1. Improved methods for
valuing EQ-5D-5L
Professor Nancy Devlin
ndevlin@ohe.org
Director of Research
Office of Health Economics, London
Spanish Health Economics Association
June 24, 2010
2. Content
• Why are new methods required?
– EQ-5D-5L
– Problems with conventional methods
– Opportunities presented by computer-aided methods
• LT-TTO
• Results from ongoing research
• DCE + VAS
• Preparing for national EQ-5D-5L value sets
Spanish Health Economics Association
June 24, 2010
3. Why develop new methods?
(i) EQ-5D-5L
• ‘Cross-over studies’ will provide interim values, but ultimately a
value set based on preferences directly elicited for EQ-5D-5L
states is required.
Challenges in valuing EQ-5D-5L: 5L
100%
3L 5L 3L 5L 3L 3L 5L 3L 5L
• More states to be valued (55 = 3125, compared with 35 = 243)
80%
• How best to elicit preferences for sufficient states across
% problem
60% descriptive system, to provide an adequate basis for
the
modelling?
40%
• Greater ‘subtlety’ between levels/labels: potential challenge
20% methods where states are considered ‘one by one’
for
• A ‘mixed methods’ approach may have more merit than reliance
0%
on TTO alone
Mobility
discomfort
Selfcare
activities
depression
Anxiety /
Usual
Pain /
Spanish Health Economics Association
June 24, 2010
4. Why develop new methods?
(ii) Addressing known problems
• The MVH protocol has come to be the de facto EQ-5D valuation
protocol.
• Yet is subject to known, non-trivial problems
• Substantial issues remain – principal among which is the
valuation of states worse than dead (< 0).
– The ‘standard’ TTO protocol cannot elicit values < 0
– A different method is required - meaning values > 0 and < 0
are non-comparable
– the method for values < 0 yields extreme values, requiring
post hoc transformation (eg. to -1)
Spanish Health Economics Association
June 24, 2010
5. Conventional TTO
Conventional TTO for state better than dead U(H) > 0
full health
Life A
Hi
Life B duration in Hi = 10 years
Conventional TTO for states worse than dead U (H) <0
Hi full health
Life A
Immediate death
Life B
Spanish Health Economics Association
June 24, 2010
6. Why develop new methods?
(iii) Exploiting new technologies
• Digital aids have largely replaced physical ‘props’ and
‘prompts’ to valuation tasks.
• Digital aids do not just replicate physical props, but offer
greater functionality, e.g.
– Built in randomisation procedures re: states and tasks
– Automated iterative procedures (greater consistency
between interviewers; less human error in prompting and
recording participant responses)
– Time stamping all responses
– Central data capture and storage; eliminating data entry
Spanish Health Economics Association
June 24, 2010
7. Lead Time TTO (LT-TTO)
‘Lead time’ TTO (state happens to be better than dead)
lead time full health
Life A
lead time Hi
Life B duration in Hi = 10 years
‘Lead time’ TTO (state happens to be worse than dead)
lead time
Life A
lead time Hi
Life B duration in Hi = 10 years
• Approach described by Robinson and Spencer (2006) Health Economics.
• LT-TTO shown to be feasible - Devlin et al (2010) Health Economics.
• A ‘lag time’ equivalent is also possible: order of states in Life B reversed – Tilling et al
(2010) Medical Decision Making.
Spanish Health Economics Association
June 24, 2010
8. Refining/testing the LT-TTO: current
research
• 1, 5, 10 year durations; lead: duration: 2:1, 5:1
• n = 208 participants, blocked into groups defined by pairs of variants/ordering
and one of two sets of states.
• Each participant valued 5 EQ-5D states using two variants i.e. 10 TTO tasks
• Data collection: May/June 2010.
