Christopher K Hoyle1, Margaret Tabberer2, Jean Brooks2. 1. Office of Health Economics, London, UK. Electronic address: chris.hoyle@btinternet.com. 2. GlaxoSmithKline Pharmaceuticals R&D, Uxbridge, Middlesex, UK.
Abstract
OBJECTIVES: To develop an algorithm for estimating EuroQol five-dimensional questionnaire (EQ-5D)-equivalent utilities from the chronic obstructive pulmonary disease (COPD) assessment test (CAT) and evaluate its use in economic evaluations as part of health technology assessments for COPD. METHODS: Data for the three-level EQ-5D (EQ-5D-3L) and the CAT were obtained from two multinational, phase III clinical trials. Three approaches were explored for estimating EQ-5D-equivalent utilities from the CAT: ordinary least-squares (OLS) regression, multinomial logistic regression, and a combination of the two. Estimated utilities were compared with actual EQ-5D-3L utilities, including treatment effect and the impact of disease severity on health, to evaluate the predictive performance of each algorithm. RESULTS: Root mean squared error and mean absolute error analyses showed that an OLS algorithm performed as well as algorithms developed using more complex modeling structures. The OLS regression included EQ-5D-3L utility weights as dependent variables and CAT items as independent variables. Within-sample validation showed systematic overestimation and underestimation within the range 0.5 ≤ EQ-5D-3L ≤ 0.9 (although all mean absolute errors were ≤0.100), with the smallest difference between estimated and actual values within 0.7 < EQ-5D-3L ≤ 0.9. The algorithm underestimated utility near full health and overestimated utility less than 0.5. As a consequence, the change from baseline was lower and the confidence intervals were narrower than those observed with actual EQ-5D-3L data. CONCLUSIONS: In the absence of EQ-5D data, the OLS regression algorithm may provide an estimate of utility for treatment models, but it is likely to underestimate treatment effects. Therefore, it is recommended that utilities be derived directly from the EQ-5D for the purposes of health technology assessments for COPD treatments in the United Kingdom.
OBJECTIVES: To develop an algorithm for estimating EuroQol five-dimensional questionnaire (EQ-5D)-equivalent utilities from the chronic obstructive pulmonary disease (COPD) assessment test (CAT) and evaluate its use in economic evaluations as part of health technology assessments for COPD. METHODS: Data for the three-level EQ-5D (EQ-5D-3L) and the CAT were obtained from two multinational, phase III clinical trials. Three approaches were explored for estimating EQ-5D-equivalent utilities from the CAT: ordinary least-squares (OLS) regression, multinomial logistic regression, and a combination of the two. Estimated utilities were compared with actual EQ-5D-3L utilities, including treatment effect and the impact of disease severity on health, to evaluate the predictive performance of each algorithm. RESULTS: Root mean squared error and mean absolute error analyses showed that an OLS algorithm performed as well as algorithms developed using more complex modeling structures. The OLS regression included EQ-5D-3L utility weights as dependent variables and CAT items as independent variables. Within-sample validation showed systematic overestimation and underestimation within the range 0.5 ≤ EQ-5D-3L ≤ 0.9 (although all mean absolute errors were ≤0.100), with the smallest difference between estimated and actual values within 0.7 < EQ-5D-3L ≤ 0.9. The algorithm underestimated utility near full health and overestimated utility less than 0.5. As a consequence, the change from baseline was lower and the confidence intervals were narrower than those observed with actual EQ-5D-3L data. CONCLUSIONS: In the absence of EQ-5D data, the OLS regression algorithm may provide an estimate of utility for treatment models, but it is likely to underestimate treatment effects. Therefore, it is recommended that utilities be derived directly from the EQ-5D for the purposes of health technology assessments for COPD treatments in the United Kingdom.
Authors: Andrew D Scarffe; Christopher J Licskai; Madonna Ferrone; Kevin Brand; Kednapa Thavorn; Doug Coyle Journal: Cost Eff Resour Alloc Date: 2022-08-12