The existence of important dissimilarities between EQ-5D-3L and EQ-5D-5L, both in terms of the health profiles and preference-based values, is a key topic in current research. This study compares the performance of the 3L and 5L versions of the EQ-5D in capturing changes in quality of life and the resulting impact on estimates of QALYs for a large cohort of cancer patients. Data were obtained from Cancer2015, a large-scale longitudinal cancer cohort study in Australia. Cancer 2015 enrols newly diagnosed, treatment-naïve cancer patients, who complete quality of life questionnaires at baseline, and at various follow-up points (approximately 3 and/or 6 months continuously). Genetic Matching techniques are used to construct a match comparison group of patients. Post-matching regression adjustment is also implemented to control for any remaining imbalances. For matched QLQ-C30 profiles, we compare 3L and 5L tariffs, as well as the magnitude of changes in quality of life at different points along the treatment trajectory of individuals. We pay special attention to the sensitivity of the measures, by exploring the impact of 3L vs 5L on larger changes in quality of life compared to smaller changes. Our analysis finds that improvements in HRQoL as measured by the QLU-C10D (which is derived from the condition specific EORTC QLQ-C30 instrument) appear to be associated with smaller changes in utility quantified by the 5L compared to the 3L. When HRQoL is deteriorating between observations then the 5L tariff is found to produce bigger utility losses. While the crosswalk (a) loses the increased sensitivity of the 5L (if it detects more change) but (b) it stretches out utility values across a larger range (the 3L range), and hence gains or losses are larger and more in line with the 3L tariffs. Author(s) and affiliation(s): Paula Lorgelly (OHE), Patricia Cubi-Molla (OHE), Mark Pennington (King's College London), Richard Norman (Curtin) Conference/meeting: EuHea 2018 Location: Maastricht, Netherlands Date: 13/07/2018

1. 12-13 June 2018
Do EQ-5D-3L and EQ-5D-5L capture the
same changes in quality of life over time?
A longitudinal study of cancer patients
Paula Lorgelly*, Patricia Cubi-Molla, Mark Pennington,
Richard Norman
*Office of Health Economics & King’s College London

2. 3L vs 5L in cancer - EuHEA 2018
EQ-5D-3L
Developed: 1990
Tariffs: 1997-
EQ-5D-5L
Developed: 2009
Tariffs: 2012-

3. 3L vs 5L in cancer - EuHEA 2018
Background/Motivation
• New instruments and their resulting data should always be
subjected to validation and comparison
• Comparisons of the 3L and 5L are particularly important as the
5L is an obvious successor to the 3L
• How the 3L and 5L capture changes in quality of life (i.e.
QALYs) is a key issue for health technology assessment
• Much has been written comparing the 3L and 5L cross-sectionally, few
longitudinal analyses

4. 3L vs 5L in cancer - EuHEA 2018
UK context
• NICE’s interim position statement:
• The 3L value set to be used for reference case analyses
• Where 5L data have been collected, reference case analyses should calculate
utilities by mapping the 5L descriptive system data onto the 3L value set
(crosswalk mapping function by van Hout et al. (2012))
• NICE supports sponsors of prospective clinical studies continuing to use 5L to
collect data on quality of life
• Effectively use the 3L until further notice …

5. 3L vs 5L in cancer - EuHEA 2018
UK/English tariff issue
• The 5L has more health states than the 3L: 3,125 vs 243
• But they are contained within a smaller space: the 5L English
tariff ranges from -0.285 to 1, compared with the 3L UK tariff
-0.594 to 1

6. 3L vs 5L in cancer - EuHEA 2018
Evidence to date to support position
• Early econometric modelling implies that the 3L and the 5L will
produce substantially different estimates of cost
effectiveness (Hernandez-Alava et al, 2018)
• Used a two-step procedure:
1. External data  mapping algorithm between 3L and 5L utilities  formula
to predict 5L responses from 3L
2. Cost-effectiveness (CE) case studies (3L version)  apply formula 
translate CE results into 5L  compare

7. 3L vs 5L in cancer - EuHEA 2018
Critique of that approach
• Individuals’ unobserved characteristics (latent factors underlying
3L and 5L responses are not uniform across both versions)
• Differential impact across the distribution of health (fixed formula
can be misleading)
• Reduction of the sensitivity (mapping can only ever
decrease sensitivity, the sensitivity introduced by
the 5 levels is then lost in the mapping to the
3L)

8. 3L vs 5L in cancer - EuHEA 2018
Objectives
1. Compare 3L and 5L results at an individual patient-level by
quantifying changes in health assessed by the 3L and 5L
instruments
2. Compare the performance of the 3L and 5L versions of the EQ-
5D in capturing changes in quality of life over time in a cohort of
cancer patients

9. 3L vs 5L in cancer - EuHEA 2018
Data: Cancer 2015
• Cancer 2015 is a longitudinal prospective
population-based genomic cancer cohort in Australia
• EQ-5D, EQ-VAS and EORTC QLQ-C30 were asked at baseline
(diagnosis) and at various follow-up points (in our sample max of
seven times/4 years follow-up)
• The 3L was used from 2012 to October 2015,
since October 2015 the 5L has been used
• Some (earlier) patients only completed the 3L, some (later)
patients only completed the 5L, while some started on the 3L and
moved to the 5L

