The document summarizes the development of an EQ-5D-5L value set for England based on stated preference data collected from 1000 members of the general public. Researchers aimed to generate a value set that reflects public preferences using time trade-off (TTO) and discrete choice experiment (DCE) data. A hybrid statistical model was developed to combine the TTO and DCE data and account for heterogeneity in individual preferences. The final model produced utility values for all 3,125 EQ-5D-5L health states based on the combined dataset.
Prof Devlin discusses the rationale for the PROMs programme and provides an overview of the various uses of the EQ-5D in England—for example by NICE in health technology assessment, in population surveys and in the English NHS PROMS program. The presentation also reviews how EQ-5D data are collected, analysed and used in the UK to inform decisions by health care providers, payers and patients.
Prof Devlin discusses the rationale for the PROMs programme and provides an overview of the various uses of the EQ-5D in England—for example by NICE in health technology assessment, in population surveys and in the English NHS PROMS program. The presentation also reviews how EQ-5D data are collected, analysed and used in the UK to inform decisions by health care providers, payers and patients.
This presentation addressed core issues in the challenge of evaluating technologies that treat rare diseases. It draws together information about current economic thinking and practices that most affect decisions, offering suggestions as to where and how the patient's input might be most important and influential.
Background: Circulation of influenza subtypes varies between influenza seasons. Little is known about patterns of circulation from one season to another. We studied the association of influenza virus subtypes detected in consecutive influenza seasons in EU/EEA countries to understand the possible predictive value of the previous season for the upcoming season.
Method: We analysed the sentinel (with systematic sampling) and non-sentinel (with convenience sampling) influenza virological surveillance data reported to the European Surveillance System from all EU/EEA countries during the seasons 2006/07-2013/14. Data were excluded if viruses were not subtyped, the number of detections exceeded the number of tested specimens or if less than 10 specimens were tested per week. Countries were excluded from analysis of any pair of consecutive seasons (cycle) if they reported for <50% of weeks in either season. We assessed the association of weekly A(H1), A(H1) pdm09, A(H3) and B virus-specific detection rates in cycles for sentinel and non-sentinel specimens. We used multilevel Poisson regression with 7 cycles as repeated measures, treated countries as cluster, and corrected for week of reporting. A sensitivity analysis was performed omitting the 2009 pandemic cycle. Associations were reported as incidence rate ratios (IRR) and 95% confidence intervals (CI).
Conclusion: Six-11 countries reported sentinel and 3-10 non-sentinel data per each cycle. The proportion of sentinel and non-sentinel influenza detections varied by (sub)type across seasons, being highest for the A(H1)pdm09 subtype during season 2009/10 (99.4%; 99.3%). The A(H3) detections were highest during 2006/07 (92.5; 91.1%). The highest proportion of influenza B was observed in 2012/13 in sentinel (64.2%) and 2007/08 in non-sentinel specimens (78.1%).
Significant associations between consecutive seasonal influenza rates were found for A(H1) (2.73;1.33-5.61, p=0.006), A(H1)pdm09 (4.31;1.92-9.67, p<0.001)><0.001) virus in the sentinel system and for A(H1) (2.70;1.00-7.30, p=0.049), A(H1)pdm09 (3.87;1.50-10.01, p=0.005) and B (0.7;0.51-0.98, p=0.039) in the non-sentinel system. When omitting the pandemic cycle, the association remained significant for A(H1) and A(H1)pdm09 in the sentinel system.
The virological influenza surveillance data suggest that influenza A(H1) and B virus circulation during any season is associated with the circulation in the forthcoming season. Vaccination coverage and vaccine effectiveness have probably an impact on the results and cause country variation as well, however, they were not within the scope of this study.
Common mistakes in measurement uncertainty calculationsGH Yeoh
The basic calculation for measurement uncertainty (MU) is through the law of propagation of uncertainty. Some find it difficult to apply and make some mistakes in the MU evaluation.
