Avai lable onl ine at www.sc iencedirect .com

journal homepage: www.elsevier .com/ locate /vhr i

The EQ-5D-5L is More Discriminative Than the EQ-5D-3L inPatients with Diabetes in SingaporePei Wang, PhD1,*, Nan Luo, PhD1, E.S. Tai, PhD2, Julian Thumboo, PhD3

1Saw Swee Hock School of Public Health, National University of Singapore, Singapore; 2Department of Medicine, Yong Loo Lin Schoolof Medicine, National University Health System, Singapore; 3Department of Rheumatology and Immunology, Singapore GeneralHospital, Singapore

A B S T R A C T

Objectives: To compare the discriminative power of the five-levelversion of the EuroQol five-dimensional questionnaire (EQ-5D-5L [5L])and its three-level version (EQ-5D-3L [3L]) in patients with diabetes inSingapore. Methods: A consecutive sample of patients with type 2diabetes mellitus (T2DM) self-completed the two versions of theEQ-5D in the clinic. The 3L index score was calculated from theSingapore 3L value set, whereas the 5L index score was mapped fromthe 5L index score using an interim scoring. The discriminative powerof the two EQ-5D indices was assessed in terms of their relativeefficiency (RE) in differentiating patients with T2DM with and withoutone of eight clinical conditions. The efficiency of the two EQ-5Dclassification systems was evaluated using the Shannon’s index (H')and in terms of ceiling effects. Results: A total of 121 patients withT2DM provided data for this study. The 3L score was systematicallyhigher than the 5L score for patients with T2DM with a condition and

systematically lower for the patients without a condition, with themean differences being 0.005 and �0.011, respectively. The 5L indexscore showed higher RE in seven of eight clinical conditions (mean RE1.87). The 5L classification system had higher H' in all dimensions:mobility (1.17 vs. 0.70), self-care (0.57 vs. 0.41), usual activities (1.01 vs.0.72), pain/discomfort (1.47 vs. 1.02), and anxiety/depression (1.36 vs.1.10). The overall ceiling effects decreased from 47.9% (3L) to 38.8%(5L). Conclusions: The EQ-5D-5L is more discriminative than theEQ-5D-3L in patients with T2DM in Singapore, supporting the use ofEQ-5D-5L in the population.Key words: EQ-5D-5L, EQ-5D-3L, discriminative power, diabetes,Singapore.

Copyright & 2016, International Society for Pharmacoeconomics andOutcomes Research (ISPOR). Published by Elsevier Inc.

Introduction

The three-level version of the EuroQol five-dimensional (EQ-5D) questionnaire (EQ-5D-3L [3L]), the original version of theEQ-5D, is a preference-based instrument that has been widelyused to describe and value health [1]. It contains a health-stateclassification system and a value set [2]. The classificationsystem defines health in five dimensions (mobility, self-care,usual activities, pain/discomfort, and anxiety/depression),each dimension with three levels (“no problems” [level 1],“some problems” [level 2], and “extreme problems” [level 3]),resulting in a total of 243 (35) health states. The system can beused as a self-completion questionnaire by a respondent toclassify his or her own health status. The value set ass-igns each health state defined by the system a singlepreference-based index score to indicate its utility to thegeneral public [2].

Although the 3L has shown validity and reliability in patientswith diabetes [3], it may be insensitive to health difference in

patients with minor problems and in those with multiple com-plications. It has also exhibited ceiling effects [4–6].

A five-level version of the EQ-5D (the EQ-5D-5L [5L]) has beenrecently developed by the EuroQol Group [7]. The 5L expands theEQ-5D classification system by adding two more responseoptions in each dimension. According to previous studiescomparing the two versions of the EQ-5D in various populations[8–15], the 5L classification system can improve the discrimi-native power of health difference and decrease the ceilingeffects. Most of the studies, however, did not compare thediscriminative power of the two EQ-5D index scores, except fora study by Pan et al. [15]. This study compared the index scoresin terms of their ability in distinguishing health severity levelsamong Chinese patients with diabetes, and indicated that theEQ-5D-5L index score is more discriminative. Given the possiblevariation in the health status of patients with diabetes, and thedifference in preference to the EQ-5D health states in differentcountries [16], a comparison of the two versions of the EQ-5D inpatients with diabetes in other countries is still needed.

