HEDS Discussion Paper 09-15: Developing preference-based health measures: using Rasch analysis to generate health state values

Background/aims: Condition specific measures may not always have independent items, and existing techniques of developing health state values from these measures are inappropriate when items are not independent. This study develops methods for deriving and valuing health states for a preference-based measure. Methods: Three key stages are presented: Rasch analysis is used to develop a health state classification system and identify a set of health states for valuation. A valuation survey of the health states using time-trade-off (TTO) methods is conducted to elicit health state values. Finally, regression models are applied to map the relationship between mean TTO values and Rasch logit values. The model is then used to estimate health state values for all possible health states. Methods are illustrated using the Flushing Symptoms Questionnaire (FSQ). Results: Rasch models were fitted to 1270 responders to the FSQ and a series of 16 health states identified for the valuation exercise. An ordinary least squares model best described the relationship between mean TTO values and Rasch logit values. (R2 = 0.958; Root mean square error = 0.042). Conclusions: We have shown how the valuation of health states can be mapped onto the Rasch scale in order to value all states defined by the FSQ. This should significantly enhance work in this field