Do EQ-5D-3L and EQ-5D-5L Capture the Same Changes in Quality of Life Over Tim...Office of Health Economics
Slides from a presentation given by OHE's Patricia Cubi-Molla and Paula Lorgelly on a EQ-5D-3L and EQ-5D-5L longitudinal study of cancer patients: do they capture the same changes in quality of life over time?
Assessing convergent and discriminant validity in the
ADHD-R IV rating scale:
User-written commands for Average Variance Extracted
(AVE), Composite Reliability (CR), and
Heterotrait-Monotrait ratio of correlations (HTMT).
This presentation addressed core issues in the challenge of evaluating technologies that treat rare diseases. It draws together information about current economic thinking and practices that most affect decisions, offering suggestions as to where and how the patient's input might be most important and influential.
Background: Circulation of influenza subtypes varies between influenza seasons. Little is known about patterns of circulation from one season to another. We studied the association of influenza virus subtypes detected in consecutive influenza seasons in EU/EEA countries to understand the possible predictive value of the previous season for the upcoming season.
Method: We analysed the sentinel (with systematic sampling) and non-sentinel (with convenience sampling) influenza virological surveillance data reported to the European Surveillance System from all EU/EEA countries during the seasons 2006/07-2013/14. Data were excluded if viruses were not subtyped, the number of detections exceeded the number of tested specimens or if less than 10 specimens were tested per week. Countries were excluded from analysis of any pair of consecutive seasons (cycle) if they reported for <50% of weeks in either season. We assessed the association of weekly A(H1), A(H1) pdm09, A(H3) and B virus-specific detection rates in cycles for sentinel and non-sentinel specimens. We used multilevel Poisson regression with 7 cycles as repeated measures, treated countries as cluster, and corrected for week of reporting. A sensitivity analysis was performed omitting the 2009 pandemic cycle. Associations were reported as incidence rate ratios (IRR) and 95% confidence intervals (CI).
Conclusion: Six-11 countries reported sentinel and 3-10 non-sentinel data per each cycle. The proportion of sentinel and non-sentinel influenza detections varied by (sub)type across seasons, being highest for the A(H1)pdm09 subtype during season 2009/10 (99.4%; 99.3%). The A(H3) detections were highest during 2006/07 (92.5; 91.1%). The highest proportion of influenza B was observed in 2012/13 in sentinel (64.2%) and 2007/08 in non-sentinel specimens (78.1%).
Significant associations between consecutive seasonal influenza rates were found for A(H1) (2.73;1.33-5.61, p=0.006), A(H1)pdm09 (4.31;1.92-9.67, p<0.001)><0.001) virus in the sentinel system and for A(H1) (2.70;1.00-7.30, p=0.049), A(H1)pdm09 (3.87;1.50-10.01, p=0.005) and B (0.7;0.51-0.98, p=0.039) in the non-sentinel system. When omitting the pandemic cycle, the association remained significant for A(H1) and A(H1)pdm09 in the sentinel system.
The virological influenza surveillance data suggest that influenza A(H1) and B virus circulation during any season is associated with the circulation in the forthcoming season. Vaccination coverage and vaccine effectiveness have probably an impact on the results and cause country variation as well, however, they were not within the scope of this study.
Common mistakes in measurement uncertainty calculationsGH Yeoh
The basic calculation for measurement uncertainty (MU) is through the law of propagation of uncertainty. Some find it difficult to apply and make some mistakes in the MU evaluation.
Do EQ-5D-3L and EQ-5D-5L Capture the Same Changes in Quality of Life Over Tim...Office of Health Economics
Slides from a presentation given by OHE's Patricia Cubi-Molla and Paula Lorgelly on a EQ-5D-3L and EQ-5D-5L longitudinal study of cancer patients: do they capture the same changes in quality of life over time?
Assessing convergent and discriminant validity in the
ADHD-R IV rating scale:
User-written commands for Average Variance Extracted
(AVE), Composite Reliability (CR), and
Heterotrait-Monotrait ratio of correlations (HTMT).
On 31 October 2019, Adrian Towse and Chris Henshall from the Office of Health Economics (OHE) presented at the G20 meeting on antimicrobial drugs R&D in Paris organised by the Wellcome Trust. The topic of their presentation was HTA and payment mechanisms for new drugs to tackle antimicrobial resistance.
