Published in:
01-05-2017 | Original Research Article
Can Mapping Algorithms Based on Raw Scores Overestimate QALYs Gained by Treatment? A Comparison of Mappings Between the Roland–Morris Disability Questionnaire and the EQ-5D-3L Based on Raw and Differenced Score Data
Published in: PharmacoEconomics | Issue 5/2017
Login to get accessAbstract
Introduction
Mapping algorithms are increasingly being used to predict health-utility values based on responses or scores from non-preference-based measures, thereby informing economic evaluations.
Objectives
We explored whether predictions in the EuroQol 5-dimension 3-level instrument (EQ-5D-3L) health-utility gains from mapping algorithms might differ if estimated using differenced versus raw scores, using the Roland–Morris Disability Questionnaire (RMQ), a widely used health status measure for low back pain, as an example.
Methods
We estimated algorithms mapping within-person changes in RMQ scores to changes in EQ-5D-3L health utilities using data from two clinical trials with repeated observations. We also used logistic regression models to estimate response mapping algorithms from these data to predict within-person changes in responses to each EQ-5D-3L dimension from changes in RMQ scores. Predicted health-utility gains from these mappings were compared with predictions based on raw RMQ data.
Results
Using differenced scores reduced the predicted health-utility gain from a unit decrease in RMQ score from 0.037 (standard error [SE] 0.001) to 0.020 (SE 0.002). Analysis of response mapping data suggests that the use of differenced data reduces the predicted impact of reducing RMQ scores across EQ-5D-3L dimensions and that patients can experience health-utility gains on the EQ-5D-3L ‘usual activity’ dimension independent from improvements captured by the RMQ.
Conclusion
Mappings based on raw RMQ data overestimate the EQ-5D-3L health utility gains from interventions that reduce RMQ scores. Where possible, mapping algorithms should reflect within-person changes in health outcome and be estimated from datasets containing repeated observations if they are to be used to estimate incremental health-utility gains.