Valuing Health at the End of Life                Koonal Shah, OHE   Rachel Baker, Glasgow Caledonian University           ...
Research and Findings to Date            Koonal Shah     Office of Health Economics       OHE Lunchtime Seminar      26 Ma...
Study Team and Note on Funding• This research is a collaboration between Koonal Shah (Office of  Health Economics) and All...
NICE End of Life Criteria• Criteria that need to be satisfied for NICE’s supplementary end-  of-life policy to apply are c...
Overview of Project    Study 1: Exploratory study•   Aim: to pilot an approach to eliciting priority setting preferences• ...
Findings from Preliminary Studies• Elicitation approach found to be feasible• No consensus set of preferences• Majority wi...
DCE Study• DCEs (discrete choice experiments) elicit people’s preferences  based on their stated preferences in hypothetic...
Attributes and LevelsAttribute                                  Unit          LevelsLife expectancy without treatment     ...
Study Design• Forced choices (no ‘neither A nor B’ option)• Generic descriptions of patients, illnesses and treatments• St...
Web-Based Surveys                Pros                                    Cons• Can recruit a vey large sample         • No...
Background Characteristics                     #     %   gen pop %Total            3,969   100         100Gender  Male    ...
Background Characteristics (2)                                                                 #    %Household composition...
Results• Best fitting model included main effects plus three interactions:   • LE without treatment against LE gain       ...
Attribute / level                              Coefficient   p-value                                                  Inte...
Transforming into Predicted Probabilities   • Following the approach used by Green and Gerard* we calculated     the relat...
Estimated Utility Score and Predicted                     Probability of Choice for All Profiles  Rank    Rank –       LE ...
Levels of QALYs without Treatment /                      Gains Associated with All 110 Profiles        6        5        4...
Most and Least Preferred Profiles                       LE without   QOL without   LE gain   QOL gain (%)       QALYs QALY...
Subgroup Analysis• We defined a selection of respondent subgroups whose choices  may be expected to differ from those of t...
Categorising According to ‘Choice Strategy’                              % choices made                     Number (%) of ...
Extension Tasks• Extension tasks showed that including information about the  amount of time that patients have known abou...
Summary of Findings• Choices driven by size of health gain• Concern about the extent to which the patient is at the end of...
Caveats and Limitations• Small range of scenarios covered – all involve poor prognoses  (some people might consider 5 year...
To enquire about additional information and analyses, please contact Koonal Shah atkshah@ohe.orgTo keep up with the latest...
Institute for Applied Health Research                       andInstitute for Society and Social Justice Research          ...
MRC Methodology panelAre health gains for terminally ill patients more valuable? Measuringsocietal views on health care re...
Outline• Why this work is important• Strengths, limitations and questions:   – Study design   – Methods   – Findings/ conc...
Are equal sized health gains ‘worth’ the sameregardless of who benefits and in what ways?
Rawlins et al Brit J of Clinical Pharmacology 2010• “The Institute recognises that the public,  generally, places special ...
Study Design• Carefully considered, rigorous design   – Preliminary and pilot work• Choice based stated preference study  ...
Methods 1: Question Framing• Choice between two patients A and B• Described in terms of 4 attributes   – LE and QoL withou...
Methods 2: Informed, C onsidered Responses• Choice types and questions of dominance     – 13 Choice types (see table 4)   ...
Methods 2: Informed, Considered Responses• Some choices between a patient who is worse off and  gains more from treatment ...
Methods 2: Informed, Considered Responses• Evidence of deliberation and carefully considered  choices .. .in web based res...
Findings 1• Large respondent sample, lots of observations   – Any ‘representative’ sample is problematic• Reporting of ‘ra...
Findings 2• Main effects with 3 interactions   – Model fits better   – Table 6 is difficult to interpret…   – Instead of c...
Findings 3• Table 8 and figures 5 and 6 summarise the  untreated QALYs and QALY gains on probability of choice• Choice is ...
Consider adding info about QALY gain                      to full rank pred prob table 7?Rank      Rank     LE      QOL   ...
