Quantifying health-state utilities: Discrete choice modeling and its relationship to fundamental measurement

Presenter: Paul Krabbe, Radboud UMCN

Abstract

Discrete choice models (DCM) are developed to establish the relative merit of subjective phenomena. DCM is believed to be the most accurate and general purpose tool currently available for making probabilistic predictions about human decision making behavior. Recently, research has started to apply DCM for quantifying stated preferences (utilities) for health states. The objective of this presentation is to present a brief overview of the relationship between DCMs and two related fundamental measurement models stemming from other science areas in the case of quantifying health states.

In the field of mathematical psychology and mathematics two different but interesting measurement models have been developed that are classified as fundamental measurement models: conjoint measurement and the Rasch model (i.e., cardinal or interval data). Conjoint measurement was developed by the mathematical psychologist Luce and statistician Tukey (1964). The significance of the theory of conjoint measurement lies in the fact that subjective properties can be quantified. Hence the quantification of such things as psychological attributes (e.g. attitudes, cognitive abilities and utility) is a logical possibility. Rasch (Danish mathematician and statistician) introduced his model for dichotomous data in 1966. In the Rasch model, the probability of a specified response is modeled as a function of item and person parameters. Therefore, Rasch models have a specific measurement property, invariance, a critical criterion for fundamental measurement. Invariance means that the comparison between two (or more) health states should be independent of the group of respondents that performed the comparisons; judgments among health states should also be independent of the set of health states being compared. Rasch pointed out that the principle of invariant comparison is characteristic of measurement in physics.

The response task in Rasch modeling is not a choice between two or more scenarios as is the case in DCM but comprises a series of monadic judgments about health states (yes/no, agree/disagree, worse/better). Therefore, at the moment the Rasch model is predominantly used in the area of health-related quality of life research for the measurement of abilities on specific health domains. However, in principle the basic operation of the Rasch response task can also be performed for overall health states. It seems that for health states instead of a representative sample of the general population judgments from a heterogeneous sample of patients are required to arrive at fundamental measurement. An example of such a patient-based judgment task will be presented. The basic operations of conjoint measurement appear to be included in the probabilistic DCM model, although formal checks whether the levels of the attributes in the health-state scenarios are satisfying the required conditions are absent.

To conclude, it seems that incorporating elements of the Rasch model into the DCM framework may provide a new and advanced measurement model with real fundamental measurement characteristics. Such a generalized measurement model should be an extension of the existing general DCM logistic model supplied by data derived from innovative and new response tasks.