❤️Zirakpur Escorts☎️7837612180☎️ Call Girl service in Zirakpur☎️ Zirakpur Cal...
An EQ-5D-5L Value Set for England- Nancy Devlin and Ben Van Hout
1. An EQ-5D-5L value set for England
Nancy Devlin & Ben van Hout
on behalf of the OHE & ScHARR research team
OHE seminar
London • October 30th 2014
2. This report is independent research commissioned and funded
by the NIHR/ Department of Health Policy Research
Programme (‘EQ-5D-5L Value Set for England’ - 070/0073).
The views expressed in this publication are those of the
author(s) and not necessarily those of the Department of
Health.
Additional data have been used from the PRET study, funded
by the MRC-NIHR Research Methodology Programme, and the
PRET-AS study, funded by the EuroQol Group.
Views expressed in the paper are those of the authors, and
not the funding bodies.
An EQ-5D-5L value set for England
October 30th 2014
Disclaimer
3. An EQ-5D-5L value set for England
October 30th 2014
Content
1. Background
2. Aims
3. Study design
4. Data
5. Modelling
6. Key results and implications
7. Remaining research questions
4. An EQ-5D-5L value set for England
October 30th 2014
Background
• EQ-5D-5L: 3,125 states cf. 243 in the EQ-5D
• Requests for 5L use now supersedes 3L: an important
instrument
• Interim utilities available by mapping 5L states to 3L
states, and using existing 3L value sets
• But ultimately, ‘bespoke’ value sets required for 5L
states
• Values are generally required to be based on
preferences of the general public (e.g. NICE 2013)
• Using ‘stated preferences’ research methods
5. 3L 5L Would we
expect
underlying
values to be the
same?
11111 11111 Yes
33333 55555 Not necessarily –
mobility ‘extreme’
vs. ‘confined to
bed’
22222 33333 Not necessarily –
‘some’ vs.
‘moderate’
An EQ-5D-5L value set for England
October 30th 2014
6. An EQ-5D-5L value set for England
October 30th 2014
Aims
Aim: To produce a set of values for the EQ-5D-5L health state
descriptive system, based on the preferences of the general public
in England, for use in decisions based on EQ-5D-5L data
We investigated the following questions:
• What is the best method to generate an EQ-5D-5L Value Set
which reflects the stated preferences of the English general
public?
• How can conceptually different types of preference data – Time
Trade Off (TTO) and Discrete Choice Experiment (DCE) – be
combined in modelling health state values?
• How are extreme negative opinions about health states best
handled?
• How do people differ in their stated preferences for quality of life;
and life and death?
7. An EQ-5D-5L value set for England
October 30th 2014
Study design
• Research protocol developed by the EuroQol Group: EQ-VT
• Stated preference data collected in face-to-face computer-assisted
personal interviews
• n = 1000 members of the adult general public of England,
selected at random from residential postcodes
• Sample recruitment sub-contracted to Ipsos MORI
• Each participant valued 10 health states using TTO, randomly
assigned from 86 health states in an underlying design; and
seven DCE tasks, randomly assigned from 196 pairs of states
• ‘Composite’ TTO approach: conventional TTO for values > 0 and
‘lead time’ TTO for values < 0
• The EuroQol Group’s Valuation Technology software (EQ-VT) was
used to present the tasks and to capture participants’ responses
9. U(Hi) = (x/t)
where t = 10
years
An EQ-5D-5L value set for England
October 30th 2014
TTO for values > 0 (states better
than dead)
Example
shown:
U(Hi) = 0.5
10. Where t = 10 years, and
lead time = 10 years,
U(Hi) = (x-10)/20-10)
and min value = -1
Example shown:
U(Hi) = -0.5
An EQ-5D-5L value set for England
October 30th 2014
TTO for values < 0 (states worse
than dead)
12. An EQ-5D-5L value set for England
October 30th 2014
Data
• Interviews conducted between Nov 2012 and May 2013
• 996 completed the valuation questionnaire (response rate
approx. 40%)
• Close attention paid to data quality: daily monitoring of
uploaded data and follow-up with interviewers
• Sample broadly representative of English adult general
public, although a somewhat larger proportion of retired
individuals and a smaller proportion of younger individuals
• In modelling, we also incorporated DCE data from a
methodological study (‘PRET’)
13. Proportions choosing A and B based on relative
An EQ-5D-5L value set for England
October 30th 2014
DCE data
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
severities of A and B
-10 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9
Misery index of state A minus misery index of state B
% B
% A
0 1
-10 -5 0 5 10
delta sum of scores
dif in misery
14. An EQ-5D-5L value set for England
October 30th 2014
TTO data
• Fewer values < 0 (worse than dead) compared to Dolan
(1997) value set – as expected.