This presentation looks at ways in which governments can set prices, including “cost plus”, value, and the external referencing of prices elsewhere. It looks at the role that competition can play in keeping down prices. In that context it briefly discusses pricing proposals being considered in Malaysia. It makes the case for using HTA to inform pricing decisions.
Adrian Towse
% GDP spending in UK, G5 countries and OECD upper middle income countries. W...Office of Health Economics
This presentation looks at rates of GDP spend on health care, distinguishing between categories of country (i.e. levels of GDP pre capita). It looks at the relationship between rates of spending and moves to universal health coverage, and explores alternative ways of increasing expenditure and making decisions about which services to provide with the money available.
The role of real world data and evidence in building a sustainable & efficien...Office of Health Economics
This presentation defines RWD and RWE in the context of digital health, and looks at potential uses for RWD and RWE. It briefly sets out the current landscape in Malaysia and looks at the challenges in using RWE. In particular, the issues of access, governance and ensuring good quality are considered.
The aim of this educational symposium was to discuss why we should seek value across the health care system and how we can apply existing research methods to measure the value of services. While considerable political attention in developed countries continues to be focused on drug spending, there is also growing awareness of the significant contribution of non-drug components of health care (e.g., hospital services and inefficient care delivery) to overall spending growth and patient affordability. At the same time, there is growing interest in making greater use of value assessment and value-based payment to control spending and better align it with care quality. In order to promote greater value, and to do so in ways that respond to the needs of payers and patients, it is essential to assess value across both drug- and non-drug interventions and health care services. This panel will offer expert viewpoints to identify and discuss gaps in value information, rationale and approaches to track and reduce system-wide low value care, and research methods for how to measure health care services.
Role Substitution, Skill Mix, and Provider Efficiency and Effectiveness : Les...Office of Health Economics
Graham participated in an organised session on Monday July 15th 2019. In the session he presented his paper with his co-author Ioannis Laliotis from the London School of Economics. The paper revisits the relationship between workforce and maternity outcomes in the English NHS in an attempt to contribute knowledge to an important policy question for which there has been a paucity of research.
This research explores the feasibility of introducing an Outcome-Based Payment approach for new cancer drugs in England. A literature review explored the current funding landscape in England, the available evidence on existing OBP schemes internationally, and
which outcomes cancer patients value most. Two focus groups and an online survey with patients and carers, as well as interviews with NHS and government stakeholders, healthcare
professionals, and pharmaceutical industry representatives, provided additional evidence on the feasibility and suitability of OBP schemes
Understanding what aspects of health and quality of life are important to peopleOffice of Health Economics
Poster presentation from the EuroQol Plenary Meeting 2019, Brussels, Belgium. By Koonal Shah, Brendan Mulhern, Patricia Cubi-Molla, Bas Janssen, and David Mott.
Koonal presented as part of an organised session on ‘moving beyond conventional economic approaches in palliative and end of life care’. He summarised the empirical evidence on the extent of pubic support for an end of life premium, before discussing some novel approaches that have been used in recent studies. His presentation was discussed by Helen Mason of Glasgow Caledonian University.
Author(s) and affiliation(s): Koonal Shah, Office of Health Economics
Event: iHEA Congress
Date: 17/07/2019
Location: Basel, Switzerland
Assessing the Life-Cycle Value Added of Second Generation Antipsychotics in S...Office of Health Economics
This research presented in a poster at HTAi 2019, Cologne (Germany) by a team of OHE and IHE researchers, estimates the value added by second generation antipsychotics over their life-cycle in the UK and Sweden. It concludes that considering the entire life-cycle, the value added by SGAs to the system is higher than the expected value estimated at launch. P&R decisions should consider how to measure, capture and take into account the value added by medicines over the long-run.
