Artikel
Mapping Health-Related Quality of Life in Multiple Myeloma: Assessment of Existing Mapping Algorithms
Suche in Medline nach
Autoren
Veröffentlicht: | 6. März 2018 |
---|
Gliederung
Text
Background: The National Institute for Clinical Excellence (NICE) recommends quality-adjusted life years as the main measure for health outcomes in cost-utility analyses (CUAs). In the absence of health utility values (HUVs), NICE recommends to use “mapping” functions to link outcomes from different measures of health-related quality-of-life. We aimed to describe already published mapping algorithms for multiple myeloma (MM) and to assess if they could be replicated and generalized to other patient-population data-sets to derive HUVs.
Methods: We searched in PubMed/MEDLINE and Web of Science to identify studies reporting mapping techniques that predict EQ-5D HUVs from EORTC questionnaires (QLQ-C30 or QLQ-MY20) in MM. In the absence of a sufficient number of studies identified, we conducted an additional hand search in the Health Economics Research Centre (HERC) database. Our study assessed the algorithms based on the ISPOR Good Practice for Outcomes Research Task Force Report. To identify if algorithms could be replicated, we extracted and summarized the algorithms based on reporting on: justifications of the statistical models, their predictive performance, the population data-set and disease severity, validation techniques, and the type of the CUA for which the mapping was applied.
Results: From the literature search, we identified only seven studies reporting on mapping techniques. Five studies reported mapping models for MM and two for other cancer diseases. The most frequently applied statistical model for mapping was the ordinary least square model. Furthermore, ordered probit regression, Tobit models, two-part models, spline models, response mapping models, limited depended variable mixture models, and multiple linear regression analysis were used. Only few studies justified the statistical mapping model used. The study populations included MM patients with follow-up periods from baseline up to 18 months. Population data-sets varied between 137 and 1600 patients. The reported R-squares were indicating a relatively good model performance.
Conclusions: Only five algorithms were fully extracted and may be used to predict HUVs for comparable patient-population data-set. Several studies suggested a range of diseases for which the results could be useful, such as MM or other cancers with similar scores. However, further research exploring predictive performance for different patient groups and validation of mapping techniques should become a recommended approach.