The mapping technique can estimate generic preference-based measure scores through a specific measure that cannot be used in economic evaluations. This study compared 2 response mapping methods to estimate EQ-5D-5L scores using the Western Ontario McMaster Universities Osteoarthritis (WOMAC).
The sample consisted of 758 patients with the hip or knee osteoarthritis recruited in baseline. Bayesian networks (BN) and multinomial logistic regression (ML) were used as response mapping models. Predictions were obtained using the 6-month follow-up as a validation sample. The mean absolute error, mean squared error, deviation from the root mean squared error and intraclass correlation coefficient were calculated as precision measures.
There was 5.5% of missing data, which was removed. The mean age was 69.6 years (standard deviation = 10.5), with 61.6% of women. The BN model presented lower mean absolute error, mean squared error, root mean squared error and higher intraclass correlation coefficient than the ML model. Only the WOMAC items pain and physical function items were related with the EQ-5D-5L dimensions.
BN response mapping models are more robust methods, with better prediction results, than ML models. The BN model also provided a graphic representation of the dependency relationships between the EQ-5D-5L dimensions and the different WOMAC items that could be useful in the clinical investigation of patients with hip or knee osteoarthritis.