Author(s) and affiliation(s): Mikel Berdud (Office of Health Economics, London), Niklas Wallin-Bernhardsson (Institute for Health Economics, Stockholm), Bernarda Zamora (Office of Health Economics, London), Peter Lindgren (Institute for Health Economics, Stockholm), Adrian Towse (Office of Health Economics, London)
Event: HTAi 2019 Annual Meeting
Date: 18/06/2019
Location: Cologne, Germany
There is growing recognition that HTA and contracting systems for antimicrobials need to be adapted to help fight the threat of antimicrobial resistance (AMR), but there is little agreement on how. This poster reports findings from a literature review, expert interviews and face-to-face discussions at a Forum on the current HTA and payment systems for antibiotics across Europe and a number of recommendations for adapting these systems to respond to the challenges of AMR.
Author(s) and affiliation(s): Margherita Neri (OHE) Grace Hampson (OHE) Christopher Henshall (OHE visiting fellow, independent consultant) Adrian Towse (OHE)
Event: HTAi annual conference 2019
Date: 18/06/2019
Location: Cologne, Germany
Assessing the Life-cycle Value Added of Second-Generation Antipsychotics in S...Office of Health Economics
This study aims to guide access decisions and drive the discussion on access and price, through recognition of the dynamic nature of value added by pharmaceutical innovation over the long-run. The analysis of the life-cycle value of risperidone estimates the value generated in the UK and Sweden. Results show that health systems were able to appropriate most of the life-cycle value generated, and this is larger than estimated at launch.
Author(s) and affiliation(s): Mikel Berdud(1), Niklas Wallin-Bernhardsson(2), Bernarda Zamora(1), Peter Lindgren(2), and Adrian Towse(1) (1) Office of Health Economics (2) The Swedish Institute for. Health Economics
Event: XXXIX JORNADAS DE ECONOMÍA DE LA SALUD
Date: 12/06/2019
Location: Albacete, Spain
Prescribed Specialised Services (PSS) Commissioning for Quality and Innovation (CQUIN) schemes were launched in 2013 in England with the aim of improving the quality of specialised care and achieving value for money. During this presentation, Marina Rodes Sanchez described the key features of the schemes and discussed its strengths and weaknesses based on international pay-for-performance literature.
Author(s) and affiliation(s): Yan Feng, Queen Mary University of London; Søren Rud Kristensen, Imperial College London; Paula Lorgelly, King’s College London; Rachel Meacock, University of Manchester; Marina Rodes Sanchez, Office of Health Economics; Luigi Siciliani, University of York; Matt Sutton, University of Manchester
Event: XXXIX Spanish Health Economics Association Conference
Date: 12/06/2019
Location: Albacete, Spain
In this session, Meng Li sets out estimates of real option value for drugs arguing that option value matters and can be calculated. Adrian Towse sets out likely payer concerns about incorporating real option value into decision making. Meng Li responds to these concerns. Jens Grueger sets out how industry considers investment opportunities, arguing that if patients (and society) have preferences these need to be reflected in P&R decisions.
Author(s) and affiliation(s): Meng Li, Postdoctoral Research Fellow, Leonard D Schaeffer Center, University of Southern California, Los Angeles, CA, USA. Adrian Towse, Emeritus Director, Office of Health Economics, London, UK Jens Grueger, formerly Head of Global Access, Senior Vice President at F. Hoffmann-La Roche
Event: ISPOR 2019
Location: New Orleans, USA
Date: 21/05/2019
MCDA OR WEIGHTED CEA BASED ON THE QALY? WHICH IS THE FUTURE FOR HTA DECISION ...Office of Health Economics
In this ISPOR session Chuck Phelps and Adrian Towse debated the case for and against using MCDA to support HTA decision making, as compared to weighting or augmenting a QALY based ICER approach. Chuck Phelps argued for use of MCDA, Adrian Towse for weighting the QALY. Nancy Devlin set the scene and moderated.