Findings 4: Extension Tasks• 8 DCE choices selected• Information about prior knowledge of disease added   – ?different res...
Overall•   Well conducted piece of research•   Raises questions about NICE end of life policy•   Quality of life and life ...
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Seminar on Studies about Valuing End-of-Life Health Care
Upcoming SlideShare
Loading in …5
×

Seminar on Studies about Valuing End-of-Life Health Care

1,731 views

Published on

OHE hosted a Lunchtime Seminar that examined both the approach and the results of research to date. OHE's Koonal Shah presented his research, which was critiqued by Dr Rachel Baker of Glasgow Caledonian University. Both presentations are included in this file.

Published in: Health & Medicine
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,731
On SlideShare
0
From Embeds
0
Number of Embeds
1,018
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Seminar on Studies about Valuing End-of-Life Health Care

  1. 1. Valuing Health at the End of Life Koonal Shah, OHE Rachel Baker, Glasgow Caledonian University OHE Lunchtime Seminar 26 March 2013 • London
  2. 2. Research and Findings to Date Koonal Shah Office of Health Economics OHE Lunchtime Seminar 26 March, 2013 • London
  3. 3. Study Team and Note on Funding• This research is a collaboration between Koonal Shah (Office of Health Economics) and Allan Wailoo, Aki Tsuchiya and Arne Risa Hola (all University of Sheffield)• The research was funded by the National Institute for Health and Clinical Excellence (NICE) through its Decision Support Unit (DSU)• The views, and any errors or omissions, expressed in this presentation are the authors’ only
  4. 4. NICE End of Life Criteria• Criteria that need to be satisfied for NICE’s supplementary end- of-life policy to apply are currently as follows. The treatment is indicated for patients with a short lifeC1 expectancy, normally less than 24 months There is sufficient evidence to indicate that the treatment offersC2 an extension to life, normally of at least an additional three months, compared to current NHS treatment The treatment is licensed or otherwise indicated, for small patientC3 populations
  5. 5. Overview of Project Study 1: Exploratory study• Aim: to pilot an approach to eliciting priority setting preferences• Aim: to explore the rationales underpinning people’s stated preferences• Small scale (n=21); convenience sample; face-to-face interviews Study 2: Preference study• Aim: to test whether there is public support for giving priority to end of life treatments• Aim: to validate the approach and worth of conducting a large scale study• Medium scale (n=50); representative sample; face-to-face interviews Study 3: Discrete choice experiment• Aim: to examine people’s preferences regarding end of life more robustly• Aim: to examine the extent to which people are willing to sacrifice overall health in order to give priority to end of life treatments• Large scale (n=3,969); representative sample; web-based survey
  6. 6. Findings from Preliminary Studies• Elicitation approach found to be feasible• No consensus set of preferences• Majority wished to give priority to the end-of-life patient, but a sizeable minority expressed the opposite preference• ‘No preference’ rarely expressed• Strong preference for treatments the improve quality of life• Preferences appear to be driven by how long patients have known about their illness (i.e. how long they have to ‘prepare for death’)• People are happy to prioritise based on characteristics of patients/disease/treatment when gains to all patients are equal in size … next step is to understand the extent to which they would sacrifice health gain to pursue equity objectives
  7. 7. DCE Study• DCEs (discrete choice experiments) elicit people’s preferences based on their stated preferences in hypothetical choices• Surveys comprise several ‘choice sets’, each containing competing alternative ‘profiles’ described using defined ‘attributes’ and a range of attribute ‘levels’• Respondents’ choices between these profiles are analysed to estimate the contribution of the attributes to overall utility
  8. 8. Attributes and LevelsAttribute Unit LevelsLife expectancy without treatment months 3, 12, 24, 36, 60Quality of life without treatment % 50, 100Life expectancy gain from treatment months 0, 1, 2, 3, 6, 12Quality of life gain from treatment % 0, 25, 50• Concept of ‘50% health’ was explained as follows: ‘Suppose there is a health state which involves some health problems. If patients tell us that being in this health state for two years is equally desirable as being in full health for one year, then we would describe someone in this health state as being in 50% health’.
