EQ-5D-5L-based quality of life normative data for patients with self-reported diabetes in Poland

Introduction The new, five-level EQ-5D generic questionnaire (EQ-5D-5L) has never been used among diabetes patients in Poland. Objectives To develop health-related quality of life (HRQoL) norms for patients with self-reported diabetes, based on a large representative sample of the general Polish population, using the EQ-5D-5L. Materials and methods Members of the general public, selected via multistage stratified sampling, filled in the EQ-5D-5L questionnaire and answered a question about the presence of diabetes. We estimated three types of EQ-5D-5L outcomes: limitations within domains, EQ VAS and EQ-5D-5L index. Multiple linear regression was used to examine the relationship between sociodemographic characteristics and HRQoL, both in patients with diabetes and the general population sample. Results Among 2,973 respondents having complete EQ-5D-5L data, 255 subjects (8.6%) self-reported diabetes. Treatment with insulin, other drugs, combination therapy or lack of drug treatment was declared by 22.0%, 48.6%, 5.1% and 24.3% of patients, respectively. Respondents with diabetes had a lower EQ VAS score (18.5 points difference on a 100-points scale) and a lower EQ-5D-5L index score (0.135 difference; scale range: 1.59). The multivariate analysis showed that the factors independently improving the HRQoL in the general population were secondary or higher education, and factors reducing HRQoL were female sex, belonging to an older age group, being treated because of diabetes with insulin, other drugs or combination treatment. Respondents diagnosed with diabetes but not treated with drugs showed a decrease in EQ VAS scores, but not in the EQ-5D-5L index. Conclusions Diabetes leads to HRQoL deterioration in all age groups when compared to matched general population respondents without diabetes. The most significant HRQoL reduction experience older patients with a basic level of education. Obtained EQ-5D-5L normative data may be used in the clinical care of patients with diabetes and health technology assessment of new anti-diabetic drugs.

predetermined study sample was proportionally allocated into layers, so as to reflect the general population structure. Multistage random sampling was carried out at three successive levels of granularity: (1) towns/cities and villages; (2) small areas (one or several adjacent streets) within the previously drawn localities; (3) according to the Polish Resident Identification Number (PESEL)-a sample of eight people living in separate dwelling/household from each of the selected areas.
The need for ethics approval for this study was waived by the Bioethical Commission of the Medical University of Warsaw (AKBE/95/2019). Written informed consent was not required for participation in the study. Oral consent was obtained. The data were analyzed anonymously.

Survey
The survey consisted of three sections in the following order: (I) sociodemographic questions, (II) self-reported presence of diabetes and (III) quality of life section (EQ-5D-5L, SF-12 and EQ-5D-3L questionnaires). In the current paper we are focusing on EQ-5D-5L results. The SF-12 and EQ-5D-3L outcomes were described elsewhere [29,30]. The current study was run as a part of a larger survey (an Omnibus study).
Sociodemographic questions covered the following: type of locality, voivodeship (province), level of education, occupational status, household income, religiosity and smoking habits. We classified respondents as having self-reported diabetes if, in response to the following question: "Have you ever been diagnosed with diabetes?", they chose one of the following answers: (1) "Yes, but I don't take any medication", (2) "Yes, I take anti-diabetic medication (other than insulin)" or (3) "Yes, I take insulin". Respondents were allowed to choose both answers (2) and (3) when they were on combined treatment.
The study used the EQ-5D-5L questionnaire, which consists of two parts: a descriptive system and a visual analog scale (EQ VAS) [31]. The descriptive part comprises five dimensions: mobility (MO), self-care (SC), usual activities (UA), pain/discomfort (PD) and anxiety/depression (AD). Each of the EQ-5D-5L items has five possible levels, of which four are common to all dimensions: (1) no problems, (2) slight problems, (3) moderate problems and (4) serious problems. The fifth answer for the dimensions MO, SC and UA was formulated as incapacity, and for PD and AD as an extreme feeling. Five scales with five possible answers result in a total of 3,125 possible health states.
Additionally, based on the respondent's answers, a weighted measure of health may be calculated-EQ-5D index. It is used in pharmacoeconomics and health technology assessment to calculate quality-adjusted life years (QALY) [32]. The EQ-5D-5L Index value scale extends from '1', for perfect health, through to '0', which corresponds to the death state, and on to negative values, which indicate states even worse than death, according to the perceptions of a given society. For the assessment of EQ-5D-5L index, Polish directly measured, time trade-off (TTO) and discrete choice experiment (DCE)-based, EQ-5D-5L value set was used [33].
EQ VAS is a visual analog scale, where values from 0 to 100 appear on a 20 cm vertical axis, where 0 means 'the worst imaginable health state' and 100 means 'the best imaginable health state'. It constitutes a subjective measure of health.