Author(s) and affiliation(s): Nancy Devlin, Director, Centre for Health Policy, University of Melbourne, Australia Adrian Towse, Emeritus Director, Office of Health Economics, London, UK Chuck Phelps, University of Rochester, Rochester, NY USA
Event: ISPOR 2019
Location: New Orleans, USA
Date: 21/05/2019
This presentation, created by Syed Faiz ul Hassan, explores the profound influence of media on public perception and behavior. It delves into the evolution of media from oral traditions to modern digital and social media platforms. Key topics include the role of media in information propagation, socialization, crisis awareness, globalization, and education. The presentation also examines media influence through agenda setting, propaganda, and manipulative techniques used by advertisers and marketers. Furthermore, it highlights the impact of surveillance enabled by media technologies on personal behavior and preferences. Through this comprehensive overview, the presentation aims to shed light on how media shapes collective consciousness and public opinion.
Collapsing Narratives: Exploring Non-Linearity • a micro report by Rosie WellsRosie Wells
Insight: In a landscape where traditional narrative structures are giving way to fragmented and non-linear forms of storytelling, there lies immense potential for creativity and exploration.
'Collapsing Narratives: Exploring Non-Linearity' is a micro report from Rosie Wells.
Rosie Wells is an Arts & Cultural Strategist uniquely positioned at the intersection of grassroots and mainstream storytelling.
Their work is focused on developing meaningful and lasting connections that can drive social change.
Please download this presentation to enjoy the hyperlinks!
1. Nancy Devlin & Ben van Hout
on behalf of the OHE & ScHARR research team
OHE seminar
London • 30 October 2014
An EQ-5D-5L value set for England
2. An EQ-5D-5L value set for England
30 October 2014
This project is independent research commissioned and funded by
the NIHR / Department of Health Policy Research Programme (‘EQ-
5D-5L Value Set for England’ - 070/0073). Additional funding was
also received from the EuroQol Research Foundation.
The views expressed in this presentation are those of the authors,
and not necessarily those of the funding bodies.
Note: The value set reported here has ‘interim’ status, until such
point as it is accepted for publication in a peer reviewed journal.
Please do not quote from or circulate these slides without
permission of the presenting authors.
Note: There are a small number of differences between the content
of these slides and of those presented at the OHE lunchtime seminar
in October 2014. The latest analyses are presented in this version.
Disclaimer
3. An EQ-5D-5L value set for England
30 October 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
30 October 2014
Background
• EQ-5D-5L: 3,125 states cf. 243 in the EQ-5D
• An important instrument: requests for 5L use now
supersede requests for 3L
• 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 the stated
preferences of the general public (e.g. NICE 2013)
5. An EQ-5D-5L value set for England
30 October 2014
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’
6. An EQ-5D-5L value set for England
30 October 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
30 October 2014
Study design
• Research protocol developed by the EuroQol Research Foundation
• 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 respondent 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 Valuation Technology software (EQ-VT) was used to
present the tasks and to capture respondents’ responses
9. An EQ-5D-5L value set for England
30 October 2014
TTO for values > 0
(states better than dead)
Example shown:
U(hi) = 5/10 = 0.5
U(hi) = (x/t)
where x is the time in
full health and t is the
time in health state hi at
the respondent’s point of
indifference
10. An EQ-5D-5L value set for England
30 October 2014
Example shown:
U(hi) = (5-10)/10
= -0.5
t = 20 years
lead time (LT) = 10 years
U(hi) = (x-LT)/(t-LT)
= (x-10)/10
Min value = -1
TTO for values < 0
(states worse than dead)
12. An EQ-5D-5L value set for England
30 October 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
13. An EQ-5D-5L value set for England
30 October 2014
DCE data
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
-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
Proportions choosing A and B based on relative
severities of A and B
% B
% A
0 1
-10-50510
delta sum of scores
difinmisery
14. An EQ-5D-5L value set for England