  9. 9. Study Design• Forced choices (no ‘neither A nor B’ option)• Generic descriptions of patients, illnesses and treatments• Steps taken to avoid bias due to task order or possibility of respondents reverting to default choices• 10 standard DCE tasks, followed by two ‘extension tasks’ designed specifically to explore whether respondents’ choices are influenced by information about how long the patient has known about their illness
  10. 10. Web-Based Surveys Pros Cons• Can recruit a vey large sample • No guarantee that respondents have quickly and cheaply listened to or understood• Avoids interviewer bias instructions• Survey highly customisable – e.g. • Concerns about effort and randomisation procedures engagement• Quality control procedures can be • High level of drop out put into place • Limited debriefing opportunity• Any less likely to be representative • Concerns about representativeness than other modes of administration? of sample
  11. 11. Background Characteristics # % gen pop %Total 3,969 100 100Gender Male 1,942 49 49 Female 2,027 51 51Age 18-24 404 10 11 25-44 1,413 36 38 45-64 1,228 31 31 65+ 924 23 21Social grade a A 221 6 4 B 1,114 28 22 C1 1,150 29 29 C2 645 16 21 D 357 9 15 E 482 12 8
  12. 12. Background Characteristics (2) # %Household composition With children 963 24 Without children 3,006 76Education No education beyond minimum school leaving age 889 22 Education beyond minimum school leaving age; no degree 1,244 31 Education beyond minimum school leaving age; degree 1,836 46Self-reported general health level Very good 1,008 25 Good 1,958 49 Fair 770 19 Poor 210 5 Very poor 23 1Experience of close friends of family with terminal illness Yes 2,689 68 No 1,197 30 Question skipped by respondent 83 2
  13. 13. Results• Best fitting model included main effects plus three interactions: • LE without treatment against LE gain • Rationale: small gains in life expectancy may be increasingly important when life expectancy without treatment is short • LE without treatment against QOL gain • Rationale: whether a quality of life improvement or a gain in life expectancy is preferred may depend on life expectancy without treatment • LE gain against QOL gain • Rationale: the important of a gain in life expectancy may depend on whether it is accompanied by a quality of life improvement
  14. 14. Attribute / level Coefficient p-value Interaction: LE without treatment # LE gain 12 months # 1 months -0.1715 0.15 12 months # 2 months -0.4220 0.00 12 months # 3 months -0.1633 0.18 12 months # 6 months -0.7294 0.00 12 months # 12 months -0.6039 0.00Attribute / level Coefficient p-value 24 months # 1 months -1.1308 0.00LE without treatment 24 months # 2 months -1.0782 0.00 24 months # 3 months -0.8614 0.00 3 months [baseline] - - 24 months # 6 months -1.2413 0.00 12 months 0.1755 0.12 24 months # 12 months -1.2601 0.00 24 months 0.9307 0.00 36 months # 1 months -0.7280 0.00 36 months 0.7841 0.00 36 months # 2 months -1.0428 0.00 60 months 1.2625 0.00 36 months # 3 months -1.2252 0.00QOL without treatment 36 months # 6 months -1.6695 0.00 36 months # 12 months -1.3963 0.00 50% [baseline] - - 60 months # 1 months -1.3159 0.00 100% 0.6730 0.00 60 months # 2 months -1.4933 0.00LE gain 60 months # 3 months -1.2558 0.00 0 months [baseline] - - 60 months # 6 months -2.0434 0.00 1 month 0.1855 0.08 60 months # 12 months -1.7114 0.00 2 months 0.8517 0.00 Interaction: LE without treatment # QOL gain 12 months # 25% 0.4562 0.00 3 months 1.0855 0.00 12 months # 