• Clusters of values at -1, 0, 0.5 and 1
• Logical inconsistencies (e.g. 55555 > than other states)
• ‘Unusual’ valuations e.g. mild states being valued < 0
• Interviewer effects apparent
15. 1 3 4 5 6 7 8
An EQ-5D-5L value set for England
October 30th 2014
-0.10 -0.05 0.00
misery coefficient
group
TTO groups
group N
all at 1 18
0<all<1 217
-1<all<1 113
-1=<all<1 53
0<all<=1 194
-1<all<=1 55
-1=<all<=1 52
16. An EQ-5D-5L value set for England
October 30th 2014
11121 mean= 0.876
value
density
-1.0 0.0 0.5 1.0
0 20 40
12111 mean= 0.868
value
density
-1.0 0.0 0.5 1.0
0 20 40
11211 mean= 0.866
value
density
-1.0 0.0 0.5 1.0
0 20 40
11221 mean= 0.862
value
density
-1.0 0.0 0.5 1.0
0 10 20
21111 mean= 0.83
value
density
-1.0 0.0 0.5 1.0
0 20 40
12121 mean= 0.823
value
density
-1.0 0.0 0.5 1.0
0 5 15
11112 mean= 0.815
value
density
-1.0 0.0 0.5 1.0
0 20 40
11122 mean= 0.806
value
density
-1.0 0.0 0.5 1.0
0 5 15
11212 mean= 0.801
value
density
-1.0 0.0 0.5 1.0
0 5 15
Distributions, by state
17. An EQ-5D-5L value set for England
October 30th 2014
54231 mean= 0.473
value
density
-1.0 0.0 0.5 1.0
0 4 8 12
33253 mean= 0.465
value
density
-1.0 0.0 0.5 1.0
0 5 15
12334 mean= 0.463
value
density
-1.0 0.0 0.5 1.0
0 4 8
23514 mean= 0.46
value
density
-1.0 0.0 0.5 1.0
0 4 8 12
43514 mean= 0.443
value
density
-1.0 0.0 0.5 1.0
0 4 8
15151 mean= 0.436
value
density
-1.0 0.0 0.5 1.0
0 5 10 15
23152 mean= 0.435
value
density
-1.0 0.0 0.5 1.0
0 5 15
31525 mean= 0.428
value
density
-1.0 0.0 0.5 1.0
0 5 10
31524 mean= 0.423
value
density
-1.0 0.0 0.5 1.0
0 2 4 6 8
Distributions, by state
18. An EQ-5D-5L value set for England
October 30th 2014
Distributions, by state
21444 mean= 0.148
value
density
-1.0 -0.5 0.0 0.5 1.0
0 10 20
53244 mean= 0.148
value
density
-1.0 -0.5 0.0 0.5 1.0
0 5 15
52455 mean= 0.12
value
density
-1.0 -0.5 0.0 0.5 1.0
0 5 10
43555 mean= 0.119
value
density
-1.0 -0.5 0.0 0.5 1.0
0 5 10 15
55555 mean= 0.016
value
density
-1.0 -0.5 0.0 0.5 1.0
0 100
NA mean= NA
value
density
-1.0 -0.5 0.0 0.5 1.0
-1.0 0.0 1.0
19. An EQ-5D-5L value set for England
October 30th 2014
Descriptive statistics
minimum value
minv
Frequency
-1.0 -0.5 0.0 0.5 1.0
0 50 100 150
maximum value
maxv
Frequency
0.0 0.2 0.4 0.6 0.8 1.0
0 200 400
standard deviation of values
(varv^0.5)
Frequency
0.0 0.2 0.4 0.6 0.8 1.0
0 50 100 150
range of values used
range
Frequency
0.0 0.5 1.0 1.5 2.0
0 50 100 150
20. • Overall, English 5L valuation data have acceptable ‘face validity’:
the worse the health state, the lower the mean and median value
An EQ-5D-5L value set for England
October 30th 2014
Face validity of the data
21. An EQ-5D-5L value set for England
October 30th 2014
Interpretation of the data
• Evidence from this study suggests that it is harder
to imagine, differentiate between, and value health
states described in terms of 5L rather than 3L
• More subtle differences between states
• Cognitive burden
• Initial model results suggested respondents did not
differentiate between ‘severe’ (level 4) and
‘extreme’ (level 5) problems on the dimension
anxiety/depression
22. Interpretation of the data
• Our process for examining the individual-level
An EQ-5D-5L value set for England
October 30th 2014
data:
• Let’s look at all our respondents
• Put expected value according to DCE on x axis
• Put values on Y axis
• And stare at 1,000 graphs
23. An EQ-5D-5L value set for England
October 30th 2014
Examination of individual-level data
24. An EQ-5D-5L value set for England
October 30th 2014
Examination of individual-level data
25. An EQ-5D-5L value set for England
October 30th 2014
Real or censored???