2212-1099$36.00 – see front matter Copyright & 2016, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).

Published by Elsevier Inc.

http://dx.doi.org/10.1016/j.vhri.2015.11.003

Conflicts of interest: The authors have indicated that they have no conflicts of interest with regard to the content of this article.

E-mail: ephwp@nus.edu.sg.

* Address correspondence to: P. Wang, Saw Swee Hock School of Public Health, National University of Singapore, 12 Science Drive, BlockMD1, Singapore 117549.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 2

We therefore aimed to assess the ability of the two EQ-5Dindices in discriminating patients with type 2 diabetes mellitus(T2DM) with different severity levels in Singapore.

Methods

Study Participants

This study was approved by the institutional review board of theNational Healthcare Group of Singapore. In the study, consec-utive outpatients with T2DM were recruited from the diabetesclinic of a tertiary hospital in Singapore in 2012. The followingwere the inclusion criteria for the patients: 1) must have adiagnosis of T2DM, 2) must be Singapore citizens/Singaporepermanent residents aged 21years or older, and 3) must havethe ability to read and converse in English.

Study Design

After their routine consultations in the clinic, consenting patientswith T2DM were asked to self-complete a questionnaire includingthe two EQ-5D versions and sociodemographic and clinicalquestions. The clinical questions asked about the presence ofdiabetes complications (i.e., retinopathy, neuropathy, dermop-athy, diabetic heart disease, diabetic cerebrovascular disease,peripheral vascular disease, diabetic foot ulcer, and lowerextremity amputation) and comorbidities (e.g., lung disease,arthritis, depression/anxiety, cancer, and musculoskeletal dis-ease). The questions were framed as follows: “Have you ever beentold by a doctor that you had …?” The 5L was administered beforethe 3L because a previous study showed that study subjectstended to avoid the use of levels 2 and 4 of the 5L when the 3Lwas used first [13]. The patients’ other clinical information (i.e.,body mass index [BMI] and hemoglobin A1c levels) was retrievedfrom their medical records. The patients were told that the 3Land the 5L were independent approaches for assessing theirhealth. An investigator was available to assist the patients incompleting the questionnaire if necessary.

Instruments

The 3L classification system defined the 243 health states bycombining different levels from each dimension, with 11111 and33333 representing full health and worst health, respectively. The5L classification system comprised the same five dimensions, buteach dimension included five levels: “no problem” (level 1),“slight problems” (level 2), “moderate problems” (level 3), “severeproblems” (level 4), and “extreme problems” (level 5) [7]. Hence,the system defined 3125 (55) health states ranging from 11111 to55555. The descriptors of levels 1, 3, and 5 in the 5L were similarbut not identical to the wording of the 3L levels. For example, inthe mobility dimension, the description “I am confined to bed”(level 3) in the 3L was changed to “I am unable to walk about”(level 5) in the 5L.

The Singapore 3L value set was adopted to translate thepatients’ responses to the 3L index scores [17]; an indirect interimmapping method was used to predict the preference scores of the5L health states from the 3L scores [18]. The details of the methodwere reported elsewhere [18]. Briefly, the preference scores of the5L health state were computed by summing the preference scoresof the 3L health states weighted by the transition probability fromthe 3L state to the 5L state. The highest and the lowest scores forboth the EQ-5D indices were 1.00 and �0.769, respectively.

Data Analysis

Redistribution Properties from the 3L to the 5L

We evaluated redistribution properties from the 3L to the 5L,defined as proportions of the 3L-5L response pairs in each 3Lresponse level (i.e., 3L-1, 3L-2, and 3L-3) [13]. An inconsistentresponse pair was defined as a 3L response that was at least twolevels away from the 5L response (e.g., level 1 [no problem] in the3L, whereas level 3 [moderate problems] in the 5L); the other pairswere regarded as consistent. We calculated the proportion ofeach consistent pair in each 3L response level and the percentageof inconsistent pairs in each dimension.