50% 0.2139 0.00 6 months 2.0433 0.00 24 months # 25% 0.2734 0.00 12 months 2.9381 0.00 24 months # 50% 0.4123 0.00QOL gain 36 months # 25% 0.8457 0.00 0% [baseline] - - 36 months # 50% 0.7374 0.00 60 months # 25% 0.5379 0.00 25% 0.0632 0.47 60 months # 50% 0.6676 0.00 50% 1.0212 0.00 Interaction: LE gain # QOL gain 1 months # 25% 0.7649 0.00 1 months # 50% 0.5254 0.00 2 months # 25% 0.3197 0.00 2 months # 50% 0.3543 0.00 3 months # 25% 0.6321 0.00 3 months # 50% 0.3163 0.00 6 months # 25% 0.6661 0.00 6 months # 50% 0.2744 0.00 12 months # 25% 0.3466 0.00 12 months # 50% [baseline] - -
  15. 15. Transforming into Predicted Probabilities • Following the approach used by Green and Gerard* we calculated the relative predicted probabilities for all of the 110 profiles • This allows us to compare the profiles that are likely to be most preferred overall with those that are likely to be least preferred overall • The predicted probability of alternative i being chosen from the complete set of alternatives (j=1,…,J) is given by: 𝑃𝑃𝑛𝑛𝑛𝑛 = 𝑗𝑗 = 1, … , J 𝑒𝑒 𝑉𝑉 𝑛𝑛𝑛𝑛 ∑J 𝑉𝑉 𝑗𝑗=1 𝑒𝑒 𝑛𝑛𝑛𝑛* Green, C. and Gerard, K. (2009) Exploring the social value of health care interventions: A statedpreference discrete choice experiment. Health Economics. 18(8), 951-976.
  16. 16. Estimated Utility Score and Predicted Probability of Choice for All Profiles Rank Rank – LE without QOL without LE gain QOL gain Utility Prob. Cumul. - best main treatment treatment (%) (mths) (%) Prob.fitting effects (mths)model model 1 1 60 50 12 50 4.17809 0.0155 0.0155 2 2 36 50 12 50 4.08461 0.0154 0.0309 3 3 24 50 12 50 4.04235 0.0153 0.0462 4 5 3 50 12 50 3.95938 0.0152 0.0614 5 4 12 50 12 50 3.74493 0.0148 0.0762 6 20 3 100 12 0 3.61116 0.0145 0.0908 - - - - - - - - - 105 107 36 50 1 0 0.24171 0.0029 0.9870 106 109 12 50 1 0 0.18955 0.0028 0.9898 107 110 3 50 1 0 0.18553 0.0028 0.9926 108 104 60 50 1 0 0.13213 0.0026 0.9952 109 94 3 50 0 25 0.06320 0.0025 0.9977 110 108 24 50 1 0 -0.01452 0.0023 1.0000
  17. 17. Levels of QALYs without Treatment / Gains Associated with All 110 Profiles 6 5 4 3QALYs 2 1 0 0.0023 0.0040 0.0055 0.0062 0.0072 0.0085 0.0100 0.0112 0.0120 0.0130 0.0140 -1 Standardised predicted probability of being chosen QALY without QALY gain Linear (QALY without) Linear (QALY gain)
  18. 18. Most and Least Preferred Profiles LE without QOL without LE gain QOL gain (%) QALYs QALYs gained treatment treatment (mths) without (mths) (%) treatment10 most preferred 27 55 11 38 1.14 1.7655 most preferred 27 57 7 31 1.27 1.2255 least preferred 27 65 2 10 1.49 0.2910 least preferred 28 50 1 3 1.18 0.06
  19. 19. Subgroup Analysis• We defined a selection of respondent subgroups whose choices may be expected to differ from those of the rest of the sample • Respondents with experience of close friends or family with terminal illness • Respondents with responsibility for children • Respondents who voluntarily left open-ended comments • Respondent who completed the survey unusually quickly• We found no substantial differences between the results for any of these subgroups and those for the full sample
