eq-5d - 3l or 5l?
TRANSCRIPT
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
The issue
■ For many years NICE has recommended use of the 3L, and its UK value set (Dolan 1997)
■ 5L available (Herdman et al 2011) with value sets eg. England (Devlin et al 2017)
■ In the UK: choice whether to use the 5L (and English value set) or the 3L (and UK value set) likely to affect estimates of QALYs and ICERs.
- QoL increasing: QALY gains likely to be reduced (to varying degrees)
- LoL increasing: QALY gains likely to be increased (to varying degrees)
■ These differences create a challenge for NICE
- potential inconsistency between past/future HTA decisions
- Inconsistency between future decisions where some evidence based on the 3L, and some based on 5L mapped to 3L values
■ Other countries may also face transitional issues
2 21-22 September 2017
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Sources of differences between the 3L and 5L
■ There are differences in both
■ The interaction between these determines the effect on QALYs
■ We have empirical evidence; and further studies underway.
3 21-22 September 2017
Measurement properties Valuation properties
How do patients respond to the questionnaire?
What are the stated preferences of the general public?
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
3L vs 5L as measures of patient health
■ 5L was developed in response to concerns about the 3L.
■ Evidence shows that the 5L, compared to the 3L, has:
4 21-22 September 2017
Reduced ceiling effects Eg. Mukuria et al (2017); Janssen et al (2013); Agborsayanga et al (2014); Feng et al (2015);
Less ‘clustering’ of data on relatively few health states
Eg. Parkin et al (2016); Zamora et al (2017a); Feng et al (2016) ; Buccholz et al (2017)*
Evidence of better discriminatory power among patient subgroups
Eg. Janssen et al (2017)*; Pan et al (2015); Wang et al (2016).
Important improvements in the 5L with respect to problems with mobility
Eg. Janssen et al (2017)*
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
EQ-5D-3L value set EQ-5D-5L crosswalk EQ-5D-5L value set
% health states worse than dead 34.6% 26.7% 5.1%
Dimension importance order Pain/Discomfort
Mobility
Anxiety/depression
Self-care
Usual Activities
Pain/Discomfort
Mobility
Anxiety/Depression
Self-care
Usual Activities
Pain/Discomfort
Anxiety/depression
Mobility
Self-care
Usual Activities
Health state values
‘Mildest’ state (11211) 0.883 0.906 0.950 (11211/12111)
‘Moderate’ state (22222 (3L) or
33333 (5L))
0.516 0.516 0.593
‘Worst’ state (33333 (3L) or
55555 (5L))
-0.594 -0.594 -0.285
5 21-22 September 2017
Comparing 3L(UK) and 5L(England) value sets
Source: Mulhern et al (2017)
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Comparing 3L(UK) and 5L(England) value sets
■ The ‘bi-modal’ nature of the UK 3L value set values is not present in the England 5L values.
6 21-22 September 2017
0.5
11
.52
Den
sity
-.5 0 .5 1Utility value
EQ-5D-5L EQ-5D-3L
Crosswalk
Source: Mulhern et al (2017)
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain7 21-22 September 2017
Comparing EQ-3L Index and EQ-5L Index data
EQ Index = Patients’ EQ-5D profile data to which the value set has been applied
Source: Parkin, Devlin, Feng (2016) Medical Decision Making
“non-normal distributions of 3L data, featuring two distinct groups, are commonly observed…this is a result of the classification system…the UK 3L value set exacerbates this by placing a larger weight on level 3 observations”
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Distributions of the EQ Index applying the van
Hout mapping vs England 5L value set.
8 21-22 September 2017
Note: these two diagrams both use the same NHS data set of 5L profile data reported by patients.
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Issues with relying on the van Hout mapping
9 21-22 September 2017
The evaluation space between 0.5 to 1.0 – where most people are – includes only 31 health states in the 3L. In the 0.8-1.0 space the situation is even worse
- The van Hout mapping allows 5L data to be valued using the 3L UK value set (so has all the problems of the latter)
PLUS
- Discontinuities in the distribution attributable to the mapping function (this issue is not specific to the UK/England)
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Reasons which apply to all countries Reasons specific to the UK/England
Difference in the preference elicitation tasks eg. improved TTO procedure for obtainingvalues < 0; inclusion of DCE tasks
There is a 20 year gap between the stated preference data collected. The composition of the general public and their preferences may have changed
A different instrument is being valued eg. mobility level 5 is descriptively different from mobility level 3 in the 3L.
The UK 3L value set is unusual in various respects eg. high % values < 0; ‘two-group’ distribution of values. Any new value set, whether for 3L or 5L, unlikely to replicate its properties.
Differences in modelling approaches Innovative modelling: heterogeneity; heteroscedasticity; hybrid; ‘censoring’
10 21-22 September 2017
Why are 3L and 5L values different?
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Is the UK 3L value set the ‘odd one out’..?
11 21-22 September 2017
■ 1998 UK ‘replication study’ (Kind & Macran 1999)
■ Values systematically higher than the UK 3L value set
■ Similar findings reported in Tsuchiya et al (2006)
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
MVH (1993)
Yo
rk (
19
98
)
UK 3L UK replication England 5L
Value for ‘pits’ -0.594 -0.126 -0.285
% states < 0 34.6% 12.3% 5.1%
Value for ‘middle’ state 0.516 0.616 0.593
www.euroqol.org34th EuroQol Plenary Meeting 2017 – Barcelona, Spain
Concluding remarks
■ Evidence points to the 5L being a better descriptive system than the 3L: it more accurately describes patients’ health.
■ The value sets for the 5L are based on more up to date preference data, and draw on improvements in methods developed over the last 20 years
■ 5L value sets are different than the 3L value sets: this doesn’t mean that there is a ‘problem’ with the 5L that needs to be ‘fixed’
■ Mulhern et al (2017): ‘QALYs may have been overestimated using the UK 3L value set’.
■ England 5L value set (correctly, given implications for NICE) subject to external validation
■ But the UK 3L value set which continues to be used was never subject to that scrutiny.
■ The transitional challenge facing NICE is inevitable in shifting to any new preference-based measure
12 21-22 September 2017