Download - E2 feng xie 2016 apr12
Transforming latent utilities to health utilities:
East doesn’t meet WestFeng Xie
Department of Clinical Epidemiology and BiostatisticsMcMaster University
Acknowledgements Coauthors: Eleanor Pullenayegum, Simon Pickard, Juan
Manuel Ramos Goni, Min-Woo Jo, Ataru Igarashi We thank Drs. Ben Van Hout, Elly Stolk, Nan Luo, Juntana
Pattanaphesaj, Juan Manuel Ramos Goñi, Min-Woo Jo, and Ataru Igarashi for sharing their data
This project was sponsored by a fast-track research grant from the EuroQol Research Foundation (#2013180)
Drs. Feng Xie is funded by the Canadian Institutes for Health Research New Investigator Award (MSH #122801). Dr. Feng Xie is also supported by McMaster University and St. Joseph’s Healthcare Hamilton.
None of the sponsors had any involvement in the design and conduct of the study, collection, analysis, and interpretation of the data, preparation, review and approval of the work.
EQ-5D-5L Valuation Study An international initiative by the EuroQol
Group Standardized protocol – EuroQol
Valuation Technology (EQ-VT) Canada, Spain, UK, the Netherlands,
Japan, Thailand, Korea, and China More countries…
Discrete choice experiment (DCE)
DCE vs TTO
Full healthDead State 1State 2
1.00.0
health utility
DCElatentutility
Cognitive challenge Online vs face-to-face interview Health utilities from TTO vs latent
utilities from DCE
The motivation Feasibility issues in conducting
interviews with a national representative sample in geographically-spread countries or those with resource constraint
DCE could be a practical alternative if an existing transforming function can be used
Hypothesis and objective The relationship between different
methods in eliciting health preference may be similar across countries given the same underlying construct being elicited
To compare generic functions with country-specific functions in transforming latent utilities to health utilities
The data sets Valuation study data from the 8 countries TTO –derived health utilities for 86 health
states 196 state pairs using DCE Each participant was asked to value 10 health
states using TTO and 7 pairs of states using DCE
Transforming L to U
1• Conditional logit model to derive latent utilities using DCE data
2• Calculating mean TTO-derived health utility for each of 86 states
3• fractional polynomial models to transform L to U (e.g. E(U|L)=β0 + β1La)
4
• Calculating mean absolute error (MAE) between predicted and observed health utility for each state without including the data from that state in modeling
Criteria for MAEs The standard deviations (SDs) of the
MAEs from 18 EQ-5D (3 level) TTO-based valuation studies
≤1 SD (0.02): acceptable; 1 SD<~<2 SDs (0.02 to 0.04): applied
with caution ≥2 SDs (0.04): unacceptable.
Study and respondent characteristics
Canada U.K. Spain Netherlands China Thailand Korea Japan
No of respondents*
1209 1221 1000 983 1299 1216 1080 1026
No of interviewers 11 60 33 19 21 6 27 31
Use of commercial survey company
N Y Y N N N Y Y
Age, years, mean±SD
47.5± 17.4 51.0 ± 17.9 43.8 ± 17.3
47.2 ± 16.8 42.3 ± 16.2
43.5 ± 15.1
45.0 ± 14.3
44.9 ± 14.9
Female, n(%) 667 (55.0%)
710 (58,2%)
525 (52.5%)
507 (51.6%)
649 (50.0%)
630 (51.8%)
548 (50.7%)
511 (49.8%)
EQ-VAS, mean±SD 82.3 ± 14.2
78.6 ± 19.0
82.3 ± 14.5
80.5 ± 14.8 86.0 ± 11.4
83.1 ± 11.9
83.0 ± 10.0
84.9 ± 11.2
-10 -8 -6 -4 -2 0
-0.2
0.0
0.2
0.4
0.6
0.8
Latent Utility (SD units)
Obs
erve
d TT
O m
ean
CanadaUnited KingdomChinaNetherlandsSpainThailandKoreaJapan
-10 -8 -6 -4 -2 0
-0.2
0.0
0.2
0.4
0.6
0.8
Latent Utility (SD units)
Obs
erve
d TT
O m
ean
CanadaUnited KingdomChinaNetherlandsSpainThailandKoreaJapan
Country-specific functions
Regional functions
Global function
The findings The differences were larger in the four
eastern countries than those in the four western countries
A global generic transforming function was associated with large increase in prediction errors
A generic function for western countries may work
Discussion DCE could be used as the sole technique
in western countries where using TTO is not feasible
Provincial value set could be derived using the national transforming function applied to provincial DCE data
Trade-off between prediction precision for health state utilities and amount of research resources to spend must be made