Valuing paediatric preference-based measures: using a discrete choice experiment with duration to value the CHU-9D
- 1. Valuing a paediatric preference-based measure: the CHU-9D-NL Donna Rowen School of Health and Related Research (ScHARR) University of Sheffield
- 2. Project team • Brendan Mulhern (University of Technology Sydney, Australia) • Katherine Stevens (University of Sheffield, UK) • Erik Vermaire (TNO, Netherlands) • Acknowledgements: Richard Norman (Curtin University)
- 3. Introduction • Child valuation • Valuing the CHU-9D-NL in the Netherlands • Outline the CHU-9D • Normative decisions • Methodology • Results • Comparison to CHU-9D-UK value set
- 4. Child valuation • Whose values? • Adult valuation arguments often focus around whether values should be elicited from general population or patients • For children the argument is less around experience of the health state and more around the population – General population? Adolescents? • Which perspective? • Elicitation technique and mode of administration? • Comparability with adult values for consistency and combined use in HTA e.g. vaccinations
- 5. CHU-9D • CHU-9D is a paediatric preference-based measure of quality of life (Stevens, 2009;2010) • Developed using qualitative interviews with over 70 school children aged 7-11 in the UK • Dimensions and wording selected using the transcripts and Framework analysis • 9 dimensional self-completed measure • Translated into 6 languages including Dutch • Suitable for self-report in ages 7-17 years • Used in over 180 studies to date
- 6. CHU-9D Classification Dimension Wording Severity levels Worried I don’t feel worried today A little bit / a bit / quite / very Sad I don’t feel sad today A little bit / a bit / quite / very Pain I don’t have any pain today A little bit / a bit / quite a lot / a lot Tired I don’t feel tired today A little bit / a bit / quite / very Annoyed I don’t feel annoyed today A little bit / a bit / quite / very School work /homework I have no problems with my schoolwork / homework today A few problems / some problems / many problems / can’t do Sleep Last night I had no problems sleeping A few problems / some problems / many problems / can’t sleep Daily routine I have no problems with my daily routine today A few problems / some problems / many problems / can’t do Able to join in activities I can join in with any activities today Most / some / a few / no
- 7. Existing value sets • UK – adults using standard gamble (SG) (Stevens, 2012) • Australia – adolescents using best-worst scaling – adults using best-worst scaling (BWS) (both anchored using time trade-off) (Ratcliffe et al, 2012;2015;2016)
- 8. Whose values? Adult preferences • Tax payers • Understanding of tasks • Able to answer questions involving ‘dead’ • Do not necessarily reflect child or young adolescent preferences Child and adolescent preferences • Children and adolescents experience the health states • Adolescents have understanding of some tasks e.g. BWS, DCE • Children 7-11 unlikely to fully understand tasks • Are adolescent preference weights more appropriate for 7-11 year olds than adult preferences? • Unable to answer questions involving ‘dead’ so require adult (or young adult) data e.g. standard gamble or time trade-off to anchor the states on 1-0 full health-dead scale • Is this preferable to using only adult values?
- 9. Perspective? If asking adults, they could be asked to imagine: • The health state in the context of a 10 year old child • Which child matters • Will incorporate respondents views about children and child health (may think it is much worse for a child to be sick, may not want to sacrifice years of life for a child) • The health states for themselves as a child • Recall bias, also some of the concerns raised above • The health state for themselves • ‘Veil of ignorance’, value is not influenced by respondents views about children and child health (comparability) • If society values child health more, QALY weighting or deliberation could be used at decision level for HTA e.g. NICE
- 10. CHU-9D-NL valuation • Whose values? • Perspective? • Elicitation technique and mode of administration? • Influenced by choice of population and perspective
- 11. CHU-9D-NL valuation • Whose values? Adult general population sample • Perspective? Themselves (reworded school work/homework dimension to work/house work) • Elicitation technique and mode of administration? Online DCE with duration
- 12. The survey and sample • Participants recruited via existing online panel, paid via points from market research agency • Information sheet, informed consent • Sociodemographic questions • CHU-9D and EQ-5D-5L • 1 practice DCE plus 12 DCE questions
- 13. Selecting profiles • Profiles selected using Ngene software taking into account regression model specifications • 3 dimensions fixed across both profiles in a pair, built into design • Duration of 1, 4, 7, 10 years – successfully used previously for other surveys • Selected 204 choice sets and allocated each to one of 17 blocks of 12 for each survey version using a D-Optimal design • Choice sets randomly ordered within a block for each participant but dimension order fixed
