imaginable health). While the EQ-5D-3L descriptive system measures core dimensions of health-related quality of life to provide a single index that reflects preferences of the general population, the EQ VAS reflects the self-rating of the respondents’ overall health including aspects without limiting to the five EQ-5D-3L dimensions [
19]. We assigned respondents with one standard deviation (SD) below the mean EQ VAS of the sample to the “poor health” group, respondents within ± 1 SD around the mean to the “average health” group, and respondents with one SD above the sample mean to the “good health” group.
Measuring acceptable health states
We performed computer-assisted personal interviews to assess the acceptability of health states in different ages between 30 and 80 years in 10-year intervals. To allow a clear separation of these ages from respondents’ own age, we will use “ageAHS” notation when referring to the hypothetical ages used to assess AHSs. In addition to separate evaluation of AHSs by the dimensions of EQ-5D-3L, we applied two novel evaluation methods: (1) joint evaluation of the acceptability of EQ-5D-3L health profiles using an adaptive testing algorithm, and (2) assessing AHSs by the EQ VAS. Reporting the results of the adaptive testing and joint evaluation is beyond the scope of this paper.
We assessed AHSs via separate evaluation of problems by dimensions of the EQ-5D-3L descriptive system according to the methods applied in previous studies on the Dutch general population [
7,
8] as well as in Hungarian patients with rheumatoid arthritis [
9]. In short, respondents were asked to indicate beyond what age they consider different levels of problems acceptable in each dimension of the EQ-5D-3L. The sample question for the mobility dimension is depicted in the Online Resource (Supplementary Fig. S1). For each respondent, we constructed acceptable health curves (AHCs) by the methods described by Wouters et al. [
7]. First, we assumed that all health problems, that were considered acceptable in a certain age
AHS separately, would also be acceptable in combination. Therefore, we aggregated the individual responses on each EQ-5D-3L domain into a single EQ-5D-3L health profile for each age
AHS, and attached the EQ-5D-3L index value for these aggregated health profiles. We will refer to this method using the term “aggregate acceptable health curve (AHC
aggregate)” [
7‐
9]. Alternatively, we assumed that respondents would only consider health problems acceptable in each domain with perfect health in mind for the remaining four domains. Therefore, for each age
AHS, we also constructed AHCs using the lowest EQ-5D-3L index value among the possible EQ-5D-3L profiles having acceptable problems in a single dimension. We will refer to this method using the term “worst acceptable health curve (AHC
worst)”. For example, if a respondent indicated moderate mobility problems and severe anxiety/depression acceptable from age
AHS 60 during separate evaluation, then the 0.345 EQ-5D-3L index of the aggregate profile 21,113 would be used in AHC
aggregate; while from the profiles 21,111 and 11,113 with EQ-5D-3L index values of 0.85 and 0.414, respectively, lower index would be chosen to construct AHC
worst. In previous studies, the difference between the AHC
aggregate and AHC
worst was substantial in ages above 60 years suggesting that these AHCs deviate from the true acceptability threshold [
7,
8].
To determine the location of the acceptability threshold, we adapted the EQ VAS instrument. In health valuation studies, multiple health states were recorded on a single EQ VAS [
20]; therefore, we asked respondents to indicate the health state that is still acceptable in different ages on the same EQ VAS. To avoid suggesting a ranking on the vertical EQ VAS, we placed ages on a horizontal line at the midpoint of the EQ VAS and asked respondents to link each age and the VAS with a line. The adapted EQ VAS is shown in the Online Resource (Supplementary Fig. S2). We constructed AHCs from the acceptable EQ VAS scores at each age
AHS, for which we use the term “VAS acceptable health curve (AHC
vas)”.
At the end of the evaluation task, we asked respondents about whom they imagined when evaluating the acceptability of health states. Based on the reference person imagined, respondents were assigned to three categories: those who thought about themselves (“reference: oneself”), those who answered with reference to others (“reference: others”), and those who had both themselves and others in their mind (“reference: mixed”). The joint evaluation of the acceptability of EQ-5D-3L profiles using the adaptive testing algorithm will be described elsewhere.
Statistical analysis
We summarised the key sample characteristics using descriptive methods. Then, we estimated the AHC
vas age profiles using a multilevel regression model. In the baseline model (Model 1, Eq.
1), AHC
vas was described by an intercept and a slope parameter, which were allowed to vary across individuals according the following equation:
$${\text{AHC}}_{\text{vasik }} = \alpha + \beta {\text{age}}_{\text{AHSik}} + \mu_{\text{i}} + \tau_{\text{i}} \; \times \;{\text{age}}_{\text{AHSik}} + \varepsilon_{\text{ik}} ,$$
(1)
where
ageAHSik denotes age when acceptable health is evaluated by respondent
i for the
kth age. We centred age
AHS on 30 years; so, the intercept denoted by represents the population average level of acceptable health at age 30 (AHC
vas30) and
μ represents the individual-specific shifts in AHC
vas30. The slope denoted by
β represents the average acceptable deterioration rate of health with age (ADR) and
τ stands for the individual-specific component of ADR. In the model,
μ and
τ were not estimated directly, but they were modelled as random effects, with only the variance of these parameters estimated.
