Gray, L.A. orcid.org/0000-0001-6365-7710, Hernandez, M. orcid.org/0000-0003-4474-5883 and Wailoo, A. orcid.org/0000-0002-9324-1617 (2017) Development of methods for the mapping of utilities using mixture models: An application to asthma. Working Paper. ScHARR HEDS Discussion Papers, 17.03 . School of Health and Related Research , University of Sheffield.
Abstract
Objectives: To develop methods for mapping to preference-based measures using mixture model approaches. These methods are compared to map from the Asthma Quality of Life Questionnaire (AQLQ) to both EQ5D-5L and HUI-3 as the target health utility measures in an international dataset.
Methods: Data from 856 patients with asthma collected as part of the Multi-Instrument Comparison (MIC) international project were used. Adjusted limited dependent variable mixture models (ALDVMMs) and beta-regression based mixture models were estimated. Optional inclusion of the gap between full health and the next value, and a mass point at the next feasible value were explored. Response-mapping could not be implemented due to missing data.
Results: In all cases, model specifications which formally modelled the gap between full health and the next value were an improvement on those which did not. Mapping to HUI3 required more components in the mixture models than mapping to EQ5D-5L due to its uneven distribution. The optimal beta-based mixture models mapping to HUI3 included a probability mass at the utility value adjacent to full health. This is not the case when estimating EQ5D-5L, due to the low proportion of observations at this point.
Conclusion: The beta-based mixture models marginally outperformed ALDVMM in this dataset when comparing models with the same number of components. This is at the expense of requiring a larger number of parameters and estimation time. Both model types are able to closely fit the data without biased characteristic of many mapping approaches. Skilled judgment is critical in determining the optimal model. Caution is required in ensuring a truly global maximum likelihood has been identified.