  20. 20. Categorising According to ‘Choice Strategy’ % choices made Number (%) of respondents who…Choice strategy according to this never followed this sometimes followed always followed this strategy strategy this strategy strategyChoose patient with largerQALY gain 0.75 1 (0.0%) 3,530 (88.9%) 438 (11.0%)Choose patient with largerLE gain 0.69 20 (0.5%) 3,405 (85.8%) 544 (13.7%)Choose patient with fewerQALYs without treatment 0.47 182 (4.6%) 3,701 (93.2%) 86 (2.2%)Choose patient with lessLE without treatment 0.45 355 (8.9%) 3,434 (86.5%) 180 (4.5%) • Multinomial logit regressions used to identify driving factor(s) behind respondents’ membership of the ‘always / never choose patient with fewer QALYs without treatment’ subgroup • Marginal effects of age and health satisfaction were found to be statistically significant, but both are small in practical terms • As age increases, the probability of always choosing the patient with fewer QALYs without treatment decreases, but even a 30-year increase in age would not be sufficient for a 1% decrease in this probability
  21. 21. Extension Tasks• Extension tasks showed that including information about the amount of time that patients have known about their prognosis has a clear impact on preferences• Holding everything else constant, respondents are less likely to choose to treat a patient if that patient has known about their illness for two years than if they have only just found out about it• Caveat: focusing effect may exaggerate importance
  22. 22. Summary of Findings• Choices driven by size of health gain• Concern about the extent to which the patient is at the end of life appears to have a negligible effect• Overall view seems to be that giving higher priority to those who are worse off is desirable, but only if the gains from treatment are substantial• No evidence of public support for giving higher priority to end-of- life treatments than to other types of treatments if the health gains offered by the treatments being ‘de-prioritised’ are larger than those offered by the end-of-life treatments
  23. 23. Caveats and Limitations• Small range of scenarios covered – all involve poor prognoses (some people might consider 5 years to be ‘end of life’)• Does not necessarily refute evidence elsewhere in the literature that people wish to pursue equity concerns• Great deal of preference heterogeneity• Limited opportunities for feedback and debriefing – cannot know for certain the extent to which the choice data truly reflect respondents’ beliefs and preferences (or whether there were adopting heuristics)• Framing effects clearly exist in stated preference studies
  24. 24. To enquire about additional information and analyses, please contact Koonal Shah atkshah@ohe.orgTo 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 (OHE)Southside, 7th Floor105 Victoria StreetLondon SW1E 6QTUnited Kingdom+44 20 7747 8850www.ohe.orgOHE’s publications may be downloaded free of charge for registered users of its website.©2013 OHE
  25. 25. Institute for Applied Health Research andInstitute for Society and Social Justice Research Valuing health at the end of life Shah et al DiscussionRachel BakerReader in Health Economicsrachel.baker@gcu.ac.ukYunus Centre for Social Business & Health
  26. 26. MRC Methodology panelAre health gains for terminally ill patients more valuable? Measuringsocietal views on health care resource allocationRachel Baker, Neil McHugh, Helen Mason, Cam Donaldson,Laura Williamson, Jon Godwin, Marissa Collins (GCU)Job van Exel (Erasmus, Rotterdam)Cathy Hutchinson (Beatson Cancer Centre, NHS Greater Glasgow &Clyde)
  27. 27. Outline• Why this work is important• Strengths, limitations and questions: – Study design – Methods – Findings/ conclusions• Future research… – MRC end of life Q methodology study
  28. 28. Are equal sized health gains ‘worth’ the sameregardless of who benefits and in what ways?