The discriminative power of the two versions of the EQ-5Dwas assessed for their index scores and classification systems.

Relative Efficiency

The discriminative power of the two EQ-5D indices was assessedaccording to their index scores’ relative efficiency (RE) in differ-entiating patients with T2DM differing in health severities. REwas defined as the ratio of F statistics of analysis of variance testsof the difference in the index scores between the two comparisongroups of subjects known to differ in health. Because a higher F-statistic value is more likely to result in statistical significance,the index score with a higher F-statistic value would be consid-ered as more discriminative. In the study, we used the F-statisticvalue of the 3L index score to be the reference (RE ¼ 1) to calculatethe RE of the 5L index score. Because patients with a clinicalcondition tend to have worse health than do patients without thecondition, RE of the two EQ-5D index scores was assessed forpatients with and without each of eight clinical conditions. Theconditions included microvascular complications (i.e., retinop-athy, neuropathy, and dermopathy), macrocomplications (i.e.,ischemic heart disease, cerebrovascular disease, and peripheralvascular disease), any complications, two or more complications,three or more complications, any comorbidities, BMI levels of 25or more, and hemoglobin A1c (Hb A1c) levels of 6.5% or more.

Classification Efficiency

The classification efficiency of the EQ-5D dimension scales wasassessed using the Shannon index (H') and the evenness index (J')[19,20]. H' is defined as follows:

H0 ¼XL

i¼1

pilog2pi

where L is the number of descriptive levels of a dimension scaleand pi is the proportion of observations in the ith level (i ¼ 1, …, L).The 3L and the 5L have three and five descriptive levels,respectively. Larger H' values indicate higher classification effi-ciency. In the case of an even (rectangular) distribution (i.e., alllevels are equally filled), H' reaches its maximum that equals log2L, which amounts to 1.58 to the 3L (i.e., log2 3) and 2.32 to the 5L (i.e., log2 5). J' is calculated as J' ¼ H'/H'max, indicating the usage ofthe system (H') given its inherent capacity (H'max). The 95%confidence intervals for H' were estimated using a nonparametricbootstrap method with replacement strategy.

The classification efficiency was also evaluated in terms ofceiling effects, described as the proportion of “no problem”

responses in each dimension and in all dimensions. Reductionin ceiling effects suggests enhanced classification efficiency.We calculated the absolute reduction and relative reduction([Ceiling3L � Ceiling5L]/Celing3L) in ceiling effects from the 3Lto the 5L.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 258

Results

A total of 121 patients with T2DM provided data for this study.The mean age of the patients was 55.5 � 12.7 years, of whom57.0% were men. Many patients (59.5%) reported having one ormore complications, with the prevalence ranging from 6.6%(diabetic amputation) to 31.4% (retinopathy) (Table 1). Most ofthe patients did not have any other comorbidities (57.9%); manypatients had a BMI level of 25 (62.0%) or more and an Hb A1c levelof 6.5% (77.0%) or more.

There were 21 and 42 3L and 5L health states, respectively,observed in the study. The mean 3L and 5L index scores were0.786 � 0.299 and 0.795 � 0.278, respectively. For preferencescores between 0 and 0.7 (approximately), and at 1.0, morehealth-state observations were made in the 3L than in the 5L(Fig. 1). The mean 3L score was higher than the mean 5L score forpatients with a clinical condition and lower than the 5L score forpatients without a clinical condition. The mean differences were0.005 and �0.011, respectively (Table 2). The exception was forpatients with and without any comorbidities, for which the 3Lscore was lower than the 5L score (0.726 vs. 0.755) for patientswith the condition and higher than the 5L score (0.830 vs. 0.817)for those without the condition.

