1 / 12

Response mapping to the EQ-5D: methods and comparative performance

Response mapping to the EQ-5D: methods and comparative performance. Oliver Rivero-Arias, Alastair Gray and Helen Dakin iHEA organised session 9 th July 2013 helen.dakin @dph.ox.ac.uk. Introduction to response mapping.

spencer
Télécharger la présentation

Response mapping to the EQ-5D: methods and comparative performance

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Response mapping to the EQ-5D: methods and comparative performance Oliver Rivero-Arias, Alastair Gray and Helen Dakin iHEAorganised session 9th July 2013 helen.dakin@dph.ox.ac.uk

  2. Introduction to response mapping • In response mapping, 5 categorical models (e.g. mlogit) predict the probability or odds that a participant is at level j on each EQ-5D domain • Combine predicted responses with tariff to estimate utilities • Advantages: • Response mapping algorithm can be used with any national EQ-5D tariff • Provides data on the distribution of patients across EQ-5D states • Gives insights into the nature of the relationship between instruments • May give better utility predictions by reflecting the distribution better • Disadvantages: • Models are more complicated to estimate: larger sample size needed? • More complicated to estimate utilities from predicted probabilities

  3. 3 methods to calculate EQ-5D utilities from predicted probabilities Pain/discomfort Self care Anxiety/depression Mobility Usual activities • Highest probability: Assume patient is at the level with the highest predicted probability • Monte Carlo: Randomly assign patients to one level • Expected value: Multiply predicted the probabilities with tariff values to obtain expected utility Highest prob: 1 2 1 -.2*.104 -.1*.214 2 1 1 -.4*.036 -.32*.094 2 3 2 -.35*.123 -.3*.386 2 1 1 -.32*.071 -.12*.236 2 1 1 =0.883 =0.760 =0.312 =0.305 1-.3*.069 -.17*.314 Monte Carlo 1: Expected value: 2: -.17*.1*.32*.3*.12*.269 -(1-(1-.53)*(1-.7)*(1-.27)*(1-.35)*(1-.66)*.081

  4. Pros and cons of methods to calculate EQ-5D utilities • Highest probability • Underestimates % of patients in rare health states (e.g. level 3)  Overestimates predicted utility • Monte Carlo (MC) • ≥1000 draws normally needed • Some studies use only one draw random variability • Expected value (EV) • Equivalent to Monte Carlo with infinite draws • Gives exact result instantly with one equation

  5. Aims • To review how response mapping to EQ-5D has been used to date • To compare the performance of response mapping with other methods

  6. Methods • A systematic review was conducted to identify studies using response mapping to predict EQ-5D responses from responses/scores on other QoL instruments • Included published and available unpublished studies • Searched Medline, Centre for Reviews and Dissemination (CRD), the Health Economists’ Study Group (HESG) website and HERC database of mapping studies (http://www.herc.ox.ac.uk/downloads/mappingdatabase) • Extracted data on: • Source instrument, models estimated • How prediction accuracy varied between models • Methods used to calculate utilities from predicted probabilities • Highest probability, expected value (EV) or Monte Carlo (MC)

  7. Characteristics of studies identified • 21 studies identified • Source instrument: SF-12 in 4 studies; EQ-5D-5L in 1 & disease-specific in 16 • 75% (6/8) of studies found predictions errors from both direct & response mapping higher for patients with utilities <0.5 than those with good health

  8. Modelling methods used • OLS was most common direct mapping method • Multinomial logit was most common response mapping model Direct mapping models Response mapping models Bayesian networks: 2 Mlogit: 14 CLAD: 7 11 2 Other: 5 3 2 2 1 3 2 9 Oprobit or generalised oprobit: 2 2 1 2 Ologit: 6 1 2-part: 5 OLS or GLS: 20 Cross-tabulation: 1

  9. Methods for estimating utilities • Monte Carlo, expected value and highest probability methods all commonly used to estimate utilities • 3 studies compared different methods Expected value: 9 Monte Carlo (n=1 or 11): 7 6 1 6 1 1 2 Highest probability: 5 Monte Carlo (n>100): 3

  10. Relative prediction accuracy • 55% of studies using EV or ≥100 Monte Carlo found response mapping gave similar or more accurate prediction errors • V.s 17% of those using highest probability or ≤11 Monte Carlo draws Predictions from response mapping are: Highest probability or ≤11 Monte Carlo Expected value or ≥100 Monte Carlo

  11. Conclusions • Response mapping and direct mapping have similar prediction errors when EV or MC (n>100) are used to calculate utilities • Highest probability method and n=1 Monte Carlo should be avoided • All mapping models perform poorly in patients with poor QoL • Response mapping has additional advantages • Insights into relationships between instruments • Gives domain-level predictions • Can be used with any EQ-5D valuation tariff • Distribution of Excel or Stata commands to estimate predictions can simplify the process for users • mrs2eq and oks2eq commands accepted for publication in Stata Journal

  12. Acknowledgements • Many thanks to Jason Madan, Kamran Kahn, Aki Tsuchiya and Richard Eldin for allowing us to cite their unpublished work

More Related