Data collection
The data were collected by professional CBOS interviewers during face-to-face interviews (April to June 2014). The EQ-5D-5L questionnaire was distributed as a paper-and-pencil version. This distribution method has predominantly been used in HRQoL studies in Poland until the present time. All other data were collected using the computer-assisted personal interviewing (CAPI) system.

Data analysis
Results were presented for the whole sample, as well as for the predefined comparisons: (1) respondents with diabetes versus respondents without diabetes; (2) treated for diabetes versus untreated; (3) treated with insulin versus treated with other drugs versus treated with combined treatment. The mean values with standard deviation, median, interquartile range and range were estimated for the continuous variables, such as EQ VAS and EQ-5D-5L index. The distribution of answers to the questions in the descriptive part of the EQ-5D-5L was computed.

Statistical analysis
Confidence intervals for proportions were calculated using the Clopper-Pearson method. The parametricity of the distribution was explored with the Shapiro-Wilk test. The statistical significance of differences in dichotomous variables was examined using Fisher's exact test and in nominal variables by using a chi-square test. The Mann-Whitney test and ANOVA were used to assess differences between two and several demographic groups, respectively, in interval data, such as the EQ-5D index or EQ VAS. We used multiple linear regression to examine the associations of sociodemographic characteristics with the EQ-5D-5L index and EQ VAS scores, both in the population of diabetic patients and the whole population or respondents. All variables, including age, were entered into the models as categorical variables. We presented the regression coefficients, together with information about the level of statistical significance. The analysis was conducted using StatsDirect 3.1.22 statistical software (StatsDirect Ltd, Altrincham, England).

Studied population
The current analysis is based on data from 2,973 (99.6%) respondents (age range 18-87 years, 46.8% men, 36.7% inhabitants of rural areas), out of 2,986, for which complete answers to the EQ-5D-5L questionnaire were available (Table 1).
Respondents with diabetes, compared to respondents without diabetes, were older (average age difference-17.5 years) and were characterized by a lower level of education, lower employment rates (20.4% vs 51.5%), a higher percentage of pensioners (72.5% vs 28.1%), former smokers (29.0% vs 15.4%) and people reporting health limitations based on the EQ-5D-5L questionnaire (90.6% vs 58.0%).
Patients with treated diabetes, compared to untreated, were also older (mean age of 67.1 vs 57.0), more often retired, less likely to be working and with more frequently reported health problems according to the EQ-5D questionnaire (93.8% vs 80.6%).  Table 2 presents the level of problems in diabetes patients according to the EQ-5D-5L dimensions.

EQ-5D-5L dimensions
In general, patients with diabetes were characterized by a similar picture of the affected domains to that of the general population or to respondents without diabetes (dimensions in order from most to least affected being: PD, AD, MO, UA, SC). The identical pattern was typical for both untreated and treated diabetes, and it only changed in the subpopulation having insulin treatment, where the number of MO health limitations exceeded that in the AD dimension.
In terms of all the domains in the EQ-5D-5L questionnaire, diabetes respondents had a higher frequency of restrictions compared to both the general and non-diabetic populations. The most significant differences in the prevalence of any problems concerned MO, PD and AD-with 35.7%, 30.0% and 26.5% more restrictions, respectively, compared to the entire study population, and 39.0%, 32.8% and 29.0% more than the non-diabetic population.
Treated diabetes patients, as compared to non-treated, had a statistically significant higher incidence of restrictions within MO, PD and UA. At the type of therapy level, in terms of SC and UA, insulin-treated patients had the most problems, whereas in terms of MO, PD and AD it was those treated with a combination therapy.

EQ-5D-5L health states in patients with diabetes
In the 255 respondents with diabetes, 121 different EQ-5D-5L health states were identified, including 39 that occurred in at least two respondents and 8 that occurred in at least five ( Table 3). The most common health condition declared was 11111 -'without any limitations' (n = 24; 9.4%), followed by 11122 (n = 17; 6.7%).