- 14. Example question Health description A Health description B You live for 10 years with the following then you die: You live for 1 year with the following then you die: You feel a little bit worried You feel a little worried You feel a bit sad You feel very sad You have a bit of pain You don’t have any pain You feel quite tired You feel quite tired You feel quite annoyed You don’t feel annoyed You can’t do work/housework You have many problems with your work/housework You have a few problems sleeping You can’t sleep at all You can’t do your daily routine You have a few problems with your daily routine You can join in with any activities You can join in with any activities
- 15. Modelling DCE with duration data Model specification (Bansback et al, 2012): 𝜇𝑖𝑗 = 𝛼𝑖 + 𝛽1 𝑡𝑖𝑗 + 𝛽′2 𝐱 𝑖𝑗 𝑡𝑖𝑗 + 𝜀𝑖𝑗 𝜇𝑖𝑗 represents the utility of individual 𝑖 for profile j 𝑡𝑖𝑗 represents time 𝛽1 is the coefficient for duration in life years t 𝛽′2 represents the coefficients on the 36 interaction terms of duration and attribute levels • Anchored using the Marginal Rate of Substitution • Divide through by the duration coefficient: 𝛽2𝑖𝑗 𝛽1 • Conditional logit model with robust standard errors
- 16. The sample Sample n=1,276 % Netherlands n=16,979,120 % Male 49.8% 49.8% Age under 30 16.5 18.7 30-39 15.4 15.3 40-49 18.8 18.9 50-59 17.0 17.7 60+ 32.2 29.4 Employed 56.0 53.6 Married 63.8 62.3 EQ-5D-5L NL Mean (s.d.) 0.795 (0.230) 0.869 (0.170)
- 18. Utility decrements for UK and The Netherlands
- 19. Regression models First model NL DCE UK SG (OLS) Statistically significant 31 (out of 37) 30 (out of 36) Incorrect sign 2 0 Inconsistencies 4 14
- 20. Examples You feel a little bit worried You feel a little bit sad You have a little bit of pain You feel a little bit tired You feel a little bit annoyed You have a few problems with your work/housework You have a few problems sleeping You have a few problems with your daily routine You can join in with most activities You feel very worried You feel very sad You have a lot of pain You feel very tired You feel very annoyed You can’t do work/housework You can’t sleep at all You can’t do your daily routine You can join in with no activities State 111111111 NL = 0.788 UK = 0.679 State 444444444 NL = -0.568 UK = 0.326
- 21. Robustness • Models re-estimated excluding: • All responses less than 5 seconds • All responses over 10 minutes • Excludes 9.3% of responses • Slightly larger coefficients • Same problems with inconsistencies and incorrect signs • One exception that work levels 3 and 4 are consistent
- 22. Discussion • Valuation of CHU-9D-NL using online DCE with duration with adult general population sample feasible and generated sensible results • Large contrast in size of utility decrements to UK SG with adult general population – with more consistent coefficients • Problems of dimension framing and interpretation of “work/housework” rather than “school work/homework” • In the Netherlands income loss from being off work due to illness is minimal • Should child or adult preferences be used to value child health states? • What is the appropriate perspective? • Does the use of ‘informed’ adult values offer a solution?
- 23. Discussion • Which values are most appropriate for informing resource allocation decisions? • Complication of generating QALYs from birth or toddlers through to adulthood and beyond • For comparability reasons could argue for use of adult general population values elicited from own perspective • Utility values are not affected by additional factors such as views around child or child health • Arguably is the health state that is important not who experiences it or the cause • Potentially raises issue of QALY weights or different threshold
- 24. References • Bansback N, Brazier J, Tsuchiya A, Anis A. Using a discrete choice experiment to estimate health state utility values. J Health Econ. 2012;31(1):306-18. • Ratcliffe J, Flynn T, Terlich F, Brazier J, Stevens K, Sawyer M. Developing adolescent specific health state values for economic evaluation: an application of profile case best worst scaling to the Child Health Utility-9D. Pharmacoeconomics 2012; 30:713-27. • Ratcliffe J, Chen G, Stevens K, Bradley S, Couzner L, Brazier J, Sawyer M, Roberts R, Huynh E, Flynn T. Valuing Child Health Utility 9D Health States with Young Adults: Insights from A Time Trade Off Study. Applied Health Economics and Health Policy, 2015; 13:485-492 • Ratcliffe J, Huynh E, Stevens K, Brazier J, Sawyer M, Flynn, T. Nothing about us without us? A comparison of adolescent and adult health-state values for the child health utility-9D using profile case best-worst scaling. Health Economics, 2016; 25: 486-496 • Rowen D, Mulhern B, Stevens K, Vermaire E. Estimating a Dutch value set for the paediatric preference-based CHU-9D using a discrete choice experiment with duration. HEDS Discussion Paper 2017, University of Sheffield, available online. • Stevens, K J. Working With Children to Develop Dimensions for a Preference-Based, Generic, Paediatric Health-Related Quality-of-Life Measure. Qualitative Health Research. 2010; vol. 20: 340 - 351 • Stevens, K J. Developing a descriptive system for a new preference-based measure of health-related quality of life for children. Quality of Life Research. 2009; 18 (8): 1105-1113 • Stevens K. Valuation of the Child Health Utility 9D Index. Pharmacoeconomics 2012; 30:8: 729-747.