30 October 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. An EQ-5D-5L value set for England
30 October 2014
1 3 4 5 6 7 8
-0.10-0.050.00
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
30 October 2014
11121 mean= 0.876
value
density
-1.0 0.0 0.5 1.002040
12111 mean= 0.868
value
density
-1.0 0.0 0.5 1.0
02040
11211 mean= 0.866
value
density
-1.0 0.0 0.5 1.0
02040
11221 mean= 0.862
value
density
-1.0 0.0 0.5 1.0
01020
21111 mean= 0.83
value
density
-1.0 0.0 0.5 1.002040
12121 mean= 0.823
value
density
-1.0 0.0 0.5 1.0
0515
11112 mean= 0.815
value
density
-1.0 0.0 0.5 1.0
02040
11122 mean= 0.806
value
density
-1.0 0.0 0.5 1.0
0515
11212 mean= 0.801
value
density
-1.0 0.0 0.5 1.0
0515
Distributions, by state
17. An EQ-5D-5L value set for England
30 October 2014
54231 mean= 0.473
value
density
-1.0 0.0 0.5 1.004812
33253 mean= 0.465
value
density
-1.0 0.0 0.5 1.0
0515
12334 mean= 0.463
value
density
-1.0 0.0 0.5 1.0
048
23514 mean= 0.46
value
density
-1.0 0.0 0.5 1.0
04812
43514 mean= 0.443
value
density
-1.0 0.0 0.5 1.0048
15151 mean= 0.436
value
density
-1.0 0.0 0.5 1.0
051015
23152 mean= 0.435
value
density
-1.0 0.0 0.5 1.0
0515
31525 mean= 0.428
value
density
-1.0 0.0 0.5 1.0
0510
31524 mean= 0.423
value
density
-1.0 0.0 0.5 1.002468
Distributions, by state
18. An EQ-5D-5L value set for England
30 October 2014
21444 mean= 0.148
value
density
-1.0 -0.5 0.0 0.5 1.0
01020
53244 mean= 0.148
value
density
-1.0 -0.5 0.0 0.5 1.0
0515
52455 mean= 0.12
value
density
-1.0 -0.5 0.0 0.5 1.0
0510
43555 mean= 0.119
value
density
-1.0 -0.5 0.0 0.5 1.0
051015
55555 mean= 0.016
value
density
-1.0 -0.5 0.0 0.5 1.0
0100
NA mean= NA
value
density
-1.0 -0.5 0.0 0.5 1.0
-1.00.01.0
Distributions, by state
19. An EQ-5D-5L value set for England
30 October 2014
minimum value
minv
Frequency
-1.0 -0.5 0.0 0.5 1.0
050100150
maximum value
maxv
Frequency
0.0 0.2 0.4 0.6 0.8 1.0
0200400
standard deviation of values
(varv^0.5)
Frequency
0.0 0.2 0.4 0.6 0.8 1.0
050100150
range of values used
range
Frequency
0.0 0.5 1.0 1.5 2.0
050100150
Descriptive statistics
20. An EQ-5D-5L value set for England
30 October 2014
• Overall, English 5L valuation data have acceptable ‘face validity’:
the worse the health state, the lower the mean and median value
TTO data
21. An EQ-5D-5L value set for England
30 October 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. An EQ-5D-5L value set for England
30 October 2014
• Our process for examining the individual-level
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
Interpretation of the data
23. An EQ-5D-5L value set for England
30 October 2014
Examination of individual-level data
24. An EQ-5D-5L value set for England
30 October 2014
Examination of individual-level data
25. An EQ-5D-5L value set for England
30 October 2014
Real or censored???
27. An EQ-5D-5L value set for England
30 October 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
30 October 2014
• 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).
• 20 parameter preferred on prior grounds
• All models were estimated for both TTO and DCE data, and
‘hybrids’ of the these
• Final model based on the hybrid
• Heterogeneity explored via random coefficient models,
which estimate value functions for every individual
member of the sample
• Values at -1 treated as censored
• Truncation of the error distribution at 1 is addressed
Key aspects of the modelling
29. An EQ-5D-5L value set for England
30 October 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
30 October 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. An EQ-5D-5L value set for England
30 October 2014
• 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
follows a:
– Normal distribution
– Lognomal distribution
– Multinomial distribution
– (3 latent classes)
Heterogeneity
-1.5
-1
-0.5
0
0.5
1
value
x'beta
35. An EQ-5D-5L value set for England
30 October 2014
Parameter estimates
-0.050
0.000
0.050
0.100
0.150
0.200
0.250
0.300
0.350
slope ~ normal
slight moderate severe unable/extreme
-0.100
0.000
0.100
0.200
0.300
0.400
0.500
0.600
slope ~ lognormal
slight moderate severe unable/extreme
-0.500
0.000
0.500
1.000
1.500
2.000
2.500
3.000
3.500
slope ~ multinomial
slight moderate severe unable/extreme
-0.050
0.000
0.050
0.100
0.150
0.200
0.250
0.300
homogeneous TTO
slight moderate severe unable/extreme
36. An EQ-5D-5L value set for England
30 October 2014
The low value of the constant
37. An EQ-5D-5L value set for England
30 October 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
Are (have) we (been) doing this
correctly?
