Modelling health state preference data

The Problem

There are currently a number of generic and condition specific preference-based measures (such as the EQ-5D, SF-6D and AQL-5D). A key problem for these measures has been the large number of unique health states that they define and the consequent need to model health state values from a valuation of a subset of possible states. Health state values present a significant challenge for conventional statistical modelling procedures due to their nature, namely: skewed, truncated, non-continuous and hierarchical (Brazier et al, 2002).

Attempts to statistically model these data have met with some success in the EQ-5D and SF-6D (Dolan, 1997 and Brazier et al, 2002). However, for both instruments there are concerns with the size of the prediction errors, and for the SF-6D there is a problem of non-monotonicity (where some better states are assigned a lower value than worse states) and an apparent systematic pattern in the prediction errors (involving over prediction of the value of the poor health states and under prediction of the value of good health states).

Research at CHEBS

CHEBS has been looking at an alternative approach to modelling health state values using a (nonparametric) Bayesian method and comparing this to conventional regression based methods. The first part of this work was to develop computer code in Matlab for applying this method and applying it to UK SF-6D valuation data (Kharroubi et al, 2005). The superior performance of this model compared to conventional regression modelling has been reported in another publication (Kharroubi et al, 2007a). An Excel spreadsheet will be available on the web very soon that will produce estimated utilities and standard deviations for health states described by the SF-6D and SF-36 instruments.

This method has been found to be useful in exploring the role of respondent background characteristics in explaining variation in health state values (Kharroubi et al, 2007). The same model has also been applied to other descriptive systems including EQ-5D and condition specific systems including the asthma instrument, AQL-5D. It has also been applied to an SF-6D data from Japan and we are currently examining its application to a similar data set from Hong Kong.

Ongoing Research at CHEBS

  • Applications to other condition specific measures.
  • Estimating differences by country in order to use the results from one country to inform the design of another. This includes a paper in preparation comparing UK and US valuation of the EQ-5D.
  • Application of Bayesian methods to analysing ordinal data, including pair wise data. Currently the subject of a CASE studentship.

Future Research at CHEBS

  • Further applications of Bayesian methods to ordinal data – possible collaboration with Monica Hernandez, Department of Economics, University of Sheffield.
  • Extending the cross country work to valuation data sets from other countries and for other systems (e.g. Hong Kong/Australia/ Brazil/Portugal/Japan) for SF-6D.
  • Exploring the importance of interactions in models.
  • Examining the use of uncertainty around health state predictions.
  • Conducting meta analyses of published health state preference data.
  • Extrapolation of published values to study populations
  • Developing Bayesian approaches to mapping between instruments and the correct estimation of uncertainty.
  • Designing valuation surveys.

Papers in submission/preparation

  • Brazier, J., Fukuhara, S., Roberts, J., Kharroubi, S. A. , O'Hagan, A, and Ikeda, S.. Estimating a preference-based index from the SF-36: the case of Japan. In preparation

Papers in press

  • Kharroubi S, O’Hagan A, Brazier J. A comparison of United States and United Kingdom EQ-5D health state valuations using a nonparametric Bayesian method. To appear in Statistics in Medicine.

Papers already published

  • Kharroubi SA, O’Hagan A, Brazier JE (2010). A comparison of United States and United Kingdom EQ-5D health states valuations using a nonparametric Bayesian method. Published online in Statistics in Medicine.
  • Kharroubi SA, Brazier JE, O’Hagan A (2007b). Modelling covariates for the SF-6D standard gamble health state preference data using a nonparametric Bayesian method. Social Science and Medicine 64:1242-52.
  • Kharroubi SA, Brazier JE, Roberts J, O’Hagan A. (2007a) Modelling SF-6D health state preference data using a nonparametric Bayesian method. Journal of Health Economics 26:597-612.
  • McCabe C, Brazier J, Gilks P, Tsuchiya A, Roberts J, O'Hagan A, Stevens K (2006). Using rank data to estimate health state utility models. Journal of Health Economics 25:418-31.
  • Kharroubi SA, O’Hagan A, Brazier JE (2005). Estimating utilities from individual health preference data: a nonparametric Bayesian method. Applied Statistics 54:879-95.