To evaluate the effect of individual characteristics on acceptable health, the model was augmented by level-2 variables as follows:
$${\text{AHC}}_{\text{vasik}} = \alpha + \beta {\text{age}}_{\text{AHSik}} + \gamma X_{\text{i}} + \delta {\text{age}}_{\text{AHSik}} \; \times \;X_{\text{i}} + \mu_{\text{i}} + \tau_{\text{i}} \; \times \;{\text{age}}_{\text{AHSik}} + \varepsilon_{\text{ik}} ,$$
(2)
where
Xi is the vector of individual characteristics of respondent
i, the vectors of coefficients are
γ and
δ representing the effects of
Xi, respectively, on the intercept and the slope. While holding
X constant at the reference values,
α and
β represent the intercept and slope parameters, respectively. The individual-specific intercept and slope components that are not explained by
X are represented by
μ and
τ, respectively.
We explored the effect of four sets of individual characteristics on the AHCvas. The model contained basic demographic characteristics, such as gender, age and education (Model 2). Respondents’ EQ VAS score and health-related lifestyle variables were included as a proxy for “experience with own health” (Model 3). We also included dummy variables to indicate respondents’ reference person when evaluating AHSs (Model 4). Finally, we explored the effect of the lifespan of close relatives and caregiver status as indicators for the “experience with others health” (Model 5).
In addition to the linear multilevel model, we also tested alternative AHCvas models. We evaluated model fit by comparing residual variance (\(\sigma_{\varepsilon }^{2}\)) and Akaike’s information criterion (AIC). First, we compared Model 1 to a restricted specification including only a random intercept, but no random slope term, and a simple linear regression model containing no random terms. Model fit was substantially inferior for these restricted specifications (\(\sigma_{\varepsilon }^{2}\) = 58.2, AIC = 8438 with random intercept only and \(\sigma_{\varepsilon }^{2}\) = 199.7, AIC = 9320 with no random term, while \(\sigma_{\varepsilon }^{2}\) = 17.2, AIC = 7657 for Model 1).
Second, we also compared Model 1 to a more general specification, allowing for nonlinear effects of age
AHS:
$${\text{AHC}}_{\text{vasik}} = \alpha + \beta {\text{age}}_{\text{AHSik}} + \theta {\text{age}}_{\text{AHSik}}^{2} + \mu_{\text{i}} + \tau_{\text{i}} \; \times \;{\text{age}}_{\text{AHSik}} + \vartheta_{\text{i}} \; \times \;{\text{age}}_{\text{AHSik}}^{2} + \varepsilon_{\text{ik}} .$$
(3)
Although the likelihood ratio test was significant (\(\chi_{{\left( {{\text{df}}\; = \;4} \right)}}^{2} = 187.1, \;p\; < \;0.0001)\) indicating better fit for the quadratic model vs the linear one (Model 1), the difference between predicted values were negligible, and the gain in fit was modest (\(\sigma_{\varepsilon }^{2}\) = 10.9, AIC = 7478). Therefore, we favoured the linear model for more straightforward interpretation of the parameters.
Discussion
To our knowledge, this is the first study that explored acceptable health states using the EQ VAS instrument. We found that the modified EQ VAS was a feasible and convenient measure of AHSs for vast majority of our respondents. Our findings confirm previous research results that people find worsening of health with age acceptable [
7‐
9]. The core hypothesis is that respondents have internal age-dependent reference points against which they judge the acceptability of health states.
Previous studies measured AHSs using the descriptive system of the EQ-5D-3L instrument. In these studies, AHSs were assessed separately by the five dimensions of the EQ-5D-3L, and assumptions were made about the acceptability simultaneous health problems in more than one EQ-5D-3L dimension. Although the acceptability of a few EQ-5D-3L health states profiles was also evaluated, current knowledge is limited about how people jointly evaluate the acceptability of simultaneous problems in multiple dimensions. One of the benefits of using the EQ VAS is that it provides a single score about where acceptable health states lie compared to the best and worst imaginable overall health status. The moderate slope of the AHC
vas with age suggested that by evaluating the acceptability of health states globally, in older ages people are less likely to accept as many combined problems as suggested by the AHC
aggregate. Moreover, the dispersion of AHC
vas was well below the variation of EQ-5D-3L descriptive system-based AHCs, suggesting that using EQ VAS may allow for defining and detecting acceptability thresholds more precisely. While the dispersion of AHC
vas was similar to that of the measures of current health, the greater dispersion of AHC
aggregate suggests that considerable measurement error arises from the construction of AHCs from the separately evaluated AHSs by dimension. However, since the AHC
aggregate is artificially constructed from EQ-5D-3L index values, which reflect societal preferences rather than each respondents’ global evaluation of individual health [
20], the AHC
aggregate and AHC
vas are not directly comparable measures of AHSs.
