This document discusses two studies that examined how the valuation of EQ-5D-5L health states is affected by: 1) The ordering of the five dimensions in time trade-off (TTO) and discrete choice experiment (DCE) tasks. The studies found the dimension ordering had a small but statistically significant impact on TTO values. 2) Whether the TTO tasks used "full health" or the best possible health state "11111" as the comparator state. While most respondents considered the descriptions different and preferred "full health", there were no significant differences in mean TTO values between study arms that used different comparators.

1. Brendan Mulhern and Koonal Shah
1st Meeting of the International Academy of Health Preference Research
Amsterdam • 8 November 2014
Does the choice of health state comparator or ordering of dimensions matter when valuing EQ-5D-5L?

2. IAHPR presentation
08/11/14 2
•This study was funded by the EuroQol Group
•We are grateful for the contributions of Nancy Devlin, Bas Janssen, Louise Longworth and Ben van Hout
Acknowledgements

3. IAHPR presentation
08/11/14 3
• EuroQol protocol for valuing
EQ-5D-5L health states
involves a combination of
TTO and DCE tasks
• Issue 1: In both tasks, the
five dimensions are always
presented in the same order
– does the ordering matter?
• Issue 2: In TTO, the health
states are valued in
comparison to “full health” –
does the choice of
comparator matter?
Background:
Valuing EQ-5D-5L
By placing a tick in one box in each group below, please indicate which statements best
describe your own health state TODAY.
MOBILITY
I have no problems in walking about 
I have slight problems in walking about 
I have moderate problems in walking about 
I have severe problems in walking about 
I am unable to walk about 
SELF-CARE
I have no problems washing or dressing myself 
I have slight problems washing or dressing myself 
I have moderate problems washing or dressing myself 
I have severe problems washing or dressing myself 
I am unable to wash or dress myself 
USUAL ACTIVITIES (e.g. work, study, housework,
family or leisure activities)
I have no problems doing my usual activities 
I have slight problems doing my usual activities 
I have moderate problems doing my usual activities 
I have severe problems doing my usual activities 
I am unable to do my usual activities 
PAIN / DISCOMFORT
I have no pain or discomfort 
I have slight pain or discomfort 
I have moderate pain or discomfort 
I have severe pain or discomfort 
I have extreme pain or discomfort 
ANXIETY / DEPRESSION
I am not anxious or depressed 
I am slightly anxious or depressed 
I am moderately anxious or depressed 
I am severely anxious or depressed 
I am extremely anxious or depressed 
1
1
1
1
1
11111

4. IAHPR presentation
08/11/14 4
•To assess the impact of different dimension orderings on the valuation of EQ-5D-5L health states using TTO and DCE
•To assess differences between alternative comparator health states
•Compare the use of “full health” and 11111 (the “best” health state defined by the EQ-5D-5L descriptive system) in TTO tasks
•Examine respondents’ interpretations of these and other descriptors using a variety of methods
Aims

5. IAHPR presentation
08/11/14 5
•Composite TTO used (as per EQ-5D-5L protocol)
•‘Classic’ TTO for states better than dead (>= 0)
•‘Lead Time’ TTO for states worse than dead (<= 0)
•Values elicited range from -1 to 1
Methods – TTO
Life ALife BLife ALife BBetter than deadWorse than deadClassic TTOLEAD time TTOt=0t=10t=10t=20t=0

6. IAHPR presentation
08/11/14 6
•Pairwise comparisons of EQ-5D-5L health states
•No indifference option offered
Methods – DCE

7. IAHPR presentation
08/11/14 7
•3x2 design – random allocation to one of six arms
•Two TTO comparators:
1.Full health
2.11111
•Three dimension orders:
1.Standard (MO-SC-UA-PD-AD)
2.Reverse (AD-PD-UA-SC-MO)
3.Block (PD-AD-MO-SC-UA)
•10/11 TTO tasks; 7 DCE tasks
•Follow-up questions relating to the full health vs. 11111 issue
Methods – study design and survey administration
•CAPI; face-to-face
•450 general public respondents from South Yorkshire, UK

8. IAHPR presentation
08/11/14 8
•Dimension ordering – TTO
•Comparison of descriptive statistics
•One- and two-way ANOVA with post-hoc Bonferroni testing
•Linear regressions (5-parameter + order; states + order)
•Dimension ordering – DCE
•χ2-test (examining propensity to choose one state or the other)
•Conditional logit regressions
•Full health vs. 11111
•Comparison of descriptive statistics
•t-test (examining differences in mean values across arms)
•Analysis of self-reported health and follow-up question responses
Methods of analysis

