The usual approach to developing a EQ-5D value set is to collect preferences for a subset of health states from a population sample using established techniques such as discrete choice experiments or time-trade-off. Multiple functional forms which describe the relationship between the selected health states and elicited preferences to these states are hypothesized and examined. A best-performed functional form is then used to develop a value set. The procedure for identifying the functional form of the responses is typically undertaken by specifying a number of hypothesized functional forms and comparing measures of model fit such as mean absolute error and mean squared error for each form. The validity of the resulting value set depends upon strict assumptions of correct model specification, yet the true functional form is unknown.

Non-parametric local-constant regression is an alternative method for estimating the relationship between health states and elicited utilities which does not require the analyst to specify any functional form prior to estimation.

When using the non-parametric method to score preference-based instruments, there is no functional form as we used to see with parametric methods. Therefore, for illustration purpose, we have created a 3D display to illustrate the concept of the non-parametric approach using R in Shiny. A searchable EQ-5D-5L value set with a graphic can be accessed by clicking here