Although it may be hypothesised that underlying health preferences influence the acceptability of health problems, the evaluation exercise of acceptable health problems is different in many aspects from health-state valuation tasks: it does not involve assumptions about death, expected lifespan, nor trade-off or risk-evaluation is involved [
12,
23]. Although the cognitive processes behind the evaluation exercise of acceptable health are yet to be explored, our results confirm previous findings that respondents’ age, health status, lifestyle, and other health-related experiences, such as informal caregiver status or the lifespan of close relatives influence the internal reference for acceptable health.
We found that ceteris paribus, the younger respondents reported in younger ages more, and in older ages less health problems acceptable compared to older ones. Other studies demonstrated greater acceptability of health problems by elderly respondents in the general population [
7], and in patients with RA [
9]. In our study, worse current health of respondents was associated with more acceptable health problems. In the Dutch general population study, suffering from a severe disorder or having a chronic condition was not associated with the acceptability of health problems [
7], while lower EQ VAS scores of Hungarian RA patients predicted more acceptable health problems [
9]. Patients’ adaptation to their conditions affects the valuation of health states [
24] which may also increase the acceptability of non-perfect health by older individuals experiencing health problems.
In addition to older respondents, the younger age group also deserves attention. The relationship of age and health-state utilities was bell-shaped in the UK TTO EQ-5D valuation study [
25] and we found increased acceptability of health problems in the 18–34-year-old age group. Perceived severity of health outcomes may play a role in the prevention of health-risk behaviours [
26], an important health concern among young adults [
27]. Acceptability of health problems may be linked to the perception of health risks by younger adults, although this hypothesis needs empirical testing. Therefore, the complex interplay between the effects of respondents’ health and age on the acceptability of health problems requires further investigation, which may provide useful insights to health valuation as well as health prevention research.
Lifestyle-related variables influenced the acceptability of health problems in both our sample and the Dutch general population, albeit in different ways. In the Dutch population, healthy diet was associated with significantly fewer acceptable health problems. In our sample, lack of exercise was associated with more (greater ADR), while high BMI and high-risk alcohol intake were associated with less acceptable health problems. Longer lifespan of close relatives was associated with more acceptable health problems in our study, while had opposite effect in the Dutch study. We also found that being a caregiver was associated with less AHSs (via lower ADR). These findings warrant deeper qualitative investigation of the experiences and attitudes about own and others’ health and the context that shape the acceptability of health problems.
We find important to note that the effects of explanatory variables need to be interpreted in the light of the potentially different measurement properties of our methods as well as different preferences of the studied populations. Wouters et al. recruited an online sample of the Dutch general population (
n = 1067), and constructed an area under curve (AUC) AHC
aggregate values between 40 years of age
AHS and the expected lifespan of respondents to study the effect of explanatory variables on AHSs [
7]. We conducted computer-assisted interviews in the Hungarian population, and evaluated the level and slope parameters of the AHC
vas separately as measures of AHSs.
In addition to the four scenarios of health preference evaluation (general public vs patients, own vs hypothetical health states), the evaluation of AHSs in different ages involves a new hypothetical situation [
23]. Our study provided insights about people’s opinion formulating mechanisms when evaluating AHSs. Despite the neutral question, approximately half of the respondents imagined themselves, one-third others and one-fifth both themselves and others during the evaluation task. Mulhern et al. [
28] found similar respondent subtypes when evaluating hypothetical health states. We also found that the reference person respondents imagined affected the way respondent’s age and health status influenced the evaluation of AHSs, which warrants further investigation as well as methodological refinement of the evaluation exercise.
Our research has three important limitations. First, our sample was a small non-representative sample of the Hungarian population, which limits the generalisation of our findings. Second, although the EQ VAS has been validated for measuring the current health status of responders, its psychometric properties has not been formally tested and validated for the measurement of AHSs. For instance, respondents may have provided systematic answers on the single VAS scale, which needs further exploration. Also, because some authors raised theoretical concerns about using the EQ VAS for health economic evaluation [
12] the theoretical and ethical basis how VAS-based measures of AHSs could inform decision-making have yet to be determined. However, our results showed that EQ VAS has promising properties in the evaluation of AHSs, warranting further exploration of the acceptable health concept.
Conclusions
We measured AHSs in different ages by adapting the EQ VAS, which proved to be a convenient and feasible measure. Our findings confirmed that health problems are increasingly acceptable with age. The comparison of VAS-based and EQ-5D-3L index-based measures suggested that the amount of acceptable health deterioration with age depends on the measurement method. In our sample, respondents’ age, health status, lifestyle, and other health-related experiences influenced the evaluation of acceptable health problems. We also found that people imagined themselves, others or both during the evaluation of acceptable health, which also influenced the evaluation of AHSs. However, our non-representative sample limits the generalizability of our findings.
AHSs may indicate societal preferences about the severity of disease and age in a single measure, and may serve as a reference point in healthcare priority setting. However, measuring AHSs need further theoretical and methodological exploration, before their practical application can be considered.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.