  29. 29. Rawlins et al Brit J of Clinical Pharmacology 2010• “The Institute recognises that the public, generally, places special value on treatments that prolong life – even for a few months – at the end of life, as long as that extension of life is of reasonable quality (at least pain-free if not disability-free). NICE has therefore provided its advisory bodies with supplementary advice about the circumstances under which they should consider advising, as cost-effective, treatments costing >£30,000 per QALY.” p 348
  30. 30. Study Design• Carefully considered, rigorous design – Preliminary and pilot work• Choice based stated preference study – Ordering effects and other biases controlled – Questions blocked by choice type• Web-based questionnaire – Diagrams and text explanation – Pilot tested and soft-launch
  31. 31. Methods 1: Question Framing• Choice between two patients A and B• Described in terms of 4 attributes – LE and QoL without treatment – LE and/or QoL gains with treatment• Individuals rather than groups of patients• QALY gain (green area) – How is QoL gain treated/ interpreted?• Indifference option (either not neither)
  32. 32. Methods 2: Informed, C onsidered Responses• Choice types and questions of dominance – 13 Choice types (see table 4) – Both patients have same LE and QoL; without treatment one patient gains more LE and QoL (11)• 10% respondents failed the dominance test. – Simple error? – Plausible rationale?• Excluding them from the analysis did not make any difference
  33. 33. Methods 2: Informed, Considered Responses• Some choices between a patient who is worse off and gains more from treatment and a patient who is better off and gains less• ?Not strictly dominated? QALY maximising choice and concern for severity are the same• 40% respondents (or in 30% of choices) chose the patient who was better off and gained less• Why?• Qualitative research/ cognitive interviewing
  34. 34. Methods 2: Informed, Considered Responses• Evidence of deliberation and carefully considered choices .. .in web based research – Lots of typed comments/ explanations? – Taking time over the survey• Speedsters! – Problem of web-based surveys – question of cut off... – < 3 mins for intro, 12 DCE questions and demographics – Quickest pilot respondent, employed/educated people with interviewer present, 6 minutes
  35. 35. Findings 1• Large respondent sample, lots of observations – Any ‘representative’ sample is problematic• Reporting of ‘raw’ data (and choice types) as well as modelling helpful – Table 4 (add majority choice for clarity?) – Main effects model (table 5) shows increasing value placed on bigger gains and – Increasing value placed on patients with better health without treatment (odd?)
  36. 36. Findings 2• Main effects with 3 interactions – Model fits better – Table 6 is difficult to interpret… – Instead of coefficients of attribute levels, Table 7: 110 profiles ranked according to probability of choice – ? Including interactions seem to take care of ‘oddness’? And untreated profile has little effect on choice (but 40% of ‘those choice types’ are still odd?) – Choices driven by QoL and LE gains
  37. 37. Findings 3• Table 8 and figures 5 and 6 summarise the untreated QALYs and QALY gains on probability of choice• Choice is driven by QALY gains and not untreated profile – Add to table 7 for all 110? – QALY gains relatively small? – Very few levels on QoL – Replication with different attribute levels?• Similar to DCE findings from SVQ study – Although modelled differently
  38. 38. Consider adding info about QALY gain to full rank pred prob table 7?Rank Rank LE QOL LE gain QOL gain Utility Prob. Cumul.- best – with/ without (mths) (%) Prob.fitting main t treatmemodel effect treat nt (%) QALY gain s (mths mode ) l (5*.5)+(1*1)1 1 60 50 12 50 =3.5 4.17809 0.0155 0.01552 2 36 50 12 50 2 4.08461 0.0154 0.03093 3 24 50 12 50 2 4.04235 0.0153 0.04624 5 3 50 12 50 1.125 3.95938 0.0152 0.06145 4 12 50 12 50 1.5 3.74493 0.0148 0.07626 20 3 100 12 0 1 3.61116 0.0145 0.0908- - - - - - - - -105 107 36 50 1 0 0.04 0.24171 0.0029 0.9870106 109 12 50 1 0 0.04 0.18955 0.0028 0.9898107 110 3 50 1 0 0.04 0.18553 0.0028 0.9926108 104 60 50 1 0 0.04 0.13213 0.0026 0.9952109 94 3 50 0 25 0.06 0.06320 0.0025 0.9977
  39. 39. Findings 4: Extension Tasks• 8 DCE choices selected• Information about prior knowledge of disease added – ?different respondents?• Responses to the extension task questions, compared with DCE responses suggest that time since diagnosis is important – We found the same in qualitative work (although not sure how important relative to other things)• Indifference option (either, not neither) – Might have helped with issue of focus and extension questions
  40. 40. Overall• Well conducted piece of research• Raises questions about NICE end of life policy• Quality of life and life extension are most important• Replication/ future research – Stretch the attributes over a wider set of levels • Esp LE without treatment • Qol levels? – Draw comparisons against patients who are less severely ill – Cognitive interviewing/ qualitative work and methods to understand rationale for ‘odd choices’

×