38. An EQ-5D-5L value set for England
30 October 2014
The low value of the constant
40. An EQ-5D-5L value set for England
30 October 2014
The
resulting
EQ-5D-
5L value
set
model
England EQ-5D-5L values 95% CIs
constant 1.003 (0.983 - 1.019)
Mobility slight 0.057 (0.043 - 0.075)
moderate 0.075 (0.057 - 0.093)
severe 0.208 (0.190 - 0.227)
unable 0.255 (0.237 - 0.275)
Self-care slight 0.058 (0.045 - 0.074)
moderate 0.083 (0.061 - 0.101)
severe 0.176 (0.157 - 0.197)
unable 0.208 (0.189 - 0.225)
Usual activities slight 0.048 (0.033 - 0.066)
moderate 0.067 (0.047 - 0.086)
severe 0.165 (0.147 - 0.180)
unable 0.165 (0.152 - 0.184)
Pain/discomfort slight 0.059 (0.042 - 0.075)
moderate 0.080 (0.059 - 0.098)
severe 0.245 (0.225 - 0.264)
extreme 0.298 (0.278 - 0.317)
Anxiety/depression slight 0.073 (0.058 - 0.089)
moderate 0.099 (0.079 - 0.119)
severe 0.282 (0.263 - 0.298)
extreme 0.282 (0.267 - 0.300)
41. An EQ-5D-5L value set for England
30 October 2014
EQ-5D-5L value set for England Example: the value for health state 23245
constant 1.003 Constant =1.003
Mobility = 2 0.057 Minus MO level 2 -0.057
Mobility = 3 0.075
Mobility = 4 0.208
Mobility = 5 0.255
Self-care = 2 0.058
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 0.080
Pain/discomfort = 4 0.245 Minus PD level 4 -0.245
Pain/discomfort = 5 0.298
Anxiety/depression = 2 0.073
Anxiety/depression = 3 0.099
Anxiety/depression = 4 0.282
Anxiety/depression = 5 0.282 Minus AD level 5 -0.282
State 23245 = 0.288
EQ-5D-5L
values for
England:
a worked
example
42. An EQ-5D-5L value set for England
30 October 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)
-0.205 -0.49 -0.594
Value of 11112* 0.927 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* 0.942 0.846 0.815
Value of 21111* 0.943 0.877 0.850
Minimum value -0.205 -0.49 -0.594
Maximum value 1 1 1
Range of values [-0.205, 1] [-0.594, 1] [-0.594, 1]
43. An EQ-5D-5L value set for England
30 October 2014
Distributions of values
0
.5
1
1.5
Density
-.5 0 .5 1
value
0
.5
1
1.5
2
Density
-.5 0 .5 1
value
0
.5
1
1.5
2
Density
-.5 0 .5 1
value
3L crosswalk
5L
44. An EQ-5D-5L value set for England
30 October 2014
Values and ‘misery scores’
-.5
0
.5
1
5 10 15
misery
eq5d3l Fitted values
-.5
0
.5
1
5 10 15 20 25
misery
eq5d5l Fitted values
-.5
0
.5
1
5 10 15 20 25
misery
eq5d5l Fitted values
3L crosswalk
5L
45. An EQ-5D-5L value set for England
30 October 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
30 October 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) (-0.205) 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
30 October 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
30 October 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