Bayesian statistical models to estimate EQ-5D utility scores from EORTC QLQ data in myeloma.

It is well documented that the modelling of health-related quality of life data is difficult as the distribution of such data is often strongly right/left skewed and it includes a significant percentage of observations at one. The objective of this study is to develop a series of two-part models (TPMs) that deal with these issues. Data from the UK Medical Research Council Myeloma IX trial were used to examine the relationship between the European Organization for Research and Treatment of Cancer (EORTC) QLQ-C30/QLQ-MY20 scores and the European QoL-5 Dimensions (EQ-5D) utility score. Four different TPMs were developed. The models fitted included TPM with normal regression, TPM with normal regression with variance a function of participant characteristics, TPM with log-transformed data, and TPM with gamma regression and a log link. The cohort of 1839 patients was divided into 75% derivation sample, to fit the different models, and 25% validation sample to assess the predictive ability of these models by comparing predicted and observed mean EQ-5D scores in the validation set, unadjusted R2 , and root mean square error. Predictive performance in the derivation dataset depended on the criterion used, with R2 /adjusted-R2 favouring the TPM with normal regression and mean predicted error favouring the TPM with gamma regression. The TPM with gamma regression performs best within the validation dataset under all criteria. TPM regression models provide flexible approaches to estimate mean EQ-5D utility weights from the EORTC QLQ-C30/QLQ-MY20 for use in economic evaluation.