27. An EQ-5D-5L value set for England
October 30th 2014
Decisions regarding the data
• Excluded 23 respondents who gave all 10 health states
the same value; and 61 respondents who valued 55555
(misery score = 25) no lower than the value they gave to
the mildest health state included in their block (misery
score = 6)
• The core modelling dataset includes 912 respondents,
with 10 TTO observations for each
• Censored 105 individuals/477 zeros with >2 states at
zero (that is out of 1,315 zeros)
• Censored 68 individuals/142 data points with inconsistent
negative data
28. An EQ-5D-5L value set for England
October 30th 2014
Modelling
• The main specifications included models with 5, 9, 10 and
20 parameters (four parameters for each of the five
dimensions reflecting a utility decrement for each severity
level)
• All models were estimated for both TTO and DCE data, and
‘hybrids’ of the these
• Heterogeneity explored via random coefficient models,
which estimate value functions for every individual
member of the sample
• Values at -1 treated as censored
29. An EQ-5D-5L value set for England
October 30th 2014
The hybrid likelihood
General
• You have a statistical model
that generates the data,
holding unknown
parameters
• You have the data
• You calculate for every set
of parameters the
probability that the data
occur
• The likelihood is the product
of all probabilities
• You calculate the
parameters at which this
product of probabilities
(likelihood) is highest
Specific
• There is a likelihood for the
DCE-data
– Assuming normal errors
• There is a likelihood for the
TTO-data
– Assuming normal errors
• The combined likelihood is
the product of both
likelihoods
33. An EQ-5D-5L value set for England
October 30th 2014
Issues with the TTO results
• Lower parameter for anxiety-depression level
5 than for anxiety/depression level 4
• Brute force
• Heterogeneity
• Latent distributions
• Low value of the intercept
• Error distributions
34. Heterogeneity
• The coefficients beta which reflect weights for
dimensions and levels are normally distributed
over the population
• The shape of the value as a function of x’beta
An EQ-5D-5L value set for England
October 30th 2014
follows a:
– Normal distribution
– Lognomal distribution
– Multinomial distribution
– (3 latent classes)
1
0.5
0
-0.5
-1
-1.5
value
x'beta
35. An EQ-5D-5L value set for England
October 30th 2014
Parameter estimates
0.300
0.250
0.200
0.150
0.100
0.050
0.000
0.350
0.300
0.250
0.200
0.150
0.100
0.050
0.000
-0.050
slope ~ normal
slight moderate severe unable/extreme
0.600
0.500
0.400
0.300
0.200
0.100
0.000
-0.100
slope ~ lognormal
slight moderate severe unable/extreme
3.500
3.000
2.500
2.000
1.500
1.000
0.500
0.000
-0.500
slope ~ multinomial
slight moderate severe unable/extreme
-0.050
homogeneous TTO
slight moderate severe unable/extreme
36. An EQ-5D-5L value set for England
October 30th 2014
The low value of the constant
37. Are (have) we (been) doing this
correctly?