The distributions of the 3L and the 5L dimension responsesand the redistribution properties from the 3L to the 5L arepresented in Table 2. Most participants did not select the levelscorresponding for “severe” or “extreme” problems no matterwhich version of the EQ-5D was used. The percentage of the 5Llevels (i.e., levels 1, 3, and 5) decreased compared with that of the3L levels in all the dimensions, suggesting the benefit of theadded levels in the 5L. When going from the 3L to the 5L, theproportion in 5L-1 was always higher than the proportion in 5L-2in each 3L-1 subgroup (range 69.6% [pain/discomfort] to 96.3%[anxiety/depression]). Similarly, the most frequent response was5L-2 in all the 3L-2 subgroups. The 3L-3 subgroups had only a fewobservations and 5L-4 was mostly frequently observed. Theproportion of inconsistency was low in all the dimensions,ranging from 2.5% (anxiety/depression) to 4.1% (usual activities).

The mean RE of the 5L scores of the eight clinical conditionswas 1.87. Among these, the 5L scores were more efficient in sevenof the eight conditions (range 1.11 [macrocomplications] to 4.00[Hb A1c Z 6.5%]). It was less efficient in differentiating betweenpatients with and without any comorbidities (RE 0.81) (Table 3).

Compared with the 3L system, the 5L system had a substantialgain in classification efficiency in each dimension, indicated byhigher H' values: mobility (1.17/0.70), self-care (0.57/0.40), usualactivities (1.02/0.72), pain/discomfort (1.47/1.02), and anxiety/depression (1.36/1.10). However, the degree of the use of systempotential was comparable between the two systems, suggestedby similar J' values of the two systems (Table 4).

For both the EQ-5D classification systems, the self-caredimension showed the highest ceiling effects (3L: 91.7%; 5L:88.4%), whereas the pain/discomfort dimension exhibited thelowest (3L: 62.0%; 5L: 57.0%). Ceiling effects reduced in alldimensions when moving from the 3L to the 5L (Table 5). Themobility dimension showed the largest reduction—the absoluteand relative reductions were 7.4% and 9.0%. The proportion of “noproblem” responses in all the dimensions decreased from 47.9%to 38.8%, indicating an absolute reduction of 9.1% and a relativereduction of 19.0% of the overall ceiling effects (Table 6).

Discussion

We found that the EQ-5D-5L is more discriminative than the EQ-5D-3L in patients with T2DM in Singapore. Its index score wasmore efficient in discriminating patients with different severitylevels, and its classification system empirically described generichealth in greater detail and reduced ceiling effects. The findingswere in line with findings from the study based on a sample ofChinese patients with T2DM [15], though the degree of superiorityof the EQ-5D-5L in our study tended to be lower. The mean RE ofthe 5L index score was 1.87 versus 1.92 of the 3L index score. Theabsolute and relative reductions in ceiling effects were 9.1% and19.1% versus 20% and 35.3%. Meanwhile, our sample had lowerEQ-5D index scores: the mean 3L and 5L scores were 0.786 and0.795 in comparison with 0.898 and 0.876 from the study in China.This may be due to two reasons. First, the Singapore EQ-5D-3Lvalue set generated lower preference scores [17]. Second, oursample had worse health-related quality of life, suggested bylower percentages of “no problem” responses in all the EQ-5Ddimensions.

Consistent with previous studies, the redistribution datashowed skewed distributions in all dimensions [8–13]. The 3L-1and 5L-1 response pair always showed the largest proportion,because many patients with T2DM displayed no symptoms on aparticular dimension no matter how refined the response scalewas. The proportions of the 3L-2 and 5L-2 and the 3L-3 and 5L-4response pairs were also large, supporting the inclusion of asecond level and a fourth level for the 5L. It also implies that

Table 1 – Characteristics of patients with T2DM.