10. 3L vs 5L in cancer - EuHEA 2018
Methods 1
• Responses were scored using the UK 3L or English 5L tariffs.
5L responses were also scored using the crosswalk.
• Responses to the QLQ-C30 were used to generate EORTC-8D and
QLU-C10D values.
• Pairs of observations (before and after) were created for each
respondent, each pair was considered a separate observation
• ‘3L pairs’ were matched with the ‘5L pairs’ using Genetic Matching
(GenMatch)

11. 3L vs 5L in cancer - EuHEA 2018
Methods 2
• Pairs were matched using the 10 dimensions of the QLU-C10D.
In addition, age, gender and whether first observation was
baseline observation were included in the matching
• The analysis considers differences in the change in utility
values for the sample as a whole and by change in C10D
quintiles
• Compared unadjusted differences and post-match regression
adjusted differences (explanatory variables in an OLS where the
matching variables)

12. 3L vs 5L in cancer - EuHEA 2018
Method: GenMatch

13. 3L vs 5L in cancer - EuHEA 2018
Methods 3
• Quality of the matches was assessed using a bootstrapped
Kolmogorov-Smirnov test for equality of distributions for
continuous data and a t-test for dichotomous data
• Differences in EQ-5D tariff were evaluated after applying the 3L
and 5L tariff values, and after using the crosswalk to generate
3L tariffs from 5L responses
• To test the quality of the matching, we also evaluated the
differences in the EORTC-8D utility scores between matched
profiles/pairs

14. 3L vs 5L in cancer - EuHEA 2018
Differencesinutility
Change in the C10D values
Utility changes are similar across subgroups
matched pairs; ns at p<0.05; matched pairs
are similar
When using EQ-5D tariffs get
divergence: when HRQoL is
improving 5L generates smaller
utility gains, where HRQoL is
worsening the 5L tariffs
generate bigger utility losses
Results

15. 3L vs 5L in cancer - EuHEA 2018
Summary of main results
• Improvements in HRQoL as measured by the QLU-C10D (which
is derived from the condition specific EORTC QLQ-C30
instrument) appear to be associated with smaller changes in
utility quantified by the 5L compared to the 3L
• When HRQoL is deteriorating between observations then the 5L
tariff is found to produce bigger utility losses
• The crosswalk (a) loses the increased sensitivity of the 5L (if it
detects more change) but (b) it stretches out utility values across
a larger range (the 3L range), and hence gains or losses are
larger and more in line with the 3L tariffs

16. 3L vs 5L in cancer - EuHEA 2018
Limitations
• If patients differed over time (pre- post-2015), result validity may
be affected
• Results are not necessarily generalisable to a non-cancer setting
• Patients were relatively ‘healthy’ on enrolment into the cohort

17. 3L vs 5L in cancer - EuHEA 2018
Conclusion
• Demonstrates the potential effects of switching between
instruments on policy recommendations resulting from cost-utility
analysis in an oncology setting
• 3L data generally suggest larger health gains from effective interventions
than 5L data, which will tend towards more attractive incremental cost-
effectiveness ratios (ICERs) for interventions that improve quality of life.
• Solutions?
• Being non-prescriptive in choice of instrument (lead to gaming rather than
scientific choice)
• Adjust the cost-effectiveness threshold for different instruments

18. 3L vs 5L in cancer - EuHEA 2018
Is it like Brexit? i.e. a UK (or English) problem
• Original UK 3L algorithm has a larger
range than other international algorithms
• 3L values cannot be replicated using more
contemporary gold-standard econometric
analysis
• The increased spread implicit in the UK 3L
algorithm will consequentially tend towards
having larger gaps between similar health
states
• Is it a descriptive system problem or a valuation problem?
• Different study …

19. 3L vs 5L in cancer - EuHEA 2018
Acknowledgements
• We would like to thank all the cancer patients who participated in the study. We
gratefully acknowledge the cooperation of the following Victorian institutions: Peter
MacCallum Cancer Center; The Andrew Love Cancer Centre, Barwon Health,
University Hospital Geelong; Oncology Trials Department, South West Regional
Cancer Centre; Royal Melbourne Hospital, Melbourne Health; Department of
Haematology & Oncology, Cabrini Institute, Cabrini Health; Centre for Health
Economics, Monash University; Department of Epidemiology and Preventative
Medicine, The Alfred Centre, Monash University.
• This study was supported by the EuroQol Foundation (EQ Project 20170410).
Cancer 2015 was funded by the Victorian Government through the Victorian Cancer
Agency Translational Research Program.
Funding

20. 3L vs 5L in cancer - EuHEA 2018
Thank you
To enquire about additional information and analyses, please contact Paula Lorgelly –
plorgelly@ohe.org
To keep up with the latest news and research, subscribe to our blog, OHE News
Follow us on Twitter @OHENews, LinkedIn and SlideShare
Office of Health Economics
Southside, 7th Floor
105 Victoria Street
London SW1E 6QT
United Kingdom
+44 20 7747 8850
www.ohe.org
OHE’s publications may be downloaded free of charge from our website