[a] [b] [c] [d] Group Variant
Duration (years): 10 1 5 5 pairs
Lead time (years): 20 5 10 1 (a) + (b)
Lag (years) 10 2 (a) + (c)
Ratio* of 2:1 5:1 2:1 2:1 3 (b) + (c)
lead(lag):duration 4 (c) + (d)
Spanish Health Economics Association
June 24, 2010
9. Initial results (i):
values by ‘category’, by variant
By variant
Categories:
1. Full health
2. Better than dead 1. U = 1 (‘non-trading’)
3. Equivalent to dead
a 4. Valued using lead time
5. valuation by extension and/or reduction
2. U > 0 and < 1
6. Refuses any trade involving health state 3. U = 0
9. missing
1. Full health
2. Better than dead
4. U < 0
3. Equivalent to dead ( & valued within the
b 4. Valued using lead time
5. valuation by extension and/or reduction
6. Refuses any trade involving health state available lead/lag
9. missing
1. Full health time)
2. Better than dead 5. U < 0
3. Equivalent to dead
c 4. Valued using lead time
5. valuation by extension and/or reduction (& valued using
6. Refuses any trade involving health state
9. missing extended led/lag
1. Full health
2. Better than dead procedure)
3. Equivalent to dead
d 4. Valued using lead time
5. valuation by extension and/or reduction
6. ‘Refuses trade’
6. Refuses any trade involving health state 7. Missing
9. missing
0 100 200 300 400 500
Count Nb: x-axis shows
frequency, not %
Spanish Health Economics Association
June 24, 2010
10. Initial results (ii): distribution of values
(all variants); and exhausting lead time,
state 33333
80
Variant % who exhaust
60
lead time in
valuing 33333
Frequency
40 a 0.21
b 0.20
20 c 0.18
d 0.28
0
-6 -4 -2 0 2
valuation
Spanish Health Economics Association
June 24, 2010
11. Initial results (iii):
mean values by state by variant
Mean values, including ‘extended lead’ values, but excluding ‘missing’ values
LT-TTO variants
EQ-5D state a b c d
11112 0.77 0.57 0.77 0.81
11122 0.67 0.36 0.61 0.56
11211 0.87 0.63 0.80 0.80
12111 0.79 0.71 0.77 0.76
22121 0.52 0.47 0.63 0.68
23232 -0.41 -1.35 -0.54 -0.26
33333 -1.0 -3.92 -1.18 -1.10
Spanish Health Economics Association
June 24, 2010
12. Initial results (iv):
distribution of values > 0 by variant
a b
15
10
5
0
Density
c d
15
10
5
0
0 .5 1 0 .5 1
valuation
Graphs by Variant
Spanish Health Economics Association
June 24, 2010
13. Discrete Choice Modeling
• DCEs widely used in health services research in the UK
• The methods are grounded in theory eg. Thurstone (1927), through to
McFadden (1974, 1989).
• Based on the idea of choices reflecting trade-offs and underlying
preferences (NICE considers this important in its choice of valuations –
NICE 2008)
• Participants are asked to choose from pairs of scenarios (EQ-5D profiles):
simple to complete.
• Potential advantages in valuing EQ-5D-5L: feasible for self-completion; each
participant can complete many DCE valuation tasks.
• But problems ‘anchoring’ values at 0 and 1
• DCE to generate preference data to supplement TTO data
• Strength of preference data may assist: DCE accompanied by VAS
• Alternative ways of modelling these data will be explored.
Spanish Health Economics Association
June 24, 2010
14. Experimentation: DCE +VAS
Compare health states A and B. Imagine for each health state that you are in that state
yourself.
A B
Confined to bed No problems in walking about
Some problems washing or dressing myself
Unable to wash or dress myself
Some problems with performing my usual
activities Unable to perform my usual activities
No pain or discomfort Extreme pain or discomfort
Not anxious or depressed Extremely anxious or depressed
Which health state is best in your opinion, A or B?
Tick the box
A B
Spanish Health Economics Association
June 24, 2010
15. The best health
you can imagine
VAS Task 100
95
90
A 85
Confined to bed 80
75
Some problems washing or dressing myself
70
Some problems with performing my usual
65
activities
60
No pain or discomfort
55
Not anxious or depressed
50
45
B 40
No problems in walking about 35
30
Unable to wash or dress myself
25
Unable to perform my usual activities
20
Extreme pain or discomfort
15
Extremely anxious or depressed 10
5
Spanish Health Economics Association 0
June 24, 2010 The worst health
you can imagine
16. Preparing for EQ-5D-5L value sets
– Issues with methodology remain
– Some of which we hope to work out prior to EQ-5D-5L
value set studies
– 4-country study: 2010/2011
– EQ-5D-5L value sets: 2011/12
Spanish Health Economics Association
June 24, 2010