Characteristic Patients withT2DM (N ¼ 121), n (%)

SexMale 69 (57.0)Female 52 (43.0)

Age group (y)21–60 40 (33.1)460 81 (66.9)

Education levelCollege and higher 24 (19.8)Middle school and lower 97 (80.2)

RaceChinese 60 (49.6)Non-Chinese 61 (50.4)

Complications*Retinopathy 38 (31.4)Neuropathy 33 (27.3)Dermopathy 17 (14.1)Diabetic heart disease 28 (23.1)Diabetic cerebrovascular disease 17 (14.1)Diabetic peripheral vascular

disease21 (17.4)

Diabetic foot ulcer 14 (11.6)Diabetic amputation 8 (6.6)

ComorbiditiesYes 51 (42.1)No 70 (57.9)

BMI levelo25 46 (38.0)Z25 75 (62.0)

Hb A1c levelo6.5% 28 (23.0)Z6.5% 93 (77.0)

BMI, body mass index; Hb A1c, hemoglobin A1c; T2DM, type 2diabetes mellitus.*Because the patients may have had multiple complications, thesum of percentages was larger than 100%.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 2 59

Fig. 1 – Cumulative percentage of the EQ-5D-5L and EQ-5D-3L index scores.

Table 2 – Response distributions of the EQ-5D-3L and the EQ-5D-5L, and redistribution properties of consistentand inconsistent responses from the EQ-5D-3L to the EQ-5D-5L..

Dimension 3L n (%) bydimension

5L n (%) bydimension

3L 5L n (%) by 3Llevel

Inconsistencies,*n (%)

Mobility 1 100 (82.6) 1 91 (75.2) 1 1 90 (91.8) 4 (3.3)2 20 (16.5) 2 20 (16.5) 2 8 (8.2)3 1 (0.8) 3 6 (5.0) 2 2 12 (66.7)

4 2 (1.7) 3 4 (22.2)5 2 (1.7) 4 2 (11.1)

3 5 1 (100)Self-care 1 111 (91.7) 1 107 (88.4) 1 1 104 (93.7) 3 (2.5)

2 10 (8.3) 2 13 (10.7) 2 7 (6.3)3 0 3 0 2 2 6 (85.7)

4 1 (0.8) 4 1 (14.3)5 0

Usual activities 1 100 (82.6) 1 93 (76.9) 1 1 90 (90.0) 5 (4.1)2 20 (16.5) 2 21 (17.4) 2 10 (10.0)3 1 (0.8) 3 4 (3.3) 2 2 11 (73.3)

4 0 3 4 (26.7)5 3 (2.5) 3 5 1 (100)

Pain/discomfort 1 75 (62.0) 1 69 (57.0) 1 1 67 (69.6) 3 (2.5)2 45 (37.2) 2 38 (31.4) 2 7 (30.4)3 1 (0.8) 3 9 (7.4) 2 2 31 (72.1)

4 5 (4.1) 3 9 (20.9)5 0 4 3 (7.0)

3 4 1 (100)Anxiety/

depression1 80 (66.1) 1 80 (66.1) 1 1 77 (96.3) 3 (2.5)2 37 (30.6) 2 28 (23.1) 2 3 (3.8)3 4 (3.3) 3 6 (5.0) 2 2 25 (66.7)

4 7 (5.8) 3 6 (31.1)5 0 4 3 (2.2)

3 4 4 (100)

EQ-5D, EuroQol five-dimensional questionnaire; EQ-5D-3L (3L), three-level version of the EQ-5D; EQ-5D-5L (5L), five-level version of the EQ-5D.* An inconsistent response pair was defined as a 3L response that was at least two levels away from the 5L response (e.g., level 1 in the 3L andlevel 3 in the 5L); the other pairs were regarded as consistent.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 260

“some” or “extreme” problems in the 3L are better described by“slight” or “severe” problems in the 5L, respectively. The lowproportion of inconsistent responses in all the dimensions meant

that the participants could consistently use the two EQ-5Dsystems.