PLOS ONE
Quality of life in diabetic patients based on EQ-5D-5L questionnaire

EQ VAS
Subjective health assessment (EQ VAS) was significantly lower in respondents with diabetes compared to non-diabetic population-a difference of 18.5 points (scale range 100 points; p <0.0001; Table 4). In diabetes patients, the subjective assessment of health was lower in treated respondents than non-treated-a difference of 8.6 points (p <0.01). A lower EQ VAS value was also observed in patients on insulin therapy versus those treated with other drugs-a difference of 8.9 points (p <0.05). The highest EQ VAS values were recorded in patients with diabetes belonging to the youngest age group of 18-49 years (69.4). They were significantly lower in the age groups of 50-64 years and above 65 years, with values of 58.7 and 52.8 respectively (p <0.001; Table 5).

EQ-5D-5L index
The results of the assessment of health, adjusted by the health preferences among Polish society (the Polish tariff-based EQ-5D-5L index) were consistent with the unweighted results and the subjective assessment. Respondents with diabetes, compared to non-diabetic ones, had a lower  Table 4). Higher EQ-5D-5L index values were characterized by patients with diabetes in younger age groups and with higher levels of education (Table 5, Fig 1).

Sociodemographic characteristics and HRQoL in patients with diabetes
The relationship between EQ-5D-5L index or EQ VAS and the sociodemographic characteristics of respondents with diabetes is summarized in Table 5, S1 and S2 Figs. The multivariate analysis showed that the factors independently reducing the quality of life of patients with diabetes (measured with EQ-5D-5L index) were being aged 65 years or above or residing in the provinces of Podlasie or Pomerania, while factor increasing EQ-5D-5L index -secondary or higher education. The subjective HRQoL assessment, measured with EQ VAS, was significantly lower when belonging to older age groups and higher when having greater levels of education. Figs 1 and 2 present 95% confidence intervals for EQ-5D-5L index, according to age group and education level respectively. Figs 3 and 4 present similar analyses for EQ VAS. Table 6 presents the relationship between the EQ-5D-5L index or EQ VAS and the sociodemographic characteristics of all respondents in the study. The multivariate analysis indicated that the factors independently improving the quality of life in the general population were secondary or higher education, and factors reducing HRQoL were female sex, belonging to an older age group, being treated because of diabetes with insulin, drugs other than insulin or combination treatment. Respondents diagnosed with diabetes but not treated with drugs showed a

PLOS ONE
Quality of life in diabetic patients based on EQ-5D-5L questionnaire decrease in EQ VAS scores, but not in the EQ-5D-5L index. S3 Fig. presents the comparison of limitations within EQ-5D-5L dimensions in respondents with or without diabetes, according to the age group.

PLOS ONE
Quality of life in diabetic patients based on EQ-5D-5L questionnaire

Discussion
We conducted the EQ-5D-5L questionnaire-based survey using a large representative sample of the general population of Poland, and developed quality of life norms for patients with selfreported diabetes. Although diabetes mellitus (DM) leads to a decrease in HRQoL across all age groups, patients with a basic level of education turned out to be a particularly vulnerable subpopulation. The developed normative data can be used in both clinical work and the health technology assessment (HTA) of new anti-diabetic drugs. This is the first study of HRQoL in patients with DM in Poland that is based on the EQ-5D-5L questionnaire.

PLOS ONE
Quality of life in diabetic patients based on EQ-5D-5L questionnaire One of the significant limitations of our study may be the moderate size of the subpopulation of patients who have declared the presence of diabetes. On the other hand, one should bear in mind that in order to identify this group, we approached nearly 3 000 representatives of the general population. The prevalence of self-reported diabetes (8.6%) and self-reported treated diabetes (6.5%) in our study was similar to that observed in the Polish-Norwegian Study (PONS; 8.4%; n = 3 854) [34], NATPOL PLUS in 2002 (6.4%; n = 3 051) [35] and NAT-POL 2011 study (6.7%; n = 2 411) [36], which confirms the appropriate selection of the population.