9. IAHPR presentation
08/11/14 9
Results – TTO
Model 1: 5 para
Model 5: State
Coef.
Sig.
Coef.
Sig
MO
-0.002
0.87
SC
-0.105
0.00
UA
-0.061
0.00
PD
-0.073
0.00
AD
-0.059
0.00
State
11112
-0.036
0.20
11121
-0.019
0.51
11223
-0.166
0.00
21111
-0.025
0.37
21232
-0.175
0.00
32442
-0.603
0.00
34155
-0.833
0.00
43331
-0.403
0.00
55233
-0.706
0.00
55555
-1.171
0.00
Order
0.029
0.06
0.029
0.06
Gender
0.034
0.19
0.034
0.19
Age
-0.001
0.09
-0.001
0.09
N
442
442
Wald
5615
5765
State
Standard
MO-SC-UA-PD-AD (n=146)
Reverse
AD-PD-UA-SC-MO (n=149)
Block
PD-AD-MO-SC-UA (n=147)
M(sd)
Range
M(sd)
Range
M(sd)
Range
Overall
Overall
0.554 (0.60)
-1 to 1
0.610 (0.55)
-1 to 1
0.610 (0.54)
-1 to 1
0.01
11111
1.000 (0.00)
1 to 1
0.998 (0.01)
0.9 to 1
0.999 (0.01)
0.95 to 1
0.36
21111
0.949 (0.14)
0.05 to 1
0.961 (0.11)
0.2 to 1
0.968 (0.10)
0 to 1
0.38
11121
0.968 (0.09)
0.5 to 1
0.967 (0.10)
0.2 to 1
0.963 (0.10)
0.1 to 1
0.90
11112
0.941 (0.20)
-1 to 1
0.956 (0.12)
0 to 1
0.948 (0.13)
0 to 1
0.71
11223
0.757(0.43)
-1 to 1
0.848 (0.26)
-1 to 1
0.849 (0.19)
0.3 to 1
0.02
21232
0.754(0.38)
-1 to 1
0.838 (0.24)
-0.45 to 1
0.837 (0.20)
0 to 1
0.01
43331
0.516 (0.55)
-1 to 1
0.590 (0.52)
-1 to 1
0.637 (0.35)
-1 to 1
0.10
32442
0.323 (0.59)
-1 to 1
0.418 (0.54)
-1 to 1
0.404 (0.54)
-1 to 1
0.28
55233
0.238 (0.62)
-1 to 1
0.310 (0.61)
-1 to 1
0.287 (0.60)
-1 to 1
0.59
34155
0.129 (0.62)
-1 to 1
0.163 (0.63)
-1 to 1
0.160 (0.61)
-1 to 1
0.87
55555
-0.260 (0.5)
-1 to 0.95
-0.144 (0.54)
-1 to 1
-0.155 (0.58)
-1 to 1
0.15
Descriptive analysis
Regression analysis

10. IAHPR presentation
08/11/14 10
Results – DCE
Descriptive analysis
Regression analysis
MO-SC-UA-PD-AD
AD-PD-UA-SC-MO
PD-AD-MO-SC-UA
Coef.
Sig.
Order
Coef.
Sig.
Order
Coef.
Sig.
Order
MO
-0.198
0.05
4
-0.264
0.09
2
-0.282
0.05
4
SC
-0.281
0.04
3
-0.260
0.05
3
-0.410
0.00
2
UA
-0.313
0.00
2
-0.240
0.00
4
-0.227
0.00
5
PD
-0.142
0.28
5
-0.204
0.11
5
-0.394
0.03
3
AD
-0.367
0.00
1
-0.404
0.00
1
-0.525
0.00
1
n
2100
2170
2100
LL
-683
-710
-674
0%
20%
40%
60%
80%
100%
31113 (state 1) /11331 (state 2)
44222 (state 1) / 22244 (state 2)
% choosing state 1
Standard
Reverse
Block

11. IAHPR presentation
08/11/14 11
•No statistically significant differences were observed across the arms in terms of mean values (t-test; p=0.10) or other key measures of data quality (e.g. inconsistencies)
•Four respondents (1.8%) gave 11111 a value other than 1 (mean=0.999)
•However, of the 224 respondents who self-reported as being in 11111, only 37 (16.5%) rated themselves as 100 on a 0- 100 scale where 100 = “best imaginable health”
•Mean (median) rating for respondents self-reporting as being in 11111 was 89 (90)
Full health vs. 11111

12. IAHPR presentation
08/11/14 12
•Q: Do you consider these descriptions [full health and 11111] to be the same as each other?
•29.9% of respondents answered NO
•Q: Please rate the descriptions on a 0-100 scale where 0 = worst imaginable health and 100 = best imaginable health
•Mean rating for full health = 98.6; 86.0% gave it a rating of 100
•Mean rating for 11111 = 95.1; 58.2% gave it a rating of 100
•Q: Which of these descriptions [perfect health; full health; best imaginable health; no health problems; 11111; healthy] would you most want to live in?
•Perfect health most often ranked best or joint best
•344 respondents (79.1%) did not rank 11111 as best or joint best
Full health vs. 11111 – follow-up questions

13. IAHPR presentation
08/11/14 13
•Dimension ordering seems to affect TTO values (standard vs. non-standard)…
•…but the effects of order do not appear to be systematic or easily explainable
•Randomisation of dimension ordering between respondents seems like a no-brainer
•Are there other hypotheses that we should have explored? Should we have designed or analysed the study differently?
•Should past studies have used randomised ordering?
•Any other suggestions for exploring the differences between full health and 11111? Are there benefits for using one anchor or the other?
Some discussion points

14. IAHPR presentation
08/11/14 14
Thank you for listening kshah@ohe.org