The finding that the 5L index score was more discriminative isexpected because the score is related to its classification system,which defines health in much greater detail than does the 3Lsystem. The finding is also similar to a previous study thatcompared the RE of the crude summary scores of the twoversions of the EQ-5D in patients with cancer with differentseverities and showed that the 5L summary score was moreefficient [10]. Moreover, our study systematically assessed theproperty through comparing the two EQ-5D indices in patientswith and without various clinical conditions. Hence, the studyprovides an important evidence for using the 5L in patients withdiabetes. The much higher RE (on average 1.87) suggested thatadopting the 5L to capture a known health difference wouldsignificantly reduce the sample size requirement compared withusing the 3L.

In general, the 3L index score was higher than the 5L indexscore if a clinical condition was present and it was lower than the5L index score if a condition was absent. The reason for theformer could be the reduction in ceiling effects from the 3L to the5L; for the latter, it could be the fact that the patients perceivedtheir health as less severe than the 3L described, because theobservations of the 5L score between 0 to approximately 0.7 wereless than those of the 3L score (Fig. 1). This systematic differenceimplies that the two EQ-5D indices would lead to different res-ults in quantifying between-group differences or within-group

Table 3 – Mean EQ-5D-3L and EQ-5D-5L index scores of patients with T2DM with and without a clinicalcondition.

Conditions (N) Patients with the condition Patients without the condition

Mean 3Lscore � SD

Mean 5Lscore � SD

Difference* Mean 3Lscore � SD

Mean 5Lscore � SD

Difference*

Microcomplications (56) 0.747 � 0.29 0.736 � 0.28 �0.011 0.822 � 0.31 0.846 � 0.26 0.024Macrocomplications (37) 0.704 � 0.35 0.692 � 0.33 �0.012 0.828 � 0.26 0.836 � 0.24 0.008Any complications (72) 0.710 � 0.34 0.705 � 0.31 �0.005 0.894 � 0.19 0.903 � 0.17 0.009Two or more

complications (42)0.686 � 0.33 0.692 � 0.33 �0.006 0.839 � 0.27 0.851 � 0.22 0.012

Three or morecomplications (27)

0.678 � 0.31 0.677 � 0.30 �0.001 0.817 � 0.29 0.829 � 0.26 0.012

Any comorbidities (47) 0.726 � 0.33 0.755 � 0.29 0.029 0.830 � 0.28 0.817 � 0.26 �0.013BMI Z 25 (75) 0.778 � 0.30 0.765 � 0.31 �0.013 0.790 � 0.31 0.816 � 0.23 0.026Hb A1c Z 6.5% (94) 0.761 � 0.33 0.743 � 0.31 �0.018 0.788 � 0.30 0.794 � 0.27 0.006Mean 0.724 0.720 �0.005 0.826 0.837 0.011

BMI, body mass index; EQ-5D, EuroQol five-dimensional questionnaire; EQ-5D-3L (3L), three-level version of the EQ-5D; EQ-5D-5L (5L), five-levelversion of the EQ-5D; Hb A1c hemoglobin A1c; T2DM, type 2 diabetes mellitus.*Reference group is mean 5L score.

Table 4 – RE of the EQ-5D-3L and the EQ-5D-5L indexscores in discriminating between patients withT2DM with and without a clinical condition..

Condition (N) RE*

Microcomplications (56) 2.49Macrocomplications (37) 1.11Any complications (72) 1.92Two or more complications (42) 1.26Three or more complications (27) 1.34Any comorbidities (47) 0.73BMI Z 25 (75) 2.13Hb A1c Z 6.5% (94) 4.00

Mean 1.87

BMI, body mass index; EQ-5D, EuroQol five-dimensional question-naire; EQ-5D-3L (3L), three-level version of the EQ-5D; EQ-5D-5L(5L), five-level version of the EQ-5D; RE, relative efficiency; Hb A1c

hemoglobin A1c; T2DM, type 2 diabetes mellitus.* Reference is 3L index score; an RE larger than 1 indicates that theEQ-5D-5L score is more discriminative.

Table 5 – Classification efficiency of the EQ-5D-3L and the EQ-5D-5L classification systems measured byShannon index (H').