PLOS ONE
Quality of life in diabetic patients based on EQ-5D-5L questionnaire Another limitation of our study results from the method used to conduct it, specifically the limitations associated with a questionnaire survey. Though we ensured the proper recruitment of respondents with the use of stratified sampling, in the study itself the respondents selfdeclared their diagnosis of diabetes. We did not verify these diagnoses with fasting plasma glucose levels, blood HbA1c levels or by using data from medical records or National Health Fund registers. Nevertheless, this is the approach widely used in epidemiological research, and our results are comparable with numerous studies undertaken on other populations [37][38][39].
Several issues may be raised in terms of the survey used. In diabetes, diet is often the only therapy in the early stage of the disease. We asked about the diagnosis of diabetes and the usage of medications, but there was no 'diet' among treatment options. As we expected respondents' answers to be less reliable, we did not collect the data on the type of diabetes (type 1, type 2). Still, instead, we obtained the information on the insulin dependence of the condition. Some other data, like self-reported weight and height (allowing calculation of Body Mass Index), disease duration, or the presence of micro and macroangiopathy, could have added valuable information about the health status of the diabetes patients. These data could improve Table 6. Relation of EQ-5D-5L index and EQ VAS with demographic characteristics of the studied general population sample (n = 2973). the applicability of the diabetes population norms obtained in this study both as a reference point in clinical assessment and in modelling of the disease in economic evaluations. The strongest point of our study is clearly the method of sample selection, based on multistage stratified sampling using 65 strata and numbers from the PESEL database. This enabled us to obtain a representative sample of the Polish population in terms of multiple criteria. A significant number of HRQoL studies among patients with diabetes in Poland have already been published. These have mainly concerned type II diabetes [40][41][42][43][44][45][46][47][48], with studies on type I diabetes [49] or both types I and II being less common [50][51][52]. Some of the research focused on precisely defined subpopulations of diabetes patients, such as diabetic foot ulceration [53], neuropathic pain [54,55], peripheral diabetic neuropathy [56], maturity onset diabetes of the young (MODY) [57], transcatheter aortic valve implantation (TAVI) [58], gestational diabetes [59] or pre-diabetes [60]. The authors willingly use disease-specific questionnaires, including ADDQoL [40,46,47,53], Diabetes Quality of Life-Brief Clinical Inventory (DQL-BCI) [41][42][43], Diabetes Symptom Checklist-Revised (DSC-R) [41][42][43], Diabetic Foot Ulcer Scale short form [49] and the PedsQL Diabetes Module 3.0 questionnaire [45]. Concerning generic questionnaires, for Polish patients with diabetes the following were used: SF-36 [36,39,48,49,52,61], World Health Organization Quality of Life-Bref (WHOQOL-Bref) [38,55], and the EQ-5D-3L, which is undoubtedly the most commonly used [36, 41-44, 50, 51, 54].

Mean (SD) Multiple linear regression coefficients
Polish researchers present a considerable heterogeneity of approaches in seizing the opportunities offered by the EQ-5D framework. Some of them use only one of the available endpoints-EQ VAS [50,51] or limitations according to dimensions of the questionnaire [36]. Some researchers estimate two outcomes-VAS and HRQoL domains [54] or VAS and EQ-5D index [41][42][43]. The practice of using the full spectrum of possible results offered by the EQ-5D and calculating all three endpoints is rare [44]. This study is the first Polish survey of HRQoL in diabetes sufferers which employs the new five-level version of the EQ-5D questionnaire.
Both versions of the EQ-5D questionnaire (EQ-5D-3L and EQ-5D-5L) were validated in patients with diabetes [27][28][29][62][63][64]. The EQ-5D-5L seems to be characterized by having a lower ceiling effect, more discriminatory power, and a higher degree of preference among the respondents. Moreover, the conditions for the use of EQ-5D in Poland were developed by the publication of Polish population norms (by age and sex) for both EQ-5D-3L [65] and EQ-5D-5L [66,67], as well as the release of country-specific value sets reflecting the health preferences of Polish society, for both versions of the questionnaire [29,68].
In our study, patients with self-reported diabetes, in comparison to the general population, were marked by a higher prevalence of health limitations across all dimensions of the EQ-5D questionnaire. The most significant differences concerned the dimensions of mobility, pain/ discomfort and anxiety/depression. A similar hierarchy of affected dimensions was observed when comparing older Chinese patients with type 2 diabetes (T2D) with their age and gendermatched controls [69]. The subjective assessment of the health of Polish respondents with diabetes was significantly lower than in the general population-by 16.9 points on the EQ VAS scale. This difference was smaller than that obtained from data collected in Poland in 2008 (average 18.8 points) [44], but higher than that observed in the German population (12.5 points) [70]. In Poland, respondents with diabetes, compared to respondents from the general population, had an EQ-5D-5L index value that was 0.123 lower. This difference was higher than that observed in Japan (0.090) [71], China (0.072) [64] or Canada-in the provinces of Quebec and Alberta (0.084 and 0.040) [72,73]. The use of EQ-5D allows international comparisons to be readily performed.

Conclusions
The paper reports EQ-5D-5L normative data for Polish patients with self-reported diabetes, based on a national representative sample. These results may be used in outcome measurement in clinical care, as well as in economic analyses and health technology assessment reports for new anti-diabetic drugs.