An EQ-5D-5L value set for England
October 30th 2014
• Variation is caused by:
• Differences of opinion
• Errors
• You can value at 0.5 or 0, but you can’t value
at 1.5 or 2
• If it is errors, and not opinions, which are
driving the lower values, the mean may not be
the right measure to reflect ‘average’ opinion
• There is error-censoring at 1
38. An EQ-5D-5L value set for England
October 30th 2014
The low value of the constant
40. An EQ-5D-5L value set for England
October 30th 2014
The final EQ-
5D-5L value
set model
EQ-5D-5L value set for England
constant 1.003
Mobility slight 0.057
moderate 0.074
severe 0.207
unable 0.255
Self care slight 0.059
moderate 0.083
severe 0.176
unable 0.208
Usual activities slight 0.048
moderate 0.067
severe 0.165
unable 0.165
Pain/discomfort slight 0.059
moderate 0.079
severe 0.244
extreme 0.298
Anxiety/depression slight 0.072
moderate 0.099
severe 0.282
extreme 0.282
41. An EQ-5D-5L value set for England
October 30th 2014
EQ-5D-5L value set for England Example: the value for health state 23245
constant 1.000 Constant =1.000
Mobility = 2 0.057 Minus MO level 2 -0.057
Mobility = 3 0.074
Mobility = 4 0.207
Mobility = 5 0.255
Self care = 2 0.059
Self care = 3 0.083 Minus SC level 3 -0.083
Self care = 4 0.176
Self care = 5 0.208
Usual activities = 2 0.048 Minus UA level 2 -0.048
Usual activities = 3 0.067
Usual activities = 4 0.165
Usual activities = 5 0.165
Pain/discomfort = 2 0.059
Pain/discomfort = 3 0.079
Pain/discomfort = 4 0.244 Minus PD level 4 -0.244
Pain/discomfort = 5 0.298
Anxiety/depression = 2 0.072
Anxiety/depression = 3 0.099
Anxiety/depression = 4 0.282
Anxiety/depression = 5 0.282 Minus AD level 5 -0.282
State 23245 =0.286
Calculating
values:
a worked
example
42. An EQ-5D-5L value set for England
October 30th 2014
Comparison with 3L and crosswalk
5L value set Crosswalk value set 3L value set
% health states
worse than dead
3.2% (100 out of
3,125)
26.66% (833 out of
3,125)
34.57% (84 out of
243)
Preferences
regarding
dimensions (from
the most important
to the least
important)
Pain/Discomfort Pain/Discomfort Pain/Discomfort
Anxiety/Depression Mobility Mobility
Mobility Anxiety/Depression Anxiety/Depression
Self-care Self-care Self-care
Usual Activities Usual Activities Usual Activities
Value of 55555
(33333)
-0.208 -0.49 -0.594
Value of 11112* 0.928 0.879 0.848
Value of 11121* 0.941 0.837 0.796
Value of 11211* 0.952 0.906 0.883
Value of 12111* 0.941 0.846 0.815
Value of 21111* 0.943 0.877 0.850
Minimum value -0.208 -0.49 -0.594
Maximum value 1 1 1
Range of values [-0.208, 1] [-0.594, 1] [-0.594, 1]
43. 3L crosswalk
An EQ-5D-5L value set for England
October 30th 2014
Distributions of values
1
1.5
.5
0
Density
-.5 0 .5 1
value
2
1
1.5
.5
0
Density
1
.5
-.5 0 .5 1
value
2
0
1.5
Density
-.5 0 .5 1
value
5L
44. 3L crosswalk
An EQ-5D-5L value set for England
October 30th 2014
Values and ‘misery scores’
-.5
1
.5
0
5 10 15
misery
eq5d3l Fitted values
-.5
1
.5
0
5 10 15 20 25
misery
eq5d5l Fitted values
-.5
1
.5
0
5 10 15 20 25
misery
eq5d5l Fitted values
5L
45. An EQ-5D-5L value set for England
October 30th 2014
Comparing 3L and 5L data
3L value set 5L value set
% logical
inconsistencies
4.89%
(166 out of 3,395)
8.43%
(84 out of 996)
% who do not give
their lowest value to
the worst health state
29.19%
(991 out of 3,395)
28.92%
(288 out of 996)
46. An EQ-5D-5L value set for England
October 30th 2014
Implications of the results
• The 5L Value set for England has a lower range of values than
the current UK EQ-5D value set
• Higher minimum value for 55555 (5L) (-0.208) than 33333
(3L) (-0.56): as expected, given known issues with the Dolan
(1997) value set
• The proportion of health states with negative values is
considerably lower
• No ‘N3’ term – it did not improve the model
• Implies treatments for very severe conditions may have lower
QALY gains than at present
• The greater descriptive sensitivity of the EQ-5D-5L will be
somewhat counteracted by the nature of the 5L value set
compared to the previous 3L value set
47. An EQ-5D-5L value set for England
October 30th 2014
Implications of the results
• For two dimensions, anxiety/depression and usual
activities, the TTO results did not differentiate between
levels 4 and 5
• e.g. interventions that reduce the level of
anxiety/depression from extreme to severe will show few
QALY gains
• Potential implications for other applications of the value
set e.g. in the PROMs programme, where it is used to
measure hospital performance
48. An EQ-5D-5L value set for England
October 30th 2014
Remaining research questions
• This presentation has focussed on the value set for
England – we have also collected data for Scotland, Wales
and NI, and will be estimating a UK value set
• How do values compare with other countries? Over a
dozen 5L value set studies underway internationally, using
a consistent methodology
• Many remaining methodological issues…for example,
– the effect of valuation full health vs. 11111 in the TTO
– Describing health states ‘in context’ of the full health
state descriptive system
– DCE with duration
– Remodelling the 3L value set with the new methods