Developing a preference-based method for mapping between (preference-based) measures of health and quality of life

The NHS has to make important decisions about how to spend its limited resources. One way to do this is to compare the value for money from different health services. One method for helping to make this decision (though there will be other factors that should influence the decision) is to compare services in terms of their cost and benefits measured in terms of health; this is done by calculating their cost per ?quality adjusted life year (QALY)? and to select those interventions with the lowest cost per QALY.

QALYs are calculated by giving a value from zero to one to a person?s state of health (where one is for full health and zero for states regarded as bad as being dead) and then multiplying with its duration. Currently, there is more than one method of assigning values to different health states, and the calculation of the cost per QALY of an intervention might be affected by which method is used. Different studies use different methods, because no one measure can capture all the dimensions of interest across every care group and medical condition. This is the problem the proposal addresses: to solve the current impossibility of comparing between studies that cover different diseases and different patient groups. This proposal seeks to address this problem by asking members of the general public to simply place a set health states (described using different instruments) in rank order, starting with the best imaginable state at the top. This information will be used to allow comparisons to be made between results obtained using different instruments. This will be very useful to researchers combining evidence from different studies and to policy makers using a cost per QALYs to help make decisions.

Recent years have seen the proliferation of preference-based measures of health and quality of life for use in economic evaluation, including a number of different generic measures, measures specific to certain populations (such as older people), and those specific to various medical conditions. This is creating a major problem for policy makers wishing to make cross-programme comparisons since there is no means of comparing scores generated by each instrument. Previous attempts to map between instruments have tended to focus on the descriptive systems, such as estimating statistical relationships between a given health state in one instrument explained in terms of the components of another instrument. However, this is not possible for some instruments if the two instruments are not meant to be used alongside each other (such as those designed for adults and those for children), and for others makes little sense where the descriptive systems are very different. What is needed is a means of relating one instrument to another using a common metric.

This is a proposal to develop a preference-based method of mapping between preference-based measures. It will be done in the first instance for five instruments, but it could be extended to other instruments. It involves asking a representative sample of the general population to value a set of health states drawn from the five instruments. Respondents would be asked to undertake four ranking exercises. The entire sample will rank a subset of health states taken from a pool, consisting of 15 health states from each instrument. These data will be analysed econometrically using rank ordered logit models, to estimate values for the 75 health states across the five instruments. Mapping functions will be estimated within each instrument between the 15 health state values estimated using the rank data and their values using current scoring procedures.

This is an exploratory study of a novel approach that could have important applications in health economics and decision science. Each instrument will have a preference-based mapping function that provides a bridge for studies seeking to synthesise data (including decision analytic models) using different instruments. This approach will be tested in terms of feasibility, completeness and consistency of responses, assessments of difficulty and understanding and cognitive interviewing a sub-sample of respondents.