Dimension 3L 5L

H'* (95% CI) J' (95% CI) H' (95% CI) J' (95% CI)

Mobility 0.70 (0.50– 0.84) 0.44 (0.32– 0.53) 1.17 (0.93– 1.32) 0.50 (0.40– 0.57)Self-care 0.41 (0.25– 0.52) 0.26 (0.16– 0.33) 0.57 (0.39– 0.70) 0.25 (0.17– 0.30)Usual activities 0.72 (0.52– 0.85) 0.46 (0.33– 0.54) 1.02 (0.71– 1.19) 0.44 (0.31– 0.51)Pain/discomfort 1.02 (0.91– 1.08) 0.65 (0.58– 0.68) 1.47 (1.31– 1.61) 0.63 (0.56– 0.69)Anxiety 1.10 (0.96– 1.21) 0.70 (0.61– 0.77) 1.36 (1.18– 1.51) 0.59 (0.51– 0.65)

CI, confidence interval; EQ-5D, EuroQol five-dimensional questionnaire; EQ-5D-3L (3L), three-level version of the EQ-5D; EQ-5D-5L (5L), five-level version of the EQ-5D.* Larger H' value indicates higher classification efficiency.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 2 61

changes. The utility gain from a transition from with a conditionto without the condition would be greater using the 5L indexscore. For example, according to our results, the gain from atransition from a BMI level of 25 or more to a BMI level of lessthan 25 would be 0.104 and 0.151 using the 3L and the 5L scores,respectively. This means a health intervention or programachieving such an improvement in health would be consideredmore efficient (e.g., more quality-adjusted life-years gained) usingthe 5L score than using the 3L score. Nevertheless, future studiesshould further examine this when the Singapore 5L value setderived from a valuation study is available.

The improved classification efficiency of the 5L was consistentwith previous studies and was anticipated because the 5Lclassification system better described the health variations.However, the degree of improvement was varied. For example,the absolute and relative reductions in overall ceiling effectsranged from 1.1% to 12.6% and from 7.1% to 49.0%, respectively[8–13]. It is not surprising because different populations havedifferent health-related quality of life.

The study had several limitations. First, the 5L index scoreswere mapped from the 3L index scores because the Singapore 5Lvalue set estimated from a valuation study had not yet beendeveloped. Hence, the results may have been different if such avalue set had been available. Second, only outpatients with T2DMjoined the study. The findings may not be generalizable toinpatients with T2DM who may have more severe health prob-lems. Third, the 5L was administered first, which might haveintroduced an order effect. Fourth, there may be a recall biasbecause the information of complications and comorbidities wasrecalled by patients themselves.

In conclusion, the EQ-5D-5L is more efficient than the EQ-5D-3L in detecting health variations. Findings from the study supportthe use of the EQ-5D-5L in patients with diabetes in Singapore.

Source of financial support: This study was supported by theNational Healthcare Group of Singapore.

R E F E R E N C E S

[1] Rabin R, de Charro F. EQ-5D: a measure of health status from theEuroQol Group. Ann Med 2001;33:337–43.

[2] Dolan P. Modeling valuations for EuroQol health states. Med Care1997;35:1095–108.

[3] Janssen MF, Lubetkin EI, Sekhobo JP, et al. The use of the EQ-5Dpreference-based health status measure in adults with type 2 diabetesmellitus. Diabet Med 2011;28:395–413.

[4] Bech P, Moses R, Gomis R. The effect of prandial glucose regulation withrepaglinide on treatment satisfaction, wellbeing and health status inpatients with pharmacotherapy naïve type 2 diabetes: a placebo-controlled, multicentre study. Qual Life Res 2008;12:413–25.

[5] Bagust A, Beale S. Modelling EuroQol health-related utility values fordiabetic complications from CODE-2 data. Health Econ 2005;14:217–30.

[6] Morgan CL, McEwan P, Morrissey M, et al. Characterization andcomparison of health-related utility in people with diabetes withvarious single and multiple vascular complications. Diabet Med2006;23:1100–5.

