WITHDRAWN: Multiple Mapping Functions From the Modified Rankin Score to EQ-5D-3L: Data From Stroke Patients in France

OBJECTIVES: This study investigates a range of mapping functions to predict EQ-5D utilities from a disease-specific questionnaire, the modified Rankin Score (mRS) among french stroke patients.

METHODS: Demographic and clinical data were collected from a multicenter, prospective randomized controlled clinical trial ASTER1 (Contact Aspiration versus Stent Retriever for Successful Revascularization after stroke). All patients were asked to complete the EQ-5D-3L and the mRS questionnaire at 12 months post stroke. A total of 227 patients (53%) have fully completed both questionnaires. The Ordinary Least Square (OLS) and the response mapping approach which use a multinomial logit were used to predict the EQ-5D-3L. Their predictive ability was assessed. The French tariff value set was used to calculate the observed EQ-5D-3L utility scores. Analyses were performed on Stata 14 software.

RESULTS: Our sample comprises 46% of male and 54% of female with a mean age of 70 yo (sd =14). The modified Rankin Score was distributed as follows: 24% (0), 20% (1), 21% (2), 15% (3), 13% (4), 6% (5). OLS model most accurately predict the mean utility value. Any significant difference was found between the observed mean utility score (0.5866) and the predicted mean utility score from OLS (0.5866); but a slight difference was observed when compared to the response mapping approach (0.5974). However, a graphical analysis of the distribution of observed and predicted values of the utility score has shown that both OLS and response mapping did not accurately predict individual utility score.

CONCLUSIONS: These results have shown that OLS and response mapping are not a good predictive model of the EQ-5D-3L from the modified Rankin Score questionnaire. Further analysis using mixture methods such as adjusted limited dependent variable mixture and beta-based mixture models will be explored.