[7] EuroQol Group. EQ-5D-5L user guide: basic information on how to usethe EQ-5D-5L instrument. 2014. Available from: http://www.euroqol.org/fileadmin/user_upload/Documenten/PDF/Folders_Flyers/UserGuide_EQ-5D-5L.pdf. [Accessed November 10, 2014].

[8] Jia YX, Cui FQ, Li L, et al. Comparison between the EQ-5D-5L and theEQ-5D-3L in patients with hepatitis B. Qual Life Res 2014;23:2355–63.

[9] Kim SH, Kim HJ, Lee SI, et al. Comparing the psychometric properties ofthe EQ-5D-3L and EQ-5D-5L in cancer patients in Korea. Qual Life Res2012;21:1065–73.

[10] Pickard AS, De leon MC, Kohlmann T, et al. Psychometric comparisonof the standard EQ-5D to a 5 level version in cancer patients. Med Care2007;45:259–63.

[11] Scalone L, Ciampichini R, Fagiuoli S, et al. Comparing the performanceof the standard EQ-5D 3L with the new version EQ-5D 5L in patientswith chronic hepatic disease. Qual Life Res 2013;22:1707–16.

[12] Janssen MF, Pickard AS, Golicki D, et al. Measurement properties of theEQ-5D-5L compared to the EQ-5D-3L across eight patient groups: amulti-country study. Qual Life Res 2013;22:1717–27.

[13] Janssen MF, Birnie E, Haagsma JA, et al. Comparing the standard EQ-5Dthree-level system with a five-level version. Value Health 2008;11:275–84.

[14] Craig BM, Pickard AS, Lubetkin EI. Health problems are more common,but less severe when measured using newer EQ-5D versions. J ClinEpidemiol 2014;67:93–9.

[15] Pan CW, Sun HP, Wang X, et al. The EQ-5D-5L index score is morediscriminative than the EQ-5D-3L index score in diabetes patients. QualLife Res 2015;24:1767–74.

[16] Wang P, Li MH, Liu GG, et al. Do Chinese have similar health-statepreferences? A comparison of mainland Chinese and SingaporeanChinese. Eur J Health Econ 2015;16:857–63.

[17] Luo N, Wang P, Thumboo J, et al. Valuation of EQ-5D-3L health states inSingapore: modeling of time trade-off values for 80 empiricallyobserved health states. Pharmacoeconomics 2014;30:495–507.

[18] van Hout B, Janssen MF, Feng YS, et al. Interim scoring for the EQ-5D-5L: mapping the EQ-5D-5L to EQ-5D-3L value sets. Value Health2012;15:708–15.

[19] Shannon CE, Weaver W. A mathematical theory of communication.Bell Syst Tech J 1948;27:161–3.

[20] Janssen MF, Birnie E, Bonsel GJ. Evaluating the discriminatory power ofEQ-5D, HUI2 and HUI3 in a US general population survey usingShannon’s indices. Qual Life Res 2007;16:895–904.

Table 6 – Proportion of “no problem” responses on the EQ-5D-3L and the EQ-5D-5L.

Dimension 3L, n (%) 5L, n (%) Absolute reduction (%) Relative reduction* (%)

Mobility 100 (82.6) 91 (75.2) 7.4 9.0Self-care 111 (91.7) 107 (88.4) 3.3 3.6Usual activities 100 (82.6) 93 (76.9) 5.7 6.9Pain/discomfort 75 (62.0) 69 (57.0) 5.0 8.1Anxiety 80 (66.1) 78 (64.5) 1.6 2.4Full health (11111) 58 (47.9) 47 (38.8) 9.1 19.0

EQ-5D, EuroQol five-dimensional questionnaire; EQ-5D-3L (3L), three-level version of the EQ-5D; EQ-5D-5L (5L), five-level version of the EQ-5D.* Relative reduction was defined as ([Ceiling3L � Ceiling5L]/Celing3L) in ceiling effects from the EQ-5D-3L to the EQ-5D-5L.

V A L U E I N H E A L T H R E G I O N A L I S S U E S 9 C ( 2 0